Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
weekly update
This week the selected systems have lost money, and I am finally caught up with the systems, because I have just as much profit as they made (the values on the table below are affected by having removed the CL_ID_05, because the blender told me it was too risky):
I studied my systems with the ROITVOR ratio and I've discovered some good systems that the sharpe ratio hadn't highlighted:
I will run them on the blender tomorrow (my Monte Carlo VaR estimator).
[...]
You know, dude, this RoI-t-VoR ratio seems pretty messy, but it came from my hard work, so I am not going to flush it down the toilet like I threatened to do with the sharpe ratio.
I'm gonna find a way to deepen my knowledge on it. I am not going to abort this amateur academic attempt of mine, mostly because actually it's not academic and i really need it. I've thought about reverting to the sharpe ratio, more than once today. But it doesn't measure Return On Investment. There's no way around this. I've also thought about reverting to something even simpler, the profit factor. But i haven't done it yet.
The sharpe ratio was neat, but deceivingly neat. For that matter, the profit factor was even neater. But then, by that rationale, i'd go back to the mere profit.
Latest version of my "developing_my_own_risk_metric.xls"
For those of my three readers who would like to participate in this little experiment, here's my latest version of the file to experiment and give me advice or questions or objections:
View attachment developing_my_own_risk_metric.xls
This week the selected systems have lost money, and I am finally caught up with the systems, because I have just as much profit as they made (the values on the table below are affected by having removed the CL_ID_05, because the blender told me it was too risky):
I studied my systems with the ROITVOR ratio and I've discovered some good systems that the sharpe ratio hadn't highlighted:
I will run them on the blender tomorrow (my Monte Carlo VaR estimator).
[...]
You know, dude, this RoI-t-VoR ratio seems pretty messy, but it came from my hard work, so I am not going to flush it down the toilet like I threatened to do with the sharpe ratio.
I'm gonna find a way to deepen my knowledge on it. I am not going to abort this amateur academic attempt of mine, mostly because actually it's not academic and i really need it. I've thought about reverting to the sharpe ratio, more than once today. But it doesn't measure Return On Investment. There's no way around this. I've also thought about reverting to something even simpler, the profit factor. But i haven't done it yet.
The sharpe ratio was neat, but deceivingly neat. For that matter, the profit factor was even neater. But then, by that rationale, i'd go back to the mere profit.
Latest version of my "developing_my_own_risk_metric.xls"
For those of my three readers who would like to participate in this little experiment, here's my latest version of the file to experiment and give me advice or questions or objections:
View attachment developing_my_own_risk_metric.xls
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
zn_on_02 vs gbp_id_02
Case study: zn_on_02 vs gbp_id_02
The ZN system was my favorite. It has a win rate of 65% vs 57% of the GBP_ID_02. It has a sharpe ratio of 5.52 vs 1.24. What else do we need? It always seemed my best system against some system, the GBP_ID_02, which I never even looked at.
And yet...
My ROITVOR ratio, and even the scatter plot, tells me that GBP_ID_02 is better than what I thought was my best system. How on earth could this be possible?
Let's study them.
They actually make the same yearly profit, but ZN requires less margin:
(same table as the one shown a few lines higher)
So how can the GBP system be better?
Found!
Stdevp is 400 for GBP and 600 for ZN. This is not a small detail: it has less variability. This is important to me. It keeps my chance of blowing out lower.
How does GBP_ID_02 achieve this with such a poor performance (sharpe ratio of 1.24 vs 5.52)?
Simple: it trades 300 times vs the 50 times traded by ZN_ON_02.
It's like the pace of the turtle. It keeps going and going, and it surpasses the ZN_ON_02 despite performing worse. Even the small steps forward and the many steps backward do not stop the GBP_ID_02 from doing better than ZN_ON_02.
It is quite shocking how sharpe ratio overlooked this, by telling me that one was worth nothing (1.24) and the other was worth gold (5.5).
ROITVOR ratio has struck again.
Now I'll have to check with my blender. And if it works there, too, then I'll know that ROITVOR also predicts the results of my Monte Carlo VaR estimator. However, this is all still at an empirical level. I do realize it.
Nope, it did get a little worse, like 1%, but it adds action so it's good.
[...]
I ended up not adding GBP_ID_02 and I also found out why adding ZN_ON_02 is ok, whereas adding GBP_ID_02 is not. Because the fact that GBP_ID_02 produces the same profit with smaller losses doesn't mean that on my blender it will produce better results and cut down on shortfall expectation (or VaR expectations, whatever it's called).
Indeed, what also counts is the frequency of those losses. If a system has small losses, but they happen twice as frequently, no matter how much money it makes, the cumulative losses will cause the shortfall estimator (or VaR estimator, whatever it's called), to tell me that I am risking a bigger fall than if I enable a system with bigger but less frequent losses.
Now... I need to find a ratio that will anticipate the VaR expectations, in other words, I am looking for a ratio that will closely forecast probability of Var, without the need for empirical tests: this will save me a lot of time.
In what proportion will adding a (profitable) system with x-size losses with x-size probability benefit the overall portfolio? I wonder if the sharpe ratio or the profit factor can tell me this. To some extent they will, but I know sharpe ratio won't do so with total precision, given that it counts upside variability as well. Profit factor is different but it has its limits, too.
This is going to be hard, but I am defining the problem better.
What I want is a ratio in line with my Monte Carlo VaR estimator. And what I understood is that we have on one side Profit over margin, which is the Return On Investment measure. And then we have on the other side the variability.
What is also clear is that a ratio, like my present ROITVOR, telling me that GBP_ID_02 is better than ZN_ON_02, merely because the average loss is smaller, and all the other things are equal, is only telling me half the story and is not assessing my systems correctly.
Instead I want to know, in probability terms, how much VaR do I have to achieve a given ROI.
Here's there's something about this:
A formula for weighing Risk / Reward / Probability
The need to use risk /reward together with the probability. I wonder if this is the same a Profit Factor.
Gross profit / gross loss which is the profit factor, is this ratio the same as (average win * probability) divided by (average loss * probability)? And what is the latter called? What is it? Expected value?
Here they say it's called "expectancy":
Why the Risk-to-Reward Ratio is Overrated
What is the difference between expectancy and profit factor and average trade? And how can they help me identify the downside risk or Var of my systems?
Expectancy/expectance is very similar to expected value, but not the same:
Expected value - Wikipedia, the free encyclopedia
I found a good article here:
Evaluation of trading systems applied to purchase and sale strategies.
...
Oh, wow. This is a gold mine:
The 101 Ways to Measure Portfolio Performance by Philippe Cogneau, Georges Hubner :: SSRN
Oh my god, another excellent book:
http://www.edhec-risk.com/performan...ce measurement for traditional investment.pdf
Always from the same article I was quoting earlier:
Evaluation of trading systems applied to purchase and sale strategies.
I am lost in this sea of information. This is going to take a while.
Case study: zn_on_02 vs gbp_id_02
The ZN system was my favorite. It has a win rate of 65% vs 57% of the GBP_ID_02. It has a sharpe ratio of 5.52 vs 1.24. What else do we need? It always seemed my best system against some system, the GBP_ID_02, which I never even looked at.
And yet...
My ROITVOR ratio, and even the scatter plot, tells me that GBP_ID_02 is better than what I thought was my best system. How on earth could this be possible?
Let's study them.
They actually make the same yearly profit, but ZN requires less margin:
(same table as the one shown a few lines higher)
So how can the GBP system be better?
Found!
Stdevp is 400 for GBP and 600 for ZN. This is not a small detail: it has less variability. This is important to me. It keeps my chance of blowing out lower.
How does GBP_ID_02 achieve this with such a poor performance (sharpe ratio of 1.24 vs 5.52)?
Simple: it trades 300 times vs the 50 times traded by ZN_ON_02.
It's like the pace of the turtle. It keeps going and going, and it surpasses the ZN_ON_02 despite performing worse. Even the small steps forward and the many steps backward do not stop the GBP_ID_02 from doing better than ZN_ON_02.
It is quite shocking how sharpe ratio overlooked this, by telling me that one was worth nothing (1.24) and the other was worth gold (5.5).
ROITVOR ratio has struck again.
Now I'll have to check with my blender. And if it works there, too, then I'll know that ROITVOR also predicts the results of my Monte Carlo VaR estimator. However, this is all still at an empirical level. I do realize it.
Nope, it did get a little worse, like 1%, but it adds action so it's good.
[...]
I ended up not adding GBP_ID_02 and I also found out why adding ZN_ON_02 is ok, whereas adding GBP_ID_02 is not. Because the fact that GBP_ID_02 produces the same profit with smaller losses doesn't mean that on my blender it will produce better results and cut down on shortfall expectation (or VaR expectations, whatever it's called).
Indeed, what also counts is the frequency of those losses. If a system has small losses, but they happen twice as frequently, no matter how much money it makes, the cumulative losses will cause the shortfall estimator (or VaR estimator, whatever it's called), to tell me that I am risking a bigger fall than if I enable a system with bigger but less frequent losses.
Now... I need to find a ratio that will anticipate the VaR expectations, in other words, I am looking for a ratio that will closely forecast probability of Var, without the need for empirical tests: this will save me a lot of time.
In what proportion will adding a (profitable) system with x-size losses with x-size probability benefit the overall portfolio? I wonder if the sharpe ratio or the profit factor can tell me this. To some extent they will, but I know sharpe ratio won't do so with total precision, given that it counts upside variability as well. Profit factor is different but it has its limits, too.
This is going to be hard, but I am defining the problem better.
What I want is a ratio in line with my Monte Carlo VaR estimator. And what I understood is that we have on one side Profit over margin, which is the Return On Investment measure. And then we have on the other side the variability.
What is also clear is that a ratio, like my present ROITVOR, telling me that GBP_ID_02 is better than ZN_ON_02, merely because the average loss is smaller, and all the other things are equal, is only telling me half the story and is not assessing my systems correctly.
Instead I want to know, in probability terms, how much VaR do I have to achieve a given ROI.
Here's there's something about this:
A formula for weighing Risk / Reward / Probability
The need to use risk /reward together with the probability. I wonder if this is the same a Profit Factor.
Gross profit / gross loss which is the profit factor, is this ratio the same as (average win * probability) divided by (average loss * probability)? And what is the latter called? What is it? Expected value?
Here they say it's called "expectancy":
Why the Risk-to-Reward Ratio is Overrated
What is the difference between expectancy and profit factor and average trade? And how can they help me identify the downside risk or Var of my systems?
Expectancy/expectance is very similar to expected value, but not the same:
Expected value - Wikipedia, the free encyclopedia
I found a good article here:
Evaluation of trading systems applied to purchase and sale strategies.
Hey, I have to find this book.
...
Oh, wow. This is a gold mine:
The 101 Ways to Measure Portfolio Performance by Philippe Cogneau, Georges Hubner :: SSRN
Oh my god, another excellent book:
http://www.edhec-risk.com/performan...ce measurement for traditional investment.pdf
Always from the same article I was quoting earlier:
Evaluation of trading systems applied to purchase and sale strategies.
I am lost in this sea of information. This is going to take a while.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
good one:
CapitalatWork
Holy cow... after days of work and criticism, it turned out that the best representation of my own systems is the scatter plot used to represent the efficient frontier. Congratulations markowitz.
I still have my ratio that puts x-axis and y-axis of the scatter plot together, but there was scales out of proportion and overlapping of information if I represented it in any other way than this (chart above).
