Noob questions

[momothebored]

1. What do people mean by "price-action"?

2. Is there any use to back-testing and creating models from there?

e.g. I take a look at the weekly and daily charts of a large-cap stock, and plot what would have worked best for the last 10 years.

So i come up with, for example, that if you bought whenever EMA(10d) crossed above EMA(20d) and sold whenever MACDH ticked down, you would have made money 99% of the time. Any value doing this?


3. What methods and books are generally considered the best? I've been using Elder's triple screen, am severely disappointed. And yes I've been implementing correctly.

tyvm!
:clap:

 

[pboyles]

So i come up with, for example, that if you bought whenever EMA(10d) crossed above EMA(20d) and sold whenever MACDH ticked down, you would have made money 99% of the time. Any value doing this?

:clap:

No because you are curve fitting. If you use enough criteria anything will look profitable.

 

[momothebored]

No because you are curve fitting. If you use enough criteria anything will look profitable.

So why wouldn't it work again?

I did in fact do it for a stock, and 22/23 times in the past, it worked.
So why not carry on doing it?

At the very least, shouldn't you get a robust model?

 

[the hare]

1. What do people mean by "price-action"?

2. Is there any use to back-testing and creating models from there?

e.g. I take a look at the weekly and daily charts of a large-cap stock, and plot what would have worked best for the last 10 years.

So i come up with, for example, that if you bought whenever EMA(10d) crossed above EMA(20d) and sold whenever MACDH ticked down, you would have made money 99% of the time. Any value doing this?


3. What methods and books are generally considered the best? I've been using Elder's triple screen, am severely disappointed. And yes I've been implementing correctly.

tyvm!
:clap:

1) different people mean different things by price action. Unfortunately, there doesn't seam to be a commonly accepted definition. Some people are focused on bar by bar patterns, whilst at the other end of the spectrum there are others (me included) who tend to think in terms of longer term wave structure.

2) back testing is useful but just not in the way you are suggesting. It certainly allows you to evaluate mechanical strategies very quickly, partially eliminating the need for time consuming forward testing. If done correctly, you might even get an idea of the expectancy and variance you might achieve in live trading, which helps in monitoring performance against expectations.

The main disadvantage is that back tests are based on the possibly flawed assumption that what happened in the past will happen in the future, and this assumption can cause potential risks as the results obtained are dependent on the characteristics of teh time-series which where exhibited during the testing period. Financial time series are statistically non stationary, often rendering the results from mechanical strategies little better than random (which is a good thing in my book 😛).

Although back-testing does not necessarily allow you to predict how a strategy will perform under future conditions, its primary benefit lies in understanding the vulnerabilities of a strategy through a simulated encounter with real-world conditions of the past. This theoretically enables the designer of a strategy to learn from their mistakes without actually having to make those mistakes with real money.

I'm not a great believer in back tests, but they are a necessary evil. [Cue bleedin obvious comments from vendors with their intelligent nonsense agenda]

Curve fitting parameters as you suggest is a sure way to failure. A lot of new traders dont seam to realize just how simple it is to design a trading system that back-tests well. Those systems invariably fail spectacularly in forward testing.

3]I'm not an advocate of burning books, but I could possibly make an exception in the case of trading related literature. Forget books, and for heavens sake forget trading forums other than 4 d lulz (although you might want to check out bbmacs thread here at the zoo for a few ideas about how the elder triple screen approach can be modified to something that makes a bit more sense)!

You just need to trade, and you need to get stuck into some serious work. Why do you want to do this anyway ?

 

[momothebored]

Curve fitting parameters as you suggest is a sure way to failure. A lot of new traders dont seam to realize just how simple it is to design a trading system that back-tests well. Those systems invariably fail spectacularly in forward testing.

3]I'm not an advocate of burning books, but I could possibly make an exception in the case of trading related literature. Forget books, and for heavens sake forget trading forums other than 4 d lulz (although you might want to check out bbmacs thread here at the zoo for a few ideas about how the elder triple screen approach can be modified to something that makes a bit more sense)!

You just need to trade, and you need to get stuck into some serious work. Why do you want to do this anyway ?

Could you please elaborate on WHY curve fitting fails?
The entire concept of the past replicating itself again mechanically in the future is the basis of TA.

Curve fitting could arguably be the creation of a new indicator in itself. Instead of a MACD, my new regression model would be the indicator. Same concept, no?


No, I need books. I can't just look at a chart without a framework (at least mentally) and trade. Not gifted that way.

Why do I want to do this?
I'm a fundamental analyst with a hedge fund. My understanding of technicals is functional. As mentioned triple screen, not much else. So when I do become a FM, I do hope to have a much more profound grasp.

 

[rawrschach]

At the very least, shouldn't you get a robust model?

If you did what you said where would your stops be? What happens if you sell on a MA crossover then the MA goes back up and there is no more cross?

Your regression model assumes what, normal i.i.d distribution? Maybe with a stochastic element, which, let me guess, is also i.i.d.? You want robustness you should start from the basis of fat tails and close to infinite variance. Livermore's mothodology is robust imo and would be ok for a longer term investment strat.

My fave books:

Reminiscences of a stock operator - Lefevre
How to trade in stocks - livermore
Popular delusions - mackay
art of contrary thinking - o'neill

 

[momothebored]

(i) MA Cross: once you're out, you're out till the next BUY signal. In the meantime, presumably you'd trade other things but you'd already be well up even if you miss the rest of the boat till the next entry.

Don't get me wrong, I'm not promoting curve fitting, i'm trying to understand why it doesn't work.

(ii) Not sure I follow why you'd need to assume a p(x) distribution at all. Are you referring to the outliers?

By regression model I simply refer to a very basic "line of best fit" approach, not an actual mathematically rigorous model. Sorry if people typically give a full system? As mentioned this isn't my area of expertise.

based on curve-fitting that historically:
A. P(Success) = 24/25 (you made money taking into account brokerage)
B. P(Failure)= 1/25

I can't remember what example i gave earlier in the thread, but assume that you entered on an EMA(10W),(20W) cross. In this case you exit on a MACDH downtick or a EMA(5W), (20W) bearish cross. You'd also have a trailing stop loss.

So what contingencies wouldn't be covered?


So you'd recommend Livermore's model?
thanks, will take a look. Do you actually use his approach in your trading?

If you did what you said where would your stops be? What happens if you sell on a MA crossover then the MA goes back up and there is no more cross?

Your regression model assumes what, normal i.i.d distribution? Maybe with a stochastic element, which, let me guess, is also i.i.d.? You want robustness you should start from the basis of fat tails and close to infinite variance. Livermore's mothodology is robust imo and would be ok for a longer term investment strat.

My fave books:

Reminiscences of a stock operator - Lefevre
How to trade in stocks - livermore
Popular delusions - mackay
art of contrary thinking - o'neill

 

[ronblack]

1. What do people mean by "price-action"?

2. Is there any use to back-testing and creating models from there?

e.g. I take a look at the weekly and daily charts of a large-cap stock, and plot what would have worked best for the last 10 years.

So i come up with, for example, that if you bought whenever EMA(10d) crossed above EMA(20d) and sold whenever MACDH ticked down, you would have made money 99% of the time. Any value doing this?


3. What methods and books are generally considered the best? I've been using Elder's triple screen, am severely disappointed. And yes I've been implementing correctly.

tyvm!
:clap:

1. Anything that is caused by price like patterns, trendlines, crossing of indicators, you name it.

2. Yes but it is not that easy and it takes a lot of work. Your system is a random artifact of selection and data snooping bias.

3. The only method that works is you actually trading a small test account and developing systems. Expect a learning curve of about 3 to 10 years depending on your skills and background.

 

[rawrschach]

Using p(x) as a single number is a bit meaningless, especially in the case of the markets. You need to look at the probability distributions.


Ya I use his model longer term.

will msg abt curve fitting later

(i) MA Cross: once you're out, you're out till the next BUY signal. In the meantime, presumably you'd trade other things but you'd already be well up even if you miss the rest of the boat till the next entry.

Don't get me wrong, I'm not promoting curve fitting, i'm trying to understand why it doesn't work.

(ii) Not sure I follow why you'd need to assume a p(x) distribution at all. Are you referring to the outliers?

By regression model I simply refer to a very basic "line of best fit" approach, not an actual mathematically rigorous model. Sorry if people typically give a full system? As mentioned this isn't my area of expertise.

based on curve-fitting that historically:
A. P(Success) = 24/25 (you made money taking into account brokerage)
B. P(Failure)= 1/25

I can't remember what example i gave earlier in the thread, but assume that you entered on an EMA(10W),(20W) cross. In this case you exit on a MACDH downtick or a EMA(5W), (20W) bearish cross. You'd also have a trailing stop loss.

So what contingencies wouldn't be covered?


So you'd recommend Livermore's model?
thanks, will take a look. Do you actually use his approach in your trading?

 

[FringFX]

Could you please elaborate on WHY curve fitting fails?
The entire concept of the past replicating itself again mechanically in the future is the basis of TA.

Curve fitting could arguably be the creation of a new indicator in itself.

Price action literally refers to the movement of price and trading off of it.

curve fitting fails because it is far from realistic.... it does not take into account externalities and eventualities. a lot of things happen in between the MA crossovers and price movement - beyond what you see in a historical chart. that is why it is essential to backtest and do live testing. live testing will account for those that backtesting missed.

assuming you get to create a 'new' indicator from curve fitting, there is a good probability it will fail (just like several other indicators). it's like a failure extracted from another failure... or cause and effect... or garbage in garbage out.