The only big difference that I introduced, not necessarily to markowitz, but to my understanding of the sharpe ratio, is that I am no longer dividing... using average trade over standard deviation (of the trades), but I am using the ROITVOR version, which is yearly profit over margin and standard deviation over profit (y-axis). I then divide the first by the second and get the ROITVOR ratio, but right now I don't use it because my scatter plot is fantastic as it is and i don't need to look at a ratio that divides the value on the x-axis by the value on the y-axis. I have all that information on the scatter plot, and it is kept separate, which is best. Then, of course, if I'll need a univocal value, that's the value I will use.
Dude, this is complex ****. I am getting better, but I still suck.
...
The info I am now missing on the chart, but it's for the best because it'd be too confusing, is the absolute margin, the absolute standard deviation, absolute loss... it's all relative now.
So, it's obvious - anyone knowing the efficient frontier concept can tell immediately that NG_ID_04 is incredibly good, returning 400% (on margin) with a very small standard deviation (on profit), but there's a problem. I do not know how much, in dollars, that standard deviation is. So I don't know if I can afford that system. Of course intuitively I'll know that I can't afford a system on silver, but for the others, anything could be possible. It could be that a NG system has a low standar deviation and a good profit, and so it scores high, or it has a large standard devation and a huge profit, and it scores high because of that.
So I am having second thoughts about this representation. But none about the y-axis. Return has to be % return on margin.
And I don't think I can divide standar deviation by margin, too, because if I do, this would happen:
And it would be equivalent to dividing Profit by Standard Deviation and then I'd lose track of Return On Investment. No doubt about this.
And yet, it's clear to me that markowitz and the others are talking about standard deviation on the capital invested. But they're talking about stocks. What is the equivalent for futures... let's do a search:
https://www.google.com/search?q=efficient+frontier+for+futures
...
Unsuccessful.
Should I just put on the bottom the standard deviation?
I can't, because a system trading silver will have a higher standard deviation, and unfairly so... as i was saying this morning...
Hold on...
I do not have to use this measure to rate anything, so this would be there only to tell me... got it!
I am going to keep it as it is, and add the absolute standard deviation next to the names, with hundreds, rounded (divided by 100 dollars):
Ok.
So now for example I can compare everything from this scatter plot, and in particular GBP_ID_02 with CAD_ID_03:
Let me make sure there's no useless info on this chart.
For one reason (high accuracy) or for another (frequent trades) GBP_ID_02 performs much better than CAD_ID_03. It returns 200% per year vs. 30% per year. It's due to the frequency of trading because it's got a profit factor which is worse, and a win rate which is lower. Smaller wins to losses, lower % of wins, and yet it performs 6 times better: I would have been fooled by profit factor, sharpe ratio, % of wins... unbelievable.
So, what else? That "4" next to both of them, tells me that I can expect a standard deviation of 400 dollars, which means an average loss more or less of that size (slightly less, since the average trade is positive, because the systems are profitable), and a max loss... who knows about that? I can check, but the standard deviation has little sensitivity to it.
So how do I know which one to trade? I know the margin, I can afford both. I can afford both in terms of losses... but my blender told me the GBP would lower my survival rate and for now it doesn't make sense to add it, because I should be making enough to keep moving forward.
So I ultimately decided not to add it. So many things to consider, also that are still not in the scatter plot.
For example, number of forward-tested trades. They're both satisfactory.
My eyes are killing me. I have to stop the reasoning and the writing here, for today.
[...]
Will write some more.
Look here:
Basically, I am quite confident about the systems in the green square. I'd be making a mistake if I didn't trade those (in the square) with a standard devation of less than 500 and a margin that I can afford.
For some reason there's something more to it than what's shown in the chart. Accuracy is important, too. It's not ROI and it's not st.dev.divided by profit.
It is profit factor, and it is sharpe ratio.
I'm about to go crazy, with all these variables and the many ways they can mix.
The principles are that I cannot make it too complex, and that I want to try to make it as complex as possible, so that's what causes this endless conflict within me: i try to push it a bit further, but then I fall back on my ass, by not understanding all the implications of my changes.
The objective is to automate everything, but in order to do this... hey, "maximize sharpe ratio" as everyone says just isn't the recipe. It's tempting to think that way, but it isn't.
Also, it's unclear if I am after either of these or both:
1) make more money, doesn't matter how
2) do things properly, doesn't matter if they produce money
Automating this whole thing may not even bring money, but slow down the process of making money.
I am endlessly moving back and forth between automating and synthesizing the whole process, and gathering good systems in a pragmatic way.
If I make money, then I can pay some math... any math professor to coach me.
So I should focus on practice, and money.
But I am afraid of quitting this thing and never picking it up again. Given that it's so unpleasant.
Also, after 4 months of math review, thousands of exercises, to be ready for this feat, I'd feel like an idiot if I quit.
At any rate, I must say that it has helped me already. The blender totally came out of this. And its predecessor, too. They came out of the probability theory I studied.
All right, **** it. Let's take a break for my brain, by being pragmatic. Let's stop for a bit to try to be coherent, logical, and putting it all together theoretically, and let's just go about finding good systems to add to my portfolio, and...
**** this whole thing.
Let's put profit factor on the x-axis. I've already got st.dev. in the ****ing name of the systems.
Let's be practical.
Profit factor tells me the accuracy of a system more clearly than anything else.
Return On Investment tells me how profitable.
One tells me how accurate, the other tells me how profitable, and the st.dev. in the name tells me how affordable. The margin, I know immediately by the name.
Ok, I'm gonna go down the practical useful path and screw the academic path.
**** the efficient frontier, **** them all. You know why? Also because I don't get why I am dividing st.dev. by profit. I want everything to be clear, or I can't do it.
These ****ing academics make it look so neat, but do not go through the trouble of explaining it to me clearly - you assholes. They only explain it to one another.
So **** them. I tried, but I could not figure it out, so **** them all.
[...]
Ok. Good. Done. I got rid of some outliers, systems that i could not trade for various reasons: mostly lack of capital and the fact that they haven't traded very much in forward-testing.
I used a log scale.. I had to do so much to take care of the scales... so that everyone would be separate and not tight together.
I am done for the day. I can't do much better anyway. This is good practical stuff.
By going into this labyrinth I did not learn all the time. Sometimes I just got confused.
But my scatter plot of today shows more than it showed a few weeks ago.
I am satisfied. Certainly this field isn't boring. There's even too much stuff to read, especially for one who's not an expert at math, and has to go the same paths, without realizing it. With a good understanding of formulas I would have saved so much time. It's like for someone who's expert at understanding laws: he goes, picks up the law, and gets done much faster than someone else would. Or like for a programmer to fix a computer vs for a regular person...
I need those math lessons, from a real person.
CapitalatWork
Holy cow... after days of work and criticism, it turned out that the best representation of my own systems is the scatter plot used to represent the efficient frontier. Congratulations markowitz.
I still have my ratio that puts x-axis and y-axis of the scatter plot together, but there was scales out of proportion and overlapping of information if I represented it in any other way than this (chart above).
The only big difference that I introduced, not necessarily to markowitz, but to my understanding of the sharpe ratio, is that I am no longer dividing... using average trade over standard deviation (of the trades), but I am using the ROITVOR version, which is yearly profit over margin and standard deviation over profit (y-axis). I then divide the first by the second and get the ROITVOR ratio, but right now I don't use it because my scatter plot is fantastic as it is and i don't need to look at a ratio that divides the value on the x-axis by the value on the y-axis. I have all that information on the scatter plot, and it is kept separate, which is best. Then, of course, if I'll need a univocal value, that's the value I will use.
Dude, this is complex ****. I am getting better, but I still suck.
...
The info I am now missing on the chart, but it's for the best because it'd be too confusing, is the absolute margin, the absolute standard deviation, absolute loss... it's all relative now.
So, it's obvious - anyone knowing the efficient frontier concept can tell immediately that NG_ID_04 is incredibly good, returning 400% (on margin) with a very small standard deviation (on profit), but there's a problem. I do not know how much, in dollars, that standard deviation is. So I don't know if I can afford that system. Of course intuitively I'll know that I can't afford a system on silver, but for the others, anything could be possible. It could be that a NG system has a low standar deviation and a good profit, and so it scores high, or it has a large standard devation and a huge profit, and it scores high because of that.
So I am having second thoughts about this representation. But none about the y-axis. Return has to be % return on margin.
And I don't think I can divide standar deviation by margin, too, because if I do, this would happen:
And it would be equivalent to dividing Profit by Standard Deviation and then I'd lose track of Return On Investment. No doubt about this.
And yet, it's clear to me that markowitz and the others are talking about standard deviation on the capital invested. But they're talking about stocks. What is the equivalent for futures... let's do a search:
https://www.google.com/search?q=efficient+frontier+for+futures
...
Unsuccessful.
Should I just put on the bottom the standard deviation?
I can't, because a system trading silver will have a higher standard deviation, and unfairly so... as i was saying this morning...
Hold on...
I do not have to use this measure to rate anything, so this would be there only to tell me... got it!
I am going to keep it as it is, and add the absolute standard deviation next to the names, with hundreds, rounded (divided by 100 dollars):
Ok.
So now for example I can compare everything from this scatter plot, and in particular GBP_ID_02 with CAD_ID_03:
Let me make sure there's no useless info on this chart.
For one reason (high accuracy) or for another (frequent trades) GBP_ID_02 performs much better than CAD_ID_03. It returns 200% per year vs. 30% per year. It's due to the frequency of trading because it's got a profit factor which is worse, and a win rate which is lower. Smaller wins to losses, lower % of wins, and yet it performs 6 times better: I would have been fooled by profit factor, sharpe ratio, % of wins... unbelievable.
So, what else? That "4" next to both of them, tells me that I can expect a standard deviation of 400 dollars, which means an average loss more or less of that size (slightly less, since the average trade is positive, because the systems are profitable), and a max loss... who knows about that? I can check, but the standard deviation has little sensitivity to it.
So how do I know which one to trade? I know the margin, I can afford both. I can afford both in terms of losses... but my blender told me the GBP would lower my survival rate and for now it doesn't make sense to add it, because I should be making enough to keep moving forward.
So I ultimately decided not to add it. So many things to consider, also that are still not in the scatter plot.
For example, number of forward-tested trades. They're both satisfactory.
My eyes are killing me. I have to stop the reasoning and the writing here, for today.
[...]
Will write some more.
Look here:
Basically, I am quite confident about the systems in the green square. I'd be making a mistake if I didn't trade those (in the square) with a standard devation of less than 500 and a margin that I can afford.
For some reason there's something more to it than what's shown in the chart. Accuracy is important, too. It's not ROI and it's not st.dev.divided by profit.
It is profit factor, and it is sharpe ratio.
I'm about to go crazy, with all these variables and the many ways they can mix.
The principles are that I cannot make it too complex, and that I want to try to make it as complex as possible, so that's what causes this endless conflict within me: i try to push it a bit further, but then I fall back on my ass, by not understanding all the implications of my changes.
The objective is to automate everything, but in order to do this... hey, "maximize sharpe ratio" as everyone says just isn't the recipe. It's tempting to think that way, but it isn't.
Also, it's unclear if I am after either of these or both:
1) make more money, doesn't matter how
2) do things properly, doesn't matter if they produce money
Automating this whole thing may not even bring money, but slow down the process of making money.
I am endlessly moving back and forth between automating and synthesizing the whole process, and gathering good systems in a pragmatic way.
If I make money, then I can pay some math... any math professor to coach me.
So I should focus on practice, and money.
But I am afraid of quitting this thing and never picking it up again. Given that it's so unpleasant.
Also, after 4 months of math review, thousands of exercises, to be ready for this feat, I'd feel like an idiot if I quit.
At any rate, I must say that it has helped me already. The blender totally came out of this. And its predecessor, too. They came out of the probability theory I studied.
All right, **** it. Let's take a break for my brain, by being pragmatic. Let's stop for a bit to try to be coherent, logical, and putting it all together theoretically, and let's just go about finding good systems to add to my portfolio, and...
**** this whole thing.
Let's put profit factor on the x-axis. I've already got st.dev. in the ****ing name of the systems.
Let's be practical.
Profit factor tells me the accuracy of a system more clearly than anything else.
Return On Investment tells me how profitable.
One tells me how accurate, the other tells me how profitable, and the st.dev. in the name tells me how affordable. The margin, I know immediately by the name.
Ok, I'm gonna go down the practical useful path and screw the academic path.
**** the efficient frontier, **** them all. You know why? Also because I don't get why I am dividing st.dev. by profit. I want everything to be clear, or I can't do it.
These ****ing academics make it look so neat, but do not go through the trouble of explaining it to me clearly - you assholes. They only explain it to one another.
So **** them. I tried, but I could not figure it out, so **** them all.
[...]
Ok. Good. Done. I got rid of some outliers, systems that i could not trade for various reasons: mostly lack of capital and the fact that they haven't traded very much in forward-testing.
I used a log scale.. I had to do so much to take care of the scales... so that everyone would be separate and not tight together.
I am done for the day. I can't do much better anyway. This is good practical stuff.
By going into this labyrinth I did not learn all the time. Sometimes I just got confused.
But my scatter plot of today shows more than it showed a few weeks ago.
I am satisfied. Certainly this field isn't boring. There's even too much stuff to read, especially for one who's not an expert at math, and has to go the same paths, without realizing it. With a good understanding of formulas I would have saved so much time. It's like for someone who's expert at understanding laws: he goes, picks up the law, and gets done much faster than someone else would. Or like for a programmer to fix a computer vs for a regular person...
I need those math lessons, from a real person.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
I woke up and I took out my scatter plot, and it looked pretty good, without being so tired as yesterday.
Then I graphically removed all the systems that I am either trading or that for some reason I can't trade, and I was left with these three:
Why am I not trading these three?
I'm gonna run the blender again.
...
Ok, listen, the forward-tested equity line goes from this:
To this:
It looks much better.
GBP_ID_02 and NQ_ID_03 are really no threat at all. The problem is that they have bad back-tested data and that is why I hadn't trusted them. The blender reflects this.
NG_ID_04 is one of the best 3 systems I have, and has an excellented back-tested record, but it has a little bit too much standard deviation (and average loss, and maximum loss).
...
Damn...
I feel so bad about it, but it would be stupid, after being prudent all this time, to now add NG_ID_04.
The blender says we go from 98% of survival to 94%. This could mean a lot more, because the future is worse.
I can always add it in a few more weeks.
Ok, let's just add the other two, that require low margin and have lower standard deviations.
...
No, let's not add anything. Because the forward-tested equity line gets worse, and these systems have long drawdowns.
Also, the back-tested record is not as good... basically I will postpone adding anything for another month, because I can't afford to make any mistakes and to be insecure. If you're insecure, then you disable systems and when you do that, you lose money. The mere insecurity is a cause of losses.
However, now I know which are the next 3 systems to enable, these three that I listed in this post.
Then I graphically removed all the systems that I am either trading or that for some reason I can't trade, and I was left with these three:
Why am I not trading these three?
I'm gonna run the blender again.
...
Ok, listen, the forward-tested equity line goes from this:
To this:
It looks much better.
GBP_ID_02 and NQ_ID_03 are really no threat at all. The problem is that they have bad back-tested data and that is why I hadn't trusted them. The blender reflects this.
NG_ID_04 is one of the best 3 systems I have, and has an excellented back-tested record, but it has a little bit too much standard deviation (and average loss, and maximum loss).
...
Damn...
I feel so bad about it, but it would be stupid, after being prudent all this time, to now add NG_ID_04.
The blender says we go from 98% of survival to 94%. This could mean a lot more, because the future is worse.
I can always add it in a few more weeks.
Ok, let's just add the other two, that require low margin and have lower standard deviations.
...
No, let's not add anything. Because the forward-tested equity line gets worse, and these systems have long drawdowns.
Also, the back-tested record is not as good... basically I will postpone adding anything for another month, because I can't afford to make any mistakes and to be insecure. If you're insecure, then you disable systems and when you do that, you lose money. The mere insecurity is a cause of losses.
However, now I know which are the next 3 systems to enable, these three that I listed in this post.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
I am at above 9k, the portfolio seems very safe (at least 90% chance of not blowing out), I'm gonna take a break from adding systems and worrying about finding systems to add. It should rise at a rate of about 1000 per week.
I will just read whatever portfolio theory I get my hands on, peacefully and without stress. I will try to stay away from looking at the markets... what else?
I just need to sleep well.
I will just read whatever portfolio theory I get my hands on, peacefully and without stress. I will try to stay away from looking at the markets... what else?
I just need to sleep well.
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
Continuing from here:
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-81.html#post1812234
Ok, I think this is going to be the post where I nail it.
As I said in the mentioned post, I was wondering why ZN_ON_02 rates better in the blender than GBP_ID_02, despite the fact that both produce the same profit, with the same margin and yet the GBP system has a lower standard deviation.
The reason is the performance of the ZN system, the fact that it has a higher % of wins.
In other words, if GBP goes from 10k to 100k by taking small steps, equally likely, of +1k and -999, the chance of them being blended next to one another is huge, whereas if ZN goes from 10k to 100k by taking steps of 75% +2k and 25% + 1.2k, then yes the loss is bigger, but the probability of a loss being mixed (by the monte carlo blender) next to another loss is so small, that even a system with a bigger average loss (or standard deviation, which is similar) will perform better in the blender, than one with a smaller loss.
So I am looking precisely for something to measure and weigh that average loss, standard deviation or what have you (a downside risk measure).
I think what I need is very related to the concepts of expectancy and/or expected value/return:
Expected Return Definition | Investopedia
Expected value - Wikipedia, the free encyclopedia
Expected Value - YouTube
****! Expected value is none other than average trade. This is not good.
How about.. and yes, expectancy is just the same, even though here he turns it around and uses a slightly different formula (since he gives losses a positive value):
Why the Risk-to-Reward Ratio is Overrated
What really pisses me off incidentally is that they call the same thing with so many different terms: expected value, expected return, expectancy, expectance... screw them, seriously.
So what are we going to use to make ZN_ON_02 stand out as better than GBP_ID_02 (and all the other systems in a situation like this one)?
We can't use standard deviation any more, because if we use the average loss and its probability, we should use it altogether. So let's see if I can get something out of multiplying the average loss by its probability. At any rate the average loss is very close to standard deviation.
Ok, this is not working and I can't figure it out with my limited skills.
I will opt for something in between, and temporary, rather than just giving up altogether: the profit factor, which I already have adopted in my scatter plot.
...
Nope, profit factor doesn't work either, because it's no good at assessing the deviation and if it see two trades losing 1000, it is the same to profit factor as one trade losing 2000, and yet to me it's not the same, so screw profit factor, too.
I need to do a probabilistic formula, which will help me both for the theoretical ratio and the practical scatter plot, so that both will match the monte carlo blender.
It all boils down to this: ZN guy has a standard deviation of 600 with a 33% probability of losing and GBP guy has a standard deviation of 400 with a 45% probability of losing. This is in the blender makes ZN preferable to GBP: now how do i figure out a formula to assess the probability of 400 dollars losses piling up for GBP and if their rate is cumulatively higher than 600 dollars losses piling up for ZN?
Let's get thinking.
Let's simplify. Let's say zn has 20% and gbp has 50%. Of course they're still both profitable and still have those standard deviations.
Ok, so 400 has another 50% to be followed by another 400, so there's 25% chance of getting a loss of 800, and a 12.5%, after 3 trades, of getting a loss of 1200. Then of course let's not forget that gbp to reach the same profit of zn with such a poor performance is trading much more frequently and so this further increases the probability of it incurring such drawdowns.
ZN system has a 20% chance of getting a loss of 600, 0.2^2=...4% of getting another loss, for a total of 1200, so bingo! The system with the bigger loss has a lower probability of incurring the same drawdown as the system with the smaller loss (and a lower win rate).
So, ok: the calculation has to be done like this. What is the probability for a system rolling a die (a hypothetical "trade die") x times to have x losses in a row, while another systems with different number of rolls has smaller losses but bigger probability of losing?
Ok, let's pretend we're rolling it 10 times.
Roll it once, and it's 20%
twice and it's 4%...
this is all about combinations and permutations i think. But i was weak on that lesson.
The question is: what is the chance of getting x consecutive losses in x trades if each loss has a probability of x% of happening?
If I then plug in the standard deviation, I get the most likely drawdown (assuming a random order of losses)...
I am dragging it on forever just because i am math illiterate. I know, but I am doing the best that I can, and once I fix this, the formula will be ready.
Ok, there's one guy saying it here:
Tossing Coins: 5 consecutive heads out of 10 attempts?
What is the probability of getting 15 or more CONSECUTIVE heads over 40 coin tosses?
I am going to start calculating the permutations possible with 10 trades:
Combinations and Permutations Calculator
n^r so it's 1024 possible permutations.
Here's a calculator on the average number of tosses necessary to get a head run of x length:
Coin Toss Runs Calculator
Wow! I am lost again.
Ah ah, I am so lost that it makes me laugh. If I had waited to figure this stuff out, before building the systems, there's no way I would have achieved anything. Instead I built them before doing portfolio theory, so I built them because no one scared telling me that it was going to be difficult. So i built them quickly, by skipping all this. Good thing. As they say, ignorance is bliss.
However, now I want to do things properly, so I am gonna figure this out, now or another time.
The question, again, is:
1) system A loses 400 in 20 random losses out of 40 trades.
2) systems B loses 600 in 3 random losses out of 10 trades.
They make the same amount of profit, and use the same margin. So, that system A would have seemed better, because it has a smaller loss and the same Return On Investment, but obviously it has a much lower accuracy and a worse performance.
But the question is now: what formula divides ROI to tell us that system B is better than system A despite its bigger loss?
Such a formula is entirely based on probability laws, given that it has to be a formula and not an empirical monte carlo simulation. And this because I don't want to have to do a simulation each time I want to assess a system.
In this formula, size of average loss (or standard deviation if you want), % of wins and number of trades are - I think - the three ingredients needed.
Average loss should be better and simpler. Let's go and check...
...
Ok, perfect correlation, and very close, so I'll use average loss.
What is the exact data for these two systems:
ZN_ON_02:
average loss of 400
percentage of losses: 35%
trades: 50
GBP_ID_02:
average loss of 300
percentage of losses: 43%
trades: 300
Question: given that the ROI is the same, find a formula that explains why ZN_ON_02 is less likely to lose money and therefore it's a better system.
For one thing the GBP is rolling the die 6 times as many. So the probability of consecutive losing trades is much higher. Then it has a higher loss rate.
I am missing the math to do this. Let's look on khan academy.
Wait, maybe I can do this. The least common multiple of the losses is 1200. To get there, GBP has to lose 4 times in a row, and ZN only 3. But one trades 50 times, and the other one trades 300 times.
So... still lost.
Frequency Probability and Unfair Coins - YouTube
Probability Density Functions - YouTube
Oh, wow... I'd have never thought i'd get lost over this. I can't solve it. Probability seems so simple and yet it's harder and more confusing than it seems.
Ok.
Forget about the size of the average loss for now. I need to focus on the probability of x consecutive losses with x number of trades. This is related to this calculator, which however does not do exactly what I want and does not show the formula:
Coin Toss Runs Calculator
This may be closer:
Probability of L consecutive losses not occurring in N rounds? - Yahoo! Answers
[...]
Using both links, I am figuring it out. This is where I'll take the formula from:
Coin Toss Runs Calculator
His formula is closely matching my monte carlo simulator.
...
Ok I reproduced the calculator on excel:
And it matches my monte carlo VaR estimator.
Now it's a matter of making this work with my systems, and getting it to compare the ZN system to the GBP system and rating them accordingly.
[...]
I think I did it. Unbelievable how resourceful I am, despite being ignorant:
This is the file:
View attachment developing_my_own_risk_metric.xls
Now I have a univocal formula that appraises downside risk, expected shortfall, monte carlo VaR, call you whatever is correct - I don't know.
This formula does it all theoretically, without a need for me to take the trades from that system and mix them with the rest of the portfolio.
It says ZN_ON_02 is far superior to GBP_ID_05, in that, while making the same profit with the same margin requirements in the same period, it has a much smaller probability of losing 1000 dollars, despite a bigger average loss. Mostly because it trades less and it has a higher win rate.
There might still be mistakes in the formula, but most of it is done, so I can go to sleep. Oh, and the formula isn't ready yet actually, because this is just the material for the scatter plot, that will have ROI on y-axis and this VaR measure on the x-axis. It's still going to take a while, but the hard part is behind me.
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-81.html#post1812234
Ok, I think this is going to be the post where I nail it.
As I said in the mentioned post, I was wondering why ZN_ON_02 rates better in the blender than GBP_ID_02, despite the fact that both produce the same profit, with the same margin and yet the GBP system has a lower standard deviation.
The reason is the performance of the ZN system, the fact that it has a higher % of wins.
In other words, if GBP goes from 10k to 100k by taking small steps, equally likely, of +1k and -999, the chance of them being blended next to one another is huge, whereas if ZN goes from 10k to 100k by taking steps of 75% +2k and 25% + 1.2k, then yes the loss is bigger, but the probability of a loss being mixed (by the monte carlo blender) next to another loss is so small, that even a system with a bigger average loss (or standard deviation, which is similar) will perform better in the blender, than one with a smaller loss.
So I am looking precisely for something to measure and weigh that average loss, standard deviation or what have you (a downside risk measure).
I think what I need is very related to the concepts of expectancy and/or expected value/return:
Expected Return Definition | Investopedia
Expected value - Wikipedia, the free encyclopedia
Expected Value - YouTube
****! Expected value is none other than average trade. This is not good.
How about.. and yes, expectancy is just the same, even though here he turns it around and uses a slightly different formula (since he gives losses a positive value):
Why the Risk-to-Reward Ratio is Overrated
What really pisses me off incidentally is that they call the same thing with so many different terms: expected value, expected return, expectancy, expectance... screw them, seriously.
So what are we going to use to make ZN_ON_02 stand out as better than GBP_ID_02 (and all the other systems in a situation like this one)?
We can't use standard deviation any more, because if we use the average loss and its probability, we should use it altogether. So let's see if I can get something out of multiplying the average loss by its probability. At any rate the average loss is very close to standard deviation.
Ok, this is not working and I can't figure it out with my limited skills.
I will opt for something in between, and temporary, rather than just giving up altogether: the profit factor, which I already have adopted in my scatter plot.
...
Nope, profit factor doesn't work either, because it's no good at assessing the deviation and if it see two trades losing 1000, it is the same to profit factor as one trade losing 2000, and yet to me it's not the same, so screw profit factor, too.
I need to do a probabilistic formula, which will help me both for the theoretical ratio and the practical scatter plot, so that both will match the monte carlo blender.
It all boils down to this: ZN guy has a standard deviation of 600 with a 33% probability of losing and GBP guy has a standard deviation of 400 with a 45% probability of losing. This is in the blender makes ZN preferable to GBP: now how do i figure out a formula to assess the probability of 400 dollars losses piling up for GBP and if their rate is cumulatively higher than 600 dollars losses piling up for ZN?
Let's get thinking.
Let's simplify. Let's say zn has 20% and gbp has 50%. Of course they're still both profitable and still have those standard deviations.
Ok, so 400 has another 50% to be followed by another 400, so there's 25% chance of getting a loss of 800, and a 12.5%, after 3 trades, of getting a loss of 1200. Then of course let's not forget that gbp to reach the same profit of zn with such a poor performance is trading much more frequently and so this further increases the probability of it incurring such drawdowns.
ZN system has a 20% chance of getting a loss of 600, 0.2^2=...4% of getting another loss, for a total of 1200, so bingo! The system with the bigger loss has a lower probability of incurring the same drawdown as the system with the smaller loss (and a lower win rate).
So, ok: the calculation has to be done like this. What is the probability for a system rolling a die (a hypothetical "trade die") x times to have x losses in a row, while another systems with different number of rolls has smaller losses but bigger probability of losing?
Ok, let's pretend we're rolling it 10 times.
Roll it once, and it's 20%
twice and it's 4%...
this is all about combinations and permutations i think. But i was weak on that lesson.
The question is: what is the chance of getting x consecutive losses in x trades if each loss has a probability of x% of happening?
If I then plug in the standard deviation, I get the most likely drawdown (assuming a random order of losses)...
I am dragging it on forever just because i am math illiterate. I know, but I am doing the best that I can, and once I fix this, the formula will be ready.
Ok, there's one guy saying it here:
Tossing Coins: 5 consecutive heads out of 10 attempts?
What is the probability of getting 15 or more CONSECUTIVE heads over 40 coin tosses?
Ok, this is getting complex.
I am going to start calculating the permutations possible with 10 trades:
Combinations and Permutations Calculator
n^r so it's 1024 possible permutations.
Here's a calculator on the average number of tosses necessary to get a head run of x length:
Coin Toss Runs Calculator
Wow! I am lost again.
Ah ah, I am so lost that it makes me laugh. If I had waited to figure this stuff out, before building the systems, there's no way I would have achieved anything. Instead I built them before doing portfolio theory, so I built them because no one scared telling me that it was going to be difficult. So i built them quickly, by skipping all this. Good thing. As they say, ignorance is bliss.
However, now I want to do things properly, so I am gonna figure this out, now or another time.
The question, again, is:
1) system A loses 400 in 20 random losses out of 40 trades.
2) systems B loses 600 in 3 random losses out of 10 trades.
They make the same amount of profit, and use the same margin. So, that system A would have seemed better, because it has a smaller loss and the same Return On Investment, but obviously it has a much lower accuracy and a worse performance.
But the question is now: what formula divides ROI to tell us that system B is better than system A despite its bigger loss?
Such a formula is entirely based on probability laws, given that it has to be a formula and not an empirical monte carlo simulation. And this because I don't want to have to do a simulation each time I want to assess a system.
In this formula, size of average loss (or standard deviation if you want), % of wins and number of trades are - I think - the three ingredients needed.
Average loss should be better and simpler. Let's go and check...
...
Ok, perfect correlation, and very close, so I'll use average loss.
What is the exact data for these two systems:
ZN_ON_02:
average loss of 400
percentage of losses: 35%
trades: 50
GBP_ID_02:
average loss of 300
percentage of losses: 43%
trades: 300
Question: given that the ROI is the same, find a formula that explains why ZN_ON_02 is less likely to lose money and therefore it's a better system.
For one thing the GBP is rolling the die 6 times as many. So the probability of consecutive losing trades is much higher. Then it has a higher loss rate.
I am missing the math to do this. Let's look on khan academy.
Wait, maybe I can do this. The least common multiple of the losses is 1200. To get there, GBP has to lose 4 times in a row, and ZN only 3. But one trades 50 times, and the other one trades 300 times.
So... still lost.
Frequency Probability and Unfair Coins - YouTube
Probability Density Functions - YouTube
Oh, wow... I'd have never thought i'd get lost over this. I can't solve it. Probability seems so simple and yet it's harder and more confusing than it seems.
Ok.
Forget about the size of the average loss for now. I need to focus on the probability of x consecutive losses with x number of trades. This is related to this calculator, which however does not do exactly what I want and does not show the formula:
Coin Toss Runs Calculator
This may be closer:
Probability of L consecutive losses not occurring in N rounds? - Yahoo! Answers
[...]
Using both links, I am figuring it out. This is where I'll take the formula from:
Coin Toss Runs Calculator
His formula is closely matching my monte carlo simulator.
...
Ok I reproduced the calculator on excel:
And it matches my monte carlo VaR estimator.
Now it's a matter of making this work with my systems, and getting it to compare the ZN system to the GBP system and rating them accordingly.
[...]
I think I did it. Unbelievable how resourceful I am, despite being ignorant:
This is the file:
View attachment developing_my_own_risk_metric.xls
Now I have a univocal formula that appraises downside risk, expected shortfall, monte carlo VaR, call you whatever is correct - I don't know.
This formula does it all theoretically, without a need for me to take the trades from that system and mix them with the rest of the portfolio.
It says ZN_ON_02 is far superior to GBP_ID_05, in that, while making the same profit with the same margin requirements in the same period, it has a much smaller probability of losing 1000 dollars, despite a bigger average loss. Mostly because it trades less and it has a higher win rate.
There might still be mistakes in the formula, but most of it is done, so I can go to sleep. Oh, and the formula isn't ready yet actually, because this is just the material for the scatter plot, that will have ROI on y-axis and this VaR measure on the x-axis. It's still going to take a while, but the hard part is behind me.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
switching from reasoning to reading/learning
There's intelligence and there's no knowledge, and only so much intelligence and reasoning and persistence can compensate for my lack of knowledge.
I have realized that, no matter how much I think, since, as that professor said "risk is one word but not one number", I am not going to figure it out nor "nail it" overnight. There's these two books that I found that I need to read: Le Sourd: Performance Measurement for traditional Investment - Literature Survey, Cogneau - Hubner: The 101 Ways to Measure Portfolio Performance. They're very recent, and summarize... the titles say what they do.
I need to listen more, read more, and learn what others have thought about. I will benefit from that more than I can benefit from going crazy over these probability formulas, as I've been doing yesterday.
Indeed, the average loss and its probability is not enough to make correct VaR estimates. If I make two losses in a row, and then one win, it will also matter how big that win is, before making another two losses in a row, so I'd need to measure that as well. And, when that formula (cfr. previous post) is already so complicated as it is, can I hope to add another similar formula and turn it into an effective measure of potential drawdown, without making mistakes? Impossible. Things are getting too complex here to avoid mistakes, especially with my math illiteracy.
I am not a mathematician and I cannot compete with these guys in building a formula that will sum it all up better than they have. And, even if I hope to do this, I need to read and I will benefit from reading what they have thought about before.
So I will start reading more, and writing less. I will take a break from developing my own recipe until I've read more. I will slow down the posting on my journal, too, because it's hard to read 100 pages of a book, if I keep on commenting here every paragraph. For once, I need to keep my thoughts to myself, or I'll be too slow to achieve what I need to achieve, which is reading a large quantity of pages.
At the moment this is where I am at:
1) assessment of individual systems
I could not manage to create or find a coherent theoretical formula that does everything (e.g.: what the sharpe ratio claims to do, but doesn't do). The work I've done so far is here. I use an empirical scatter plot with Return On Investment (yearly profit divided by margin required) on the y-axis and profit factor on the x-axis:
2) assessment of portfolio
Here, too, I haven't found a coherent theoretical formula that does everything (e.g.: once again, at a portfolio level, what the sharpe ratio claims to do, but doesn't do).
Once again, I use an empirical method that I call "the blender", and that from what I've understood could be called a "Monte Carlo VaR estimator". This is what it looks like, on the final sheet:
Of course, the objective is always the same as it's been since September of last year: to devise two univocal theoretical formulas that will tell me which systems are the best, and, at a portfolio level, which systems and how many contracts I should trade.
There's intelligence and there's no knowledge, and only so much intelligence and reasoning and persistence can compensate for my lack of knowledge.
I have realized that, no matter how much I think, since, as that professor said "risk is one word but not one number", I am not going to figure it out nor "nail it" overnight. There's these two books that I found that I need to read: Le Sourd: Performance Measurement for traditional Investment - Literature Survey, Cogneau - Hubner: The 101 Ways to Measure Portfolio Performance. They're very recent, and summarize... the titles say what they do.
I need to listen more, read more, and learn what others have thought about. I will benefit from that more than I can benefit from going crazy over these probability formulas, as I've been doing yesterday.
Indeed, the average loss and its probability is not enough to make correct VaR estimates. If I make two losses in a row, and then one win, it will also matter how big that win is, before making another two losses in a row, so I'd need to measure that as well. And, when that formula (cfr. previous post) is already so complicated as it is, can I hope to add another similar formula and turn it into an effective measure of potential drawdown, without making mistakes? Impossible. Things are getting too complex here to avoid mistakes, especially with my math illiteracy.
I am not a mathematician and I cannot compete with these guys in building a formula that will sum it all up better than they have. And, even if I hope to do this, I need to read and I will benefit from reading what they have thought about before.
So I will start reading more, and writing less. I will take a break from developing my own recipe until I've read more. I will slow down the posting on my journal, too, because it's hard to read 100 pages of a book, if I keep on commenting here every paragraph. For once, I need to keep my thoughts to myself, or I'll be too slow to achieve what I need to achieve, which is reading a large quantity of pages.
At the moment this is where I am at:
1) assessment of individual systems
I could not manage to create or find a coherent theoretical formula that does everything (e.g.: what the sharpe ratio claims to do, but doesn't do). The work I've done so far is here. I use an empirical scatter plot with Return On Investment (yearly profit divided by margin required) on the y-axis and profit factor on the x-axis:
2) assessment of portfolio
Here, too, I haven't found a coherent theoretical formula that does everything (e.g.: once again, at a portfolio level, what the sharpe ratio claims to do, but doesn't do).
Once again, I use an empirical method that I call "the blender", and that from what I've understood could be called a "Monte Carlo VaR estimator". This is what it looks like, on the final sheet:
Of course, the objective is always the same as it's been since September of last year: to devise two univocal theoretical formulas that will tell me which systems are the best, and, at a portfolio level, which systems and how many contracts I should trade.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
Ok, this approach (cfr. previous post) has produced the result of more thinking... I can't help it. I am a 90% thinking and 10% learning person.
The whole situation revolves around the following problem.
We first of all need to relativize profit, which the sharpe ratio doesn't do. So we do need to measure yearly profit and divide it by the investment required (the margin). And that part is taken care of.
Then, just as the sharpe ratio does, we need to see what kind of variability we have to put up with, in order to achieve the Return On Investment described above. Once again we have to divide the variability but, unlike what I was saying here, we don't divide it by profit but by margin again.
And then, they do cancel out, because both profit and variability are divided by the same margin:
And then we end up with simply yearly profit (not average profit like in the sharpe ratio) divided by standard deviation or whatever measure of risk you want to adopt.
But then we come up against the ZN_ON_02 vs GBP_ID_02 dilemma, discussed here:
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-81.html#post1812234
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-82.html#post1812678
Whereby the ZN (short for ZN_ON_02), with a better performance, is considered worse than GBP (short for GBP_ID_02) because it has a larger standard deviation, and the probability of losses is ignored, but the blender rewards ZN, so there must be something wrong, and I was driven crazy by this all last night.
So, dividing yearly profit by standard deviation does not accurately forecast the VaR probability. And so this cannot be accepted as a ratio.
These are the ingredients and where I have doubts:
1) profit, margin: mixed together to produce Return On Investment - no problem so far.
2) variability, margin: mixed together to produce Variability On Investment.
The problem is only with variability. Of four ingredients, I have a problem with only one of them. The rest is all clear.
How do i find a way to get ZN and GBP to be assessed correctly?
I can't divide yearly profit by standard deviation, but I need to modify that standard deviation, maybe by using the sharpe ratio itself.
Indeed, recapitulating what i said yesterday, a loss of 600 that happens once every 10 trades, is not as bad as a loss of 400 happening every two trades, because losses do not always happen in the middle of two wins, but could happen consecutively. And that's more likely to happen for the system with a lower win rate. Not only this, but also the size and frequency of wins matters, because if you have a run of loses interrupted by a win before resuming, then it's important how big that win is.
In other words, I need to measure performance and include it in my ratio.
Forget for a second using the term "sharpe ratio", because it's got many more ingredients than I am using and than I can even understand (risk-free rate, annualization ingredients... things that even academics cannot agree on).
As I said, there are no doubts about the fact that I have to divide yearly profit by standard deviation or similar (since the margin on the denominator of one cancels out along with the margin on the numerator of the other), but the problem is "or similar", in that standard deviation alone doesn't assess correctly that zn is better than gbp.
So I can't divide by standard deviation but I need to divide by something that will also assess performance and as I said I can't go crazy with those probability formulas as i did yesterday, because they only did half of the job anyway (only to the downside).
So, since I can't use profit factor, because it considers a loss of 2000 the same as two losses of 1000, totally stealing from the sharpe ratio, I could divide average win by standard deviation, but then that value cannot sit on the denominator, because the higher it is the better a system is, and yet by sitting on the denominator the higher it is, the more my final ratio will go down.
So I could do something like this instead: divide standard deviation by the average trade (the end result is always the same):
I don't think this is like the sharpe ratio, but if it is, and I am reinventing the wheel, so be it. At least I understand what I am doing.
In fact, I will use the sharpe ratio, since I've got the sharpe ratio all over my sheets, and it's calculated similarly (except that on the numerator it multiplies everything by the square root of 250 days, to annualize it or whatever it is, since it's not my formula - i got from the investors).
So basically, weeks of work, to go back to the sharpe ratio but my change is that I multiply it by yearly profit. And I got to this by replacing SD in the previous ratio with SD/AV.TRADE, which, after being reversed becomes the sharpe ratio:
...
But still something doesn't seem right, because on my scatter plot now I'd have not ROI but just plain yearly profit on the y-axis. Haven't I lost track of ROI? Margin got eliminated when I also divided standard deviation by margin, which left me with just (yearly) profit over standard deviation.
But then that wasn't good and I ended up multiplying that by average trade, which is the same as saying that I am multiplying yearly profit by the sharpe ratio.
Hey, what matters to me is a formula that tells me if a system is better than another and if it will make my blender happy or not.
I'm gonna call this the travis ratio, since it keeps on changing.
Ok, here's the updated file:
View attachment developing_my_own_risk_metric.xls
This thing seems crappy but it works, you can test it on the file itself.
I am satisfied. Now I'll change my other files. And in the next few days I'll see if it works in every situation.
So in my table now the progress is as follows:
1) individual systems assessment: both theoretical and empirical methods implemented.
2) portfolio assessement: by the same rationale, both might be taken care of, but this will require a lot more testing to see if the formula works in all situations.
And this is my scatter plot, with yearly profit on the y-axis and sharpe ratio on the x-axis:
For now I am satisfied and I can take a break. Hopefully I will get started reading all those things, because there's many gaps to be filled - it's all a huge gap of ignorance. But for now i am happy.
Sharpe Ratio multiplied by yearly profit seems to be a formula capable of telling me which systems, given a similar standard deviation, will perform better on my blender. Or maybe I am so exhausted mentally that I am making myself believe that I am satisfied.
[...]
After several hours of thinking and some double-checking, putting ROI (yearly profit divided by margin) on the y-axis seems much better and it also it seems much better for my formula. I don't know what the implications are, mathematically speaking, but it's so perfect that I will change even before checking anything mathematically.
This is the new scatter plot:
Both at a formula level and at a scatter plot level, this ratio is dead on target. So it is finalized (for today) and the version is: Return On Investment (yearly profit divided by margin) multplied by the Sharpe Ratio.
I still do not know why, but I can tell by the systems that it brings up that this works perfectly. This was one neat job. I don't care about my limits, but I came up with a fantastic final outcome, and the final formula is totally univocal: I mean I could totally select systems based entirely on the ratio I get from those two factors (cfr.previous paragraph). Yeah, I ought to thank sharpe.
This is the new excel file, with the summary of my work and the template to experiment with new hypothetical systems:
View attachment developing_my_own_risk_metric_UPDATED.xls
The whole situation revolves around the following problem.
We first of all need to relativize profit, which the sharpe ratio doesn't do. So we do need to measure yearly profit and divide it by the investment required (the margin). And that part is taken care of.
Then, just as the sharpe ratio does, we need to see what kind of variability we have to put up with, in order to achieve the Return On Investment described above. Once again we have to divide the variability but, unlike what I was saying here, we don't divide it by profit but by margin again.
And then, they do cancel out, because both profit and variability are divided by the same margin:
And then we end up with simply yearly profit (not average profit like in the sharpe ratio) divided by standard deviation or whatever measure of risk you want to adopt.
But then we come up against the ZN_ON_02 vs GBP_ID_02 dilemma, discussed here:
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-81.html#post1812234
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-82.html#post1812678
Whereby the ZN (short for ZN_ON_02), with a better performance, is considered worse than GBP (short for GBP_ID_02) because it has a larger standard deviation, and the probability of losses is ignored, but the blender rewards ZN, so there must be something wrong, and I was driven crazy by this all last night.
So, dividing yearly profit by standard deviation does not accurately forecast the VaR probability. And so this cannot be accepted as a ratio.
These are the ingredients and where I have doubts:
1) profit, margin: mixed together to produce Return On Investment - no problem so far.
2) variability, margin: mixed together to produce Variability On Investment.
The problem is only with variability. Of four ingredients, I have a problem with only one of them. The rest is all clear.
How do i find a way to get ZN and GBP to be assessed correctly?
I can't divide yearly profit by standard deviation, but I need to modify that standard deviation, maybe by using the sharpe ratio itself.
Indeed, recapitulating what i said yesterday, a loss of 600 that happens once every 10 trades, is not as bad as a loss of 400 happening every two trades, because losses do not always happen in the middle of two wins, but could happen consecutively. And that's more likely to happen for the system with a lower win rate. Not only this, but also the size and frequency of wins matters, because if you have a run of loses interrupted by a win before resuming, then it's important how big that win is.
In other words, I need to measure performance and include it in my ratio.
Forget for a second using the term "sharpe ratio", because it's got many more ingredients than I am using and than I can even understand (risk-free rate, annualization ingredients... things that even academics cannot agree on).
As I said, there are no doubts about the fact that I have to divide yearly profit by standard deviation or similar (since the margin on the denominator of one cancels out along with the margin on the numerator of the other), but the problem is "or similar", in that standard deviation alone doesn't assess correctly that zn is better than gbp.
So I can't divide by standard deviation but I need to divide by something that will also assess performance and as I said I can't go crazy with those probability formulas as i did yesterday, because they only did half of the job anyway (only to the downside).
So, since I can't use profit factor, because it considers a loss of 2000 the same as two losses of 1000, totally stealing from the sharpe ratio, I could divide average win by standard deviation, but then that value cannot sit on the denominator, because the higher it is the better a system is, and yet by sitting on the denominator the higher it is, the more my final ratio will go down.
So I could do something like this instead: divide standard deviation by the average trade (the end result is always the same):
I don't think this is like the sharpe ratio, but if it is, and I am reinventing the wheel, so be it. At least I understand what I am doing.
In fact, I will use the sharpe ratio, since I've got the sharpe ratio all over my sheets, and it's calculated similarly (except that on the numerator it multiplies everything by the square root of 250 days, to annualize it or whatever it is, since it's not my formula - i got from the investors).
So basically, weeks of work, to go back to the sharpe ratio but my change is that I multiply it by yearly profit. And I got to this by replacing SD in the previous ratio with SD/AV.TRADE, which, after being reversed becomes the sharpe ratio:
...
But still something doesn't seem right, because on my scatter plot now I'd have not ROI but just plain yearly profit on the y-axis. Haven't I lost track of ROI? Margin got eliminated when I also divided standard deviation by margin, which left me with just (yearly) profit over standard deviation.
But then that wasn't good and I ended up multiplying that by average trade, which is the same as saying that I am multiplying yearly profit by the sharpe ratio.
Hey, what matters to me is a formula that tells me if a system is better than another and if it will make my blender happy or not.
I'm gonna call this the travis ratio, since it keeps on changing.
Ok, here's the updated file:
View attachment developing_my_own_risk_metric.xls
This thing seems crappy but it works, you can test it on the file itself.
I am satisfied. Now I'll change my other files. And in the next few days I'll see if it works in every situation.
So in my table now the progress is as follows:
1) individual systems assessment: both theoretical and empirical methods implemented.
2) portfolio assessement: by the same rationale, both might be taken care of, but this will require a lot more testing to see if the formula works in all situations.
And this is my scatter plot, with yearly profit on the y-axis and sharpe ratio on the x-axis:
For now I am satisfied and I can take a break. Hopefully I will get started reading all those things, because there's many gaps to be filled - it's all a huge gap of ignorance. But for now i am happy.
Sharpe Ratio multiplied by yearly profit seems to be a formula capable of telling me which systems, given a similar standard deviation, will perform better on my blender. Or maybe I am so exhausted mentally that I am making myself believe that I am satisfied.
[...]
After several hours of thinking and some double-checking, putting ROI (yearly profit divided by margin) on the y-axis seems much better and it also it seems much better for my formula. I don't know what the implications are, mathematically speaking, but it's so perfect that I will change even before checking anything mathematically.
This is the new scatter plot:
Both at a formula level and at a scatter plot level, this ratio is dead on target. So it is finalized (for today) and the version is: Return On Investment (yearly profit divided by margin) multplied by the Sharpe Ratio.
I still do not know why, but I can tell by the systems that it brings up that this works perfectly. This was one neat job. I don't care about my limits, but I came up with a fantastic final outcome, and the final formula is totally univocal: I mean I could totally select systems based entirely on the ratio I get from those two factors (cfr.previous paragraph). Yeah, I ought to thank sharpe.
This is the new excel file, with the summary of my work and the template to experiment with new hypothetical systems:
View attachment developing_my_own_risk_metric_UPDATED.xls
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
latest version of the travis ratio
Holy cow... with my persistence I am figuring it out! Despite being so bad at math...
Updated file:
View attachment developing_my_own_risk_metric.xls
Sneak preview:
Needless to say, I went over it again, and I changed it again, improving it. I have become very methodical with my excel file for testing.
I have identified four scenarios where Sharpe Ratio and Profit Factor both fail to assess a better system, and I managed to fix travis ratio so to succeed.
Amazing.
Yeah...
...
I now have tried to see if my formula could be simplified but for all I know it has to be kept as it is:
...
This is the new scatter plot:
So the scatter plot is perfect now, because if you multiply, for a given system, the value on the x-axis by the value on the y-axis you get the travis ratio.
On the x-axis you have the sharpe ratio plain and simple. On the y-axis it is the ROI divided by the standard deviation, because, as I woke up this morning, i realized that, while the formula cannot be simplified, it can be turned around, and the division can be anticipated and applied to the ROI. In other words, multiplying ROI and SR and then dividing their product by SD, is the same as dividing ROI by SD and then multiplying it by SR, because:
(10 * 20)/5=10*20/5=10/5*20=40
I don't know what it's called but it must be some x property of math.
Conceptually, it's clearer if I put it like this (dividing ROI by SD first), because I am evaluating/matching the ROI against the standard deviation required to achieve it. Then I also multiply it by 1000, but just for a problem of scale, so that's totally irrelevant.
I would sum it up like this:
This is a beautiful level of synthesis I achieved, provided that it works. In case anyone wants to test it, in the mentioned file, I have a sheet called "scenario_template":
View attachment developing_my_own_risk_metric.xls
You could post here your objections, or write me a private message. It's all ready to go, and all you need to do is tweak the trades and see if the travis ratio assesses correctly an improvement or worsening of the system.
As before, there might be problems with it that I am not seeing, but right now I don't see any.
Ennio Morricone - Monaco - Here's to you - YouTube
I can't remember another song as repetitive and yet as beautiful as this one.
JebVita's Channel - YouTube
Holy cow... with my persistence I am figuring it out! Despite being so bad at math...
Updated file:
View attachment developing_my_own_risk_metric.xls
Sneak preview:
Needless to say, I went over it again, and I changed it again, improving it. I have become very methodical with my excel file for testing.
I have identified four scenarios where Sharpe Ratio and Profit Factor both fail to assess a better system, and I managed to fix travis ratio so to succeed.
Amazing.
Yeah...
...
I now have tried to see if my formula could be simplified but for all I know it has to be kept as it is:
...
This is the new scatter plot:
So the scatter plot is perfect now, because if you multiply, for a given system, the value on the x-axis by the value on the y-axis you get the travis ratio.
On the x-axis you have the sharpe ratio plain and simple. On the y-axis it is the ROI divided by the standard deviation, because, as I woke up this morning, i realized that, while the formula cannot be simplified, it can be turned around, and the division can be anticipated and applied to the ROI. In other words, multiplying ROI and SR and then dividing their product by SD, is the same as dividing ROI by SD and then multiplying it by SR, because:
(10 * 20)/5=10*20/5=10/5*20=40
I don't know what it's called but it must be some x property of math.
Conceptually, it's clearer if I put it like this (dividing ROI by SD first), because I am evaluating/matching the ROI against the standard deviation required to achieve it. Then I also multiply it by 1000, but just for a problem of scale, so that's totally irrelevant.
I would sum it up like this:
This is a beautiful level of synthesis I achieved, provided that it works. In case anyone wants to test it, in the mentioned file, I have a sheet called "scenario_template":
View attachment developing_my_own_risk_metric.xls
You could post here your objections, or write me a private message. It's all ready to go, and all you need to do is tweak the trades and see if the travis ratio assesses correctly an improvement or worsening of the system.
As before, there might be problems with it that I am not seeing, but right now I don't see any.
Ennio Morricone - Monaco - Here's to you - YouTube
I can't remember another song as repetitive and yet as beautiful as this one.
JebVita's Channel - YouTube
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
using khan academy to get closer to portfolio theory formulas
Yeah, dude...
I was doing my daily 20 (yeah, today I got 25!) review exercises at khan academy and I came across one that reminded me of the efficient frontier formulas, which I still have not understood neither from Shiller's nor from Geanakoplos' lectures.
So I tried to adapt it and I think I came pretty close to it
In other words, for a given level of profit, you can only have a given combination of stocks and bonds producing it, but it should also be the one with the lowest variability. And I don't know how they get to that part, because they didn't help me out, neither them nor anyone else (I'll have to pay for those math private lessons).
I am not there yet, because no one really wrote a portfolio theory for dummies manual. But I'll get there, if I live long enough.
It has become a fascinating personal challenge. I want at once to make money, and to be as knowledgeable as an academic about how I do it. I am not asking for a nobel prize - just to know all the valid theory in the field that concerns me.
Yeah, dude...
I was doing my daily 20 (yeah, today I got 25!) review exercises at khan academy and I came across one that reminded me of the efficient frontier formulas, which I still have not understood neither from Shiller's nor from Geanakoplos' lectures.
So I tried to adapt it and I think I came pretty close to it
In other words, for a given level of profit, you can only have a given combination of stocks and bonds producing it, but it should also be the one with the lowest variability. And I don't know how they get to that part, because they didn't help me out, neither them nor anyone else (I'll have to pay for those math private lessons).
I am not there yet, because no one really wrote a portfolio theory for dummies manual. But I'll get there, if I live long enough.
It has become a fascinating personal challenge. I want at once to make money, and to be as knowledgeable as an academic about how I do it. I am not asking for a nobel prize - just to know all the valid theory in the field that concerns me.
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
Benchmarks for mechanical trading systems?
As I was reading a paper from my bank, that talks about risk management (an index of risk made by a consulting firm and applied to all our financial products) and investment funds, I came across the term "benchmark", and wondered, for once, what it really meant, because it seems unclear. I remember reading of how they use a benchmark to compare their performance with... or rather:
Benchmark Definition | Investopedia
1) how did they choose the benchmark, and that's ok, because probably there's a standard choice for a given investment fund, but that's debatable, too.
2) what does it mean that they performed better than a given benchmark if they achieved that return only at the end of the period, and fluctuated wildly during the period? A benchmark can hardly be a measure of return without variability. Variability is needed.
But these questions weren't answered, because I don't know enough about investment funds to find out if, as it seems to me, they abuse the concept of benchmark, and maybe that concept should not even be used, because it enables an investment fund to manipulate investors and make them think that it's trustworthy, by giving them a false sense of security.
But then I moved on to another question: is there such a thing as a benchmark for trading systems?
I think it doesn't make sense for investment funds, let alone for trading systems. However, I did my search and here's what i found:
http://www.trade2win.com/boards/mec...73-benchmarks-mechanical-trading-systems.html
It is one lively and interesting discussion on the subject, even though it didn't last too long. But, compared to the usual threads, it is an interesting one. Well, actually all the threads in that section are interesting threads.
Well, in this case I am going to end the post here, without providing much of an answer. Sometimes I write a post to just remind myself that I have an unanswered question, and in this case the question is: what the **** is a benchmark? And another question is: isn't bringing up a benchmark a scam to begin with?
I mean, an investment fund always brings up a benchmark in its brochures for all I know, nowadays online. So this would be as if I wrote in my curriculum vitae: "my math skills are better than benchmark joe" and joe is my retarded cousin. I would never write "my math skills are worse than mark", whoever mark is, so using a benchmark only makes sense if it's chosen by an independent judge - it makes no sense if the benchmark is chosen by the fund itself.
So if that's the case, since I am against dishonesty, **** the benchmark and **** all the investment funds using it.
...
Bingo!
Since I am a 90% reasoning and 10% reading guy, I did my 10% of reading and bingo!
Fund Benchmarks | Mutual Fund Performance
As I was reading a paper from my bank, that talks about risk management (an index of risk made by a consulting firm and applied to all our financial products) and investment funds, I came across the term "benchmark", and wondered, for once, what it really meant, because it seems unclear. I remember reading of how they use a benchmark to compare their performance with... or rather:
Benchmark Definition | Investopedia
But the problem is that a benchmark is a little more than that for investment funds: it is a way of saying that they're doing better than the benchmark. But then two questions arise:
1) how did they choose the benchmark, and that's ok, because probably there's a standard choice for a given investment fund, but that's debatable, too.
2) what does it mean that they performed better than a given benchmark if they achieved that return only at the end of the period, and fluctuated wildly during the period? A benchmark can hardly be a measure of return without variability. Variability is needed.
But these questions weren't answered, because I don't know enough about investment funds to find out if, as it seems to me, they abuse the concept of benchmark, and maybe that concept should not even be used, because it enables an investment fund to manipulate investors and make them think that it's trustworthy, by giving them a false sense of security.
But then I moved on to another question: is there such a thing as a benchmark for trading systems?
I think it doesn't make sense for investment funds, let alone for trading systems. However, I did my search and here's what i found:
http://www.trade2win.com/boards/mec...73-benchmarks-mechanical-trading-systems.html
It is one lively and interesting discussion on the subject, even though it didn't last too long. But, compared to the usual threads, it is an interesting one. Well, actually all the threads in that section are interesting threads.
Well, in this case I am going to end the post here, without providing much of an answer. Sometimes I write a post to just remind myself that I have an unanswered question, and in this case the question is: what the **** is a benchmark? And another question is: isn't bringing up a benchmark a scam to begin with?
I mean, an investment fund always brings up a benchmark in its brochures for all I know, nowadays online. So this would be as if I wrote in my curriculum vitae: "my math skills are better than benchmark joe" and joe is my retarded cousin. I would never write "my math skills are worse than mark", whoever mark is, so using a benchmark only makes sense if it's chosen by an independent judge - it makes no sense if the benchmark is chosen by the fund itself.
So if that's the case, since I am against dishonesty, **** the benchmark and **** all the investment funds using it.
...
Bingo!
Since I am a 90% reasoning and 10% reading guy, I did my 10% of reading and bingo!
Fund Benchmarks | Mutual Fund Performance
Even in all my ignorance, my brilliant reasoning (hey, there's no readers left to compliment me so I have to take care of it) fills up the gaps. I knew the answer by simply reasoning before even doing any reading on it. I realized that the concept of benchmark doesn't make any sense in the way it's used by the financial community. It would be like having a benchmark "joe" in my curriculum, whom I compare myself to. This excellent brain of mine never ceases to surprise me.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
progress but now I am stuck on travis ratio
I am totally stuck on scenario "trades f":
View attachment developing_my_own_risk_metric.xls
I went about checking on my systems if things worked according to my metric and I nailed it - I nailed the problem, but I can't solve it now.
The ingredients are clearly defined and are always the same: ROI (yearly profit divided by margin), sharpe ratio, and standard deviation.
But I just cannot find the right recipe. To make all scenarios work and be appraised correctly.
Essentially I multiply ROI by Sharpe Ratio, and then I divide both by... and here comes the problem.
If I divide them by the standard deviation, I have a problem with scenario "trades f", because it's not fair that a system with twice the leverage gets punished because, obviously, the standard deviation of its trades is also going to double.
So the system is the same, but it gets rated with half the score, because it's unfairly punished by the doubled standard deviation (normal consequence of the doubled leverage).
If instead I take ROI * SR and then divide them by a standard deviation that first gets divided by margin, then "trades f" works, but all other scenarios do not work and it's as if margin gets disabled because ROI is profit divided by margin, but standard deviation divided by margin, when you flip it.... look here:
So, I know I need standard deviation, but also I need to preserve ROI and margin.
And I can't find a way to make all the scenarios happy... I would need a math guy, but I got into fights with all the math people I knew when I asked them what is -2^2 and they answered incorrectly that it's 4.
So, I am stuck here for today:
I am totally stuck on scenario "trades f":
View attachment developing_my_own_risk_metric.xls
I went about checking on my systems if things worked according to my metric and I nailed it - I nailed the problem, but I can't solve it now.
The ingredients are clearly defined and are always the same: ROI (yearly profit divided by margin), sharpe ratio, and standard deviation.
But I just cannot find the right recipe. To make all scenarios work and be appraised correctly.
Essentially I multiply ROI by Sharpe Ratio, and then I divide both by... and here comes the problem.
If I divide them by the standard deviation, I have a problem with scenario "trades f", because it's not fair that a system with twice the leverage gets punished because, obviously, the standard deviation of its trades is also going to double.
So the system is the same, but it gets rated with half the score, because it's unfairly punished by the doubled standard deviation (normal consequence of the doubled leverage).
If instead I take ROI * SR and then divide them by a standard deviation that first gets divided by margin, then "trades f" works, but all other scenarios do not work and it's as if margin gets disabled because ROI is profit divided by margin, but standard deviation divided by margin, when you flip it.... look here:
So, I know I need standard deviation, but also I need to preserve ROI and margin.
And I can't find a way to make all the scenarios happy... I would need a math guy, but I got into fights with all the math people I knew when I asked them what is -2^2 and they answered incorrectly that it's 4.
So, I am stuck here for today:
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
latest developments
Latest developments:
View attachment developing_my_own_risk_metric.xls
I totally got rid of sharpe ratio and even standard deviation, in favor of average loss and profit factor.
Latest developments:
View attachment developing_my_own_risk_metric.xls
I totally got rid of sharpe ratio and even standard deviation, in favor of average loss and profit factor.
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
Yet another change...
View attachment developing_my_own_risk_metric.xls
This time it's empirical. In other words the formula won't exactly assess if a system is precisely twice as good as another, which was done by previous formulas, which however had a huge problem when the margin or the standard deviation changed.
I went from theoretical, neat and beautiful, which I could not get to work (probably due to my math illiteracy), to an empirical and practical formula which works very well, and assesses my systems very correctly and close to what I could be doing by analyzing them one by one (as I did, so I know it matches it):
Once again, the ratio works as a whole, but won't assess correctly if a system is twice as good as another. For example if I do trades +2000 and -1000 and require a margin of 2000 and then 1000, the ratio won't tell you that the second system is twice as good, as it should - and this is my big regret. It will tell you that the system went from 2.8 to 4. It's close but not scientific - it's crap. But overall, it's better than the sharpe ratio, or at least in the ways that I could get the sharpe ratio to work, since on the web I could find about ten different versions of it.
Or let's say a system does +2000 and -1000 and another one does +2000 and -1000, and then again +2000 and -1000 - both in the same period and with the same margin. Sharpe ratio will tell you they are the same.
My ratio should tell you that the second system is twice as good as the first, but instead (another regret) it tells you that... actually in this case it works perfectly (cfr. attached excel file, see link above, at the start of this post).
However what counts is that overall it assesses my systems very well.
Another thing I have to add: the usual rule applies and if you multiply what's on the x-axis by what's on the y-axis, you get the value of my ratio. By this rationale, a system that has a profit factor of 4 is awesome, but if it trades three times a year and makes little money compared to its margin and its standard deviation, then it's as if it didn't exist (and this might be a limit of my ratio in some rare cases).
View attachment developing_my_own_risk_metric.xls
This time it's empirical. In other words the formula won't exactly assess if a system is precisely twice as good as another, which was done by previous formulas, which however had a huge problem when the margin or the standard deviation changed.
I went from theoretical, neat and beautiful, which I could not get to work (probably due to my math illiteracy), to an empirical and practical formula which works very well, and assesses my systems very correctly and close to what I could be doing by analyzing them one by one (as I did, so I know it matches it):
Once again, the ratio works as a whole, but won't assess correctly if a system is twice as good as another. For example if I do trades +2000 and -1000 and require a margin of 2000 and then 1000, the ratio won't tell you that the second system is twice as good, as it should - and this is my big regret. It will tell you that the system went from 2.8 to 4. It's close but not scientific - it's crap. But overall, it's better than the sharpe ratio, or at least in the ways that I could get the sharpe ratio to work, since on the web I could find about ten different versions of it.
Or let's say a system does +2000 and -1000 and another one does +2000 and -1000, and then again +2000 and -1000 - both in the same period and with the same margin. Sharpe ratio will tell you they are the same.
My ratio should tell you that the second system is twice as good as the first, but instead (another regret) it tells you that... actually in this case it works perfectly (cfr. attached excel file, see link above, at the start of this post).
However what counts is that overall it assesses my systems very well.
Another thing I have to add: the usual rule applies and if you multiply what's on the x-axis by what's on the y-axis, you get the value of my ratio. By this rationale, a system that has a profit factor of 4 is awesome, but if it trades three times a year and makes little money compared to its margin and its standard deviation, then it's as if it didn't exist (and this might be a limit of my ratio in some rare cases).
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
No one answers my emails and requests for math help, I can't get any sleep at night, I am overweight, I am feeling some sort of pain around my heart area, I may not have much longer to live. All in all, I considered my chances of blowing out (and chances of dying), and I have enabled another two systems, twins, doing the same thing on CL and NG. These systems are excellent but they also have a bit of a leverage which right now I may not be able to afford. In other words I am increasing my "monte carlo blender" chances of blowing out from 2% to 8%. This is just as far as backtesting, which means in real trading they might be as high as 20%.
I might as well do so, because:
1) If i don't, I might engage in discretionary trading out of frustration - frustrationary trading. And then I'd blow out my account anyway, so I might as well let my systems gamble rather than me doing it.
2) I might die any time.
3) risk of shooting rampage at work, and if that happens, I can't keep trading from jail.
With the new systems on, there's going to be a lot of action. And I either blow out within the next 3 weeks, or it will be heaven. I mean making about 6k per month. That's a huge amount of money for me.
The risk is only in the first month, because if it doesn't go down, then it takes off, and if it takes off, it's not looking back, and the chances of blowing out will grow rapidly smaller and smaller, infinitesimal.
...
This journal is also pissing me off - I mean the forum. Not the management, I mean the readers. There's three readers, but they don't write. Yeah, of course they're afraid of getting into arguments with me, but I'd like some interaction every once in a while. Besides, the arguments have been very frequent and angry, but only with those who disrespected me. Especially now I'd like some interaction, because I need math help with those formulas, and instead no one is even answering my private messages. Everyone was busting my balls in September and later, with bull**** advice, but when you ask for practical help (not just berating), no one is willing or capable of giving you any. They just want to spend time harassing you with bull**** advice. Real help? They don't have enough time or intelligence for that.
Because of all these reasons, especially if I die, I might write less in the future. Maybe like a few posts per week.
And what should I say of the ratings? This forum is full of despicable people, according to how I've been rated. So, unlike what I thought at the start, the forum is not as crappy as the real world, where the idiots are 95%, but it's closer to it than I thought. Maybe here the idiots are 66%. No: 75%.
Listen, I'll take the rest of the day off. I am going home, getting drunk with some Heineken that I'll buy now, and then I'm going to try to sleep, because tonight there's two colleagues coming to have dinner with me. I am totally disgusted with these idiots in the room next to mine, who are laughing loudly. Bunch of monkeys.
I might as well do so, because:
1) If i don't, I might engage in discretionary trading out of frustration - frustrationary trading. And then I'd blow out my account anyway, so I might as well let my systems gamble rather than me doing it.
2) I might die any time.
3) risk of shooting rampage at work, and if that happens, I can't keep trading from jail.
With the new systems on, there's going to be a lot of action. And I either blow out within the next 3 weeks, or it will be heaven. I mean making about 6k per month. That's a huge amount of money for me.
The risk is only in the first month, because if it doesn't go down, then it takes off, and if it takes off, it's not looking back, and the chances of blowing out will grow rapidly smaller and smaller, infinitesimal.
...
This journal is also pissing me off - I mean the forum. Not the management, I mean the readers. There's three readers, but they don't write. Yeah, of course they're afraid of getting into arguments with me, but I'd like some interaction every once in a while. Besides, the arguments have been very frequent and angry, but only with those who disrespected me. Especially now I'd like some interaction, because I need math help with those formulas, and instead no one is even answering my private messages. Everyone was busting my balls in September and later, with bull**** advice, but when you ask for practical help (not just berating), no one is willing or capable of giving you any. They just want to spend time harassing you with bull**** advice. Real help? They don't have enough time or intelligence for that.
Because of all these reasons, especially if I die, I might write less in the future. Maybe like a few posts per week.
And what should I say of the ratings? This forum is full of despicable people, according to how I've been rated. So, unlike what I thought at the start, the forum is not as crappy as the real world, where the idiots are 95%, but it's closer to it than I thought. Maybe here the idiots are 66%. No: 75%.
Listen, I'll take the rest of the day off. I am going home, getting drunk with some Heineken that I'll buy now, and then I'm going to try to sleep, because tonight there's two colleagues coming to have dinner with me. I am totally disgusted with these idiots in the room next to mine, who are laughing loudly. Bunch of monkeys.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
more on summation notation
Continuing from here:
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-70.html#post1804764
Wow, I am back to life with this link:
sigma(n=1to5)(n+2)^2 - Wolfram|Alpha
They have a wonderfully simple calculator of summation notation on wolfram alpha. This is fascinating:
And you know, summation notation is the first step in the conquest of financial math.
...
And with this I've finished these exercises, too:
Summation, or Sigma, Notation
This is useful, too:
Instructions for using Windows Alt Codes
ALT Codes - Alt Codes for Maths / Mathematics
Alt-codes for characters
I need it for entering "infinity" and similar... look: on my HP laptop, I first press "fn" and bloc num, and then ALT + 236, and here it is:
∞
Continuing from here:
http://www.trade2win.com/boards/trading-journals/140032-my-journal-3-a-70.html#post1804764
Wow, I am back to life with this link:
sigma(n=1to5)(n+2)^2 - Wolfram|Alpha
They have a wonderfully simple calculator of summation notation on wolfram alpha. This is fascinating:
And you know, summation notation is the first step in the conquest of financial math.
...
And with this I've finished these exercises, too:
Summation, or Sigma, Notation
This is useful, too:
Instructions for using Windows Alt Codes
ALT Codes - Alt Codes for Maths / Mathematics
Alt-codes for characters
I need it for entering "infinity" and similar... look: on my HP laptop, I first press "fn" and bloc num, and then ALT + 236, and here it is:
∞
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
Very odd... I looked several times at the feature on the bottom of this page, "Currently Active Users Viewing This Thread", and it never showed anyone. It must be the first time it happens for an entire day - I don't remember anything like this before. Usually I see at least one guest. Today, I must have checked 10 times already, and no one was ever reading my journal. This is all random of course, even though I would instinctively take it personally. Or it could be a technical problem. Or my journal has been removed from the main page of the journals.
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
the different meaning of "risk" in finance
After increasing both the variability (and therefore "risk") and expected return of my systems I am feeling a little bit better: I feel like they'll take care of the action for me. Of course I am feeling better about the increase in expected profit and not the increase in variability.
Oh, and today this choice of enabling those two extra systems has paid off immediately with a lot of money.
I went to dinner with my two colleagues, and now I am enjoying this celebrity island tv show (we only have it in Italy and I don't know the english translation):
Video Rai.TV - L'isola dei famosi 2012 - Nuova zizzania, stessa protagonista: Nina
I will probably manage to sleep a little better tonight. Finally. I found peace as far as trading by enabling more systems. Sometimes increasing risk by the systems can help you decrease your disretionary trading, and therefore an unprofitable activity: I would not call it "risky" because in finance "risk" is intended as variability rather than as "probability of losing". It's the probability of a temporary loss, which is also known as variability.
Indeed, increasing risk doesn't mean you go from a system that makes a yearly 20% to a system that makes a yearly 10%. In common ordinary everyday sense, that is what you would expect, because you're increasing the risk of losing. In finance it's almost the opposite: in fact most typically you increase risk by going from a system that makes a yearly 20% to a system that makes a yearly 30% - not for that reason itself of course but for the fact that, in order to make that yearly 30%, you're often trading something that will go up and down more than in the other case, because otherwise you wouldn't it. Of course it's not always like this. But what I am saying is that in finance "risk" is not connected (as an outsider would instead naturally think) to the probability of profit but to the variability of profit, and therefore to the probability of profit in the short term, too.
Let me see if I can find any traces of my concepts and understanding in the definition given for "risk" by the online encyclopedias:
http://en.wikipedia.org/wiki/Risk#Economic_risk
http://en.wikipedia.org/wiki/Risk#Finance
I think the risk I was referring to is called "equity risk", and it's only one of the many risks included in "financial risk", but I am not satisfied by the definition given:
http://en.wikipedia.org/wiki/Equity_risk
I found a great video here:
http://www.investopedia.com/video/play/risk-and-time-horizon#axzz1pha6UEjq
It says a meaningful thing about how risk (intended as variability) should be compensated by expected return - but not always is, so it's wrong to say as they do here:
http://www.investopedia.com/terms/r/risk.asp#axzz1pha6UEjq
Summary. Risk in finance is not like in regular life, where a bigger risk of dying is a higher probability of dying while doing something. An investment with a bigger risk is not an investment with a bigger risk of losing the money (as some of those definitions said) - otherwise we wouldn't be talking about it, because no one would be interested in it. No wait there is some truth to that.
Yes, there is a higher risk of losing the money just like we use the term in everyday life. And this is the permanent side of risk: the sense is that if you kept the money in the bank instead of trading futures with your trading systems, it'd be safer - and this is true. Then there is the variability meaning. It's also an undesirable side of risk, but it's only referring to the potential temporary loss of money. Furthermore the markets are not efficient and you could find a way to get a higher return with a lower risk. For example, my systems are safer than the argentine bonds, but they return more. This is the first meaning: they don't fluctuate, so they don't have the variability problem. Another example for the variability is that they will make money every month, while a stock won't, so this is less safe in every aspect.
Well, it's not totally clear, but clearer than before. The way i see it is that there are these two aspects to (trading) risk:
1) permanent loss
2) temporary loss (variability)
1) for the first, the only safer thing I can see relative to my systems is keeping the money on my savings account
2) for the second, it's just the same - of course the selection of systems will affect my variability, but there's no stocks that could match the little variability of my systems, and bonds would have less variability of course, but they don't produce any return so I am not interested.
...
I feel my readers have decreased, but as long as there'll be the hope for 1 reader, I'll feel that it's worth it to keep writing here. My writing style may even improve as the consequence of this sensation I am having that fewer and fewer people are reading this journal. I think it's making me write more what I feel like writing, because I feel "since I won't get any readers by writing about this and that, let's write whatever I feel like": I may even stop posting some weekly updates, who knows.
After increasing both the variability (and therefore "risk") and expected return of my systems I am feeling a little bit better: I feel like they'll take care of the action for me. Of course I am feeling better about the increase in expected profit and not the increase in variability.
Oh, and today this choice of enabling those two extra systems has paid off immediately with a lot of money.
I went to dinner with my two colleagues, and now I am enjoying this celebrity island tv show (we only have it in Italy and I don't know the english translation):
Video Rai.TV - L'isola dei famosi 2012 - Nuova zizzania, stessa protagonista: Nina
I will probably manage to sleep a little better tonight. Finally. I found peace as far as trading by enabling more systems. Sometimes increasing risk by the systems can help you decrease your disretionary trading, and therefore an unprofitable activity: I would not call it "risky" because in finance "risk" is intended as variability rather than as "probability of losing". It's the probability of a temporary loss, which is also known as variability.
Indeed, increasing risk doesn't mean you go from a system that makes a yearly 20% to a system that makes a yearly 10%. In common ordinary everyday sense, that is what you would expect, because you're increasing the risk of losing. In finance it's almost the opposite: in fact most typically you increase risk by going from a system that makes a yearly 20% to a system that makes a yearly 30% - not for that reason itself of course but for the fact that, in order to make that yearly 30%, you're often trading something that will go up and down more than in the other case, because otherwise you wouldn't it. Of course it's not always like this. But what I am saying is that in finance "risk" is not connected (as an outsider would instead naturally think) to the probability of profit but to the variability of profit, and therefore to the probability of profit in the short term, too.
Let me see if I can find any traces of my concepts and understanding in the definition given for "risk" by the online encyclopedias:
http://en.wikipedia.org/wiki/Risk#Economic_risk
Different meaning here. This is in line with the intuitive (but not financial) idea of a "probability of permanent loss", whereas financially it's "probability of temporary loss". Basically it all boils down to a distinction of "permanent" vs "temporary".
http://en.wikipedia.org/wiki/Risk#Finance
Much closer but not exactly what I said.
I think the risk I was referring to is called "equity risk", and it's only one of the many risks included in "financial risk", but I am not satisfied by the definition given:
http://en.wikipedia.org/wiki/Equity_risk
I found a great video here:
http://www.investopedia.com/video/play/risk-and-time-horizon#axzz1pha6UEjq
It says a meaningful thing about how risk (intended as variability) should be compensated by expected return - but not always is, so it's wrong to say as they do here:
http://www.investopedia.com/terms/r/risk.asp#axzz1pha6UEjq
There's a lot of stupid investors around who increase their risk without increasing the expected return. They should have phrased it "the greater the potential return... should be". The video sets the record straight. Also, the video makes a good remark about the time horizon and how it should influence your evaluation of an investment based on its risk ("variability" should be the term used - risk is so abused, misleading and unclear!): the longer your horizon, the more variability you can accept. Of course in my case a long horizon may even mean just a few days. So in my case risk assessment is not really about the time horizon but the size of the account, but this is because I am investing in futures, and so I cannot go down too much because I am trading on margin. It's almost as if a time horizon were forced on me, because it's like I am borrowing money when I use margin.
Summary. Risk in finance is not like in regular life, where a bigger risk of dying is a higher probability of dying while doing something. An investment with a bigger risk is not an investment with a bigger risk of losing the money (as some of those definitions said) - otherwise we wouldn't be talking about it, because no one would be interested in it. No wait there is some truth to that.
Yes, there is a higher risk of losing the money just like we use the term in everyday life. And this is the permanent side of risk: the sense is that if you kept the money in the bank instead of trading futures with your trading systems, it'd be safer - and this is true. Then there is the variability meaning. It's also an undesirable side of risk, but it's only referring to the potential temporary loss of money. Furthermore the markets are not efficient and you could find a way to get a higher return with a lower risk. For example, my systems are safer than the argentine bonds, but they return more. This is the first meaning: they don't fluctuate, so they don't have the variability problem. Another example for the variability is that they will make money every month, while a stock won't, so this is less safe in every aspect.
Well, it's not totally clear, but clearer than before. The way i see it is that there are these two aspects to (trading) risk:
1) permanent loss
2) temporary loss (variability)
1) for the first, the only safer thing I can see relative to my systems is keeping the money on my savings account
2) for the second, it's just the same - of course the selection of systems will affect my variability, but there's no stocks that could match the little variability of my systems, and bonds would have less variability of course, but they don't produce any return so I am not interested.
...
I feel my readers have decreased, but as long as there'll be the hope for 1 reader, I'll feel that it's worth it to keep writing here. My writing style may even improve as the consequence of this sensation I am having that fewer and fewer people are reading this journal. I think it's making me write more what I feel like writing, because I feel "since I won't get any readers by writing about this and that, let's write whatever I feel like": I may even stop posting some weekly updates, who knows.
Last edited:
Yamato
Legendary member
- Messages
- 9,840
- Likes
- 246
I just woke up, and had this incredibly intelligent dream that I must briefly mention before going to work. I was calculating, correctly, how much my father spent to send me to study in the States for all those years I've studied there, and it's incredible but I've realized, while dreaming, that he has spent all his savings (he was just a university professor and my mother for many years wasn't working) for my tuition.
This is something I had never realized before. One more reason that I am glad to have stopped being spendthrift, one more reason to make some serious money and show him that I have put to good use what he has done for me. He was still an asshole, but other than that, he has done everything for me. It's amazing what a parent can do for his son, even when he abuses him emotionally.
This is something I had never realized before. One more reason that I am glad to have stopped being spendthrift, one more reason to make some serious money and show him that I have put to good use what he has done for me. He was still an asshole, but other than that, he has done everything for me. It's amazing what a parent can do for his son, even when he abuses him emotionally.
Last edited:
