Vrothdar's Journal

[Vrothdar]

I'm starting this journal as a record of my progress from knowing a bit to, with any luck, knowing a bit more with the intention of using the time required to maintain it as an opportunity to reflect on what I've been doing. Thoughts, suggestions and general comments are, of course, most welcome as I value alternative viewpoints highly - I'd value them even more so if I weren't quite so stubborn

I'll start by outlining what I've done so far, where that's got me and what my aims are for the future. Subsequent posts will describe what I've been doing, my successes (hopefully) and my failures (definately!) and of course responses to any discussions that my journal may generate.

The journey so far

When I first started considering trading I was very, very green around the ears. I couldn't understand why a spread betting company would be unlikely to peg their prices against the cash value of an index, didn't understand how many, if any, of the "real" markets worked, understood very little trading jargon and had been trying to come up with a system of some sort using advfn charts and going back up to a month and writing down my trades. Basically I thought I knew more than I did.

After a few weeks of messing about like that I read a post on these boards that changed my outlook along the lines of "It doesn't matter what market you trade in so long as you understand how the market you're trading in works". Upon reading that post I came to the realisation that I hadn't got a clue. I then set out to discover a market I understood the workings of, what I want to get out of trading and take serious consideration to the limits of my knowledge.

Since then I've come to the following conclusions:

1. I'm interesting in Forex and have a good understanding of how the market works. This is the market I'm going to focus all my efforts on.
   
2. I'll rely technical analysis only. The reason for this is that I'm a perfectionist and know that if I try to take into account fundamentals I'll either suffer from paralysis of analysis or end up applying my trading methods inconsistently.
   
3. I'd be more comfortable trading on an intraday timescale. I'm more comfortable with the idea of small profits and small loses than large profits and large losses; I don't want to get rich (or poor) quickly. In my opinion this timescale will help minimise the effect of fundamental influences on price, such as news stories, on the long term outcome of my trading.
4. I could follow a conversation between experienced trades and perhaps contribute a little to it too. To somebody who has no interest in trading I could sound like I know my stuff. In other words: I'm aware of my limits and in a position to make some decent progress.

My immediate aims

1. Develop a simple algorithm for trading cable using 15min bars - tested against the historical data I've pulled from MT4 (about 14 days of 15min bars)
   
2. Test that algorithm using data from the same time period for around 20 other currency pairs.
   
3. If it's performance is acceptable against other currency pairs find (much) more historical data to test against.
   
4. If it's still successful determine several situations that will indicate that the algorithm no longer works (indications other than blowing my account!)
   
5. Trade a demo account.

 

[rathcoole_exile]

you asked for thoughts & comments ?

two such spring immediately to mind -
first, 14 days is nowhere near long enough for "historical" data; and
secondly, put aside some funds you can afford to contribute to your market education, and trade a small live account. A demo will teach you fcuk all squared.

best of luck, r_e

oh, and a third suggestion, drop the stupid grinning faces there are more than enough of those to go around posted by that half-brained idiot elsewhere

 

[Vrothdar]

first, 14 days is nowhere near long enough for "historical" data;

Oh aye, I realise that and that's where "If it's performance is acceptable against other currency pairs find (much) more historical data to test against." comes in. I'd like to be testing against that amount of data from the start however the only source I've been able to find so far that looks like it'll give me the data I want in the format I want (opentick) isn't accepting new users at the moment.

I should probably look into datafeeds a bit more, my knowledge of them is pretty weak, with a view to subscribing to one before I start developing my system, not after it. After I've finished typing this reply I'm going to make a post about the methods I'll be using to analyse historical data and develop a trading system. Perhaps you (or anybody else) might be able to recommend a data feed that would suit my needs.

secondly, put aside some funds you can afford to contribute to your market education, and trade a small live account. A demo will teach you fcuk all squared.

The reason I'm going to trade a demo account for a while is to ensure I'm comfortable with the way whatever interface I'm using works. I can live with losing money because a particular trade didn't work out or my decision of stop loss placement wasn't quite right but losing money because I screwed up and accidentally placed a limit order instead of a stop order is completely unacceptable. Once I'm comfortable with the mechanics of trading on a demo account I'll move to a small live account to learn what it's like to trade for real.

best of luck, r_e

Thanks.

oh, and a third suggestion, drop the stupid grinning faces there are more than enough of those to go around posted by that half-brained idiot elsewhere

I agree with you on this one really. I'm putting my use of smilies down to misspent time playing MMORPGs and old habits dieing hard.

 

[Vrothdar]

Keep it simple stupid. I want to keep that in the forefront of my mind whenever I'm thinking about trading. If it requires judgment it's too complicated. I know that I'm vulnerable to paralysis of analysis so my goal is to have a system that will generate trades for me (even if I place them manually) so I'll always be asking the question; could I automate this?

With KISS in mind I'm looking to use 2 or 3 indicators to determine entries, a further couple to determine stoploss and target and that's it. I'll be looking for my trades to complete within 1 - 3 bars of my entry regardless of the time frame I'm looking at. Initially it'll be the 15min time frame although I'm sure I'll look at longer ones too.

I'm doing all my analysis in a spreadsheet with my data looking like the price data attachment. The colours are simply there to make the data a little easier to read. The colour of the close price is the colour a candlestick would be, the colours on high/low are there to show whether or not there is an upper/lower shadow and the grey on then open is just there cause I felt like it.

To the right I've got calculations for my various indicators and cells showing whether or not a setup has occured and where one has the target and stoploss for the trade. I then have a fourth column showing whether or not the trade was successful or not during the next bar. If neither the stoploss or the target were hit it currently returns unsure.

This gives me a summary as can be seen in the summary attachment. Later on I'll write a macro for determining the outcome of all trades so that the only trades that return an unsure outcome is only returned when both the stoploss and target are hit in the same bar.

Using my summary data I'll use the random equity curve simulator (Random Equity Curve Simulator of a trading system. Learn it before you trade), posted by shadowninja in another thread, as an initial evaluation of any set of indicators before doing a more detailed evaluation of the trades.

My next step is to develop my spreadsheet further so that I can paste in data from any currency pair and time frame, adjust my indicators to whichever values I'd like to test and then click a button to generate summary data.

After that I'll be taking r_e's comments on board and looking for much more extensive historical data to test with.

 

[Vrothdar]

Well I finished putting together a spreadsheet to automatically generate entries, stoploss, target and outcome for any historical data and had a bit of a play around with some different indicators and settings for those indicators. I mainly messed about with using 2 -3 SMAs to find an entry and then setting target and stoploss as a factor of the X bar ATR.

I took r_e's advice and was mainly using daily and hourly bars as the historical data from MT4 goes back further and as such is more suitable for testing.

I got some interesting results using my entry condition as one SMA crossing a second in the direction that the third, longer period, SMA was moving and then setting stoploss and targets as a factor of the ATR calculated over the same period as the fastest SMA.

I'll post the results the system achieved for GBPEUR, EURUSD and USDCAD when I get home from work.

Next I'm going to try having two ATRs at different speeds with another condition of entry being that the faster ATR is above the slower one and using the faster ATR to set stoploss and target. Hopefully this will weed out some of the signals that are being generated when the market is consolidating.

 

[Vrothdar]

Turns out I've not got results for EURUSD or USDCAD saved, must've deleted them. Still, the GBPUSD results on 60min bars were:

Each win is 2.5 times the size of each loss. Over 2159 bars 163 signals were generated of which 60 were successful and 100 unsuccessful. The results of 3 signals was unknown and I've counted these as being losses.

36.81% of trades were successful.

Whilst putting those stats into the equity curve generator I posted previously shows interesting and encouraging results it also highlights one of the issues I'm going to face if I choose to use a system with a low hit rate but good win:loss ratio - waiting for a statistical indication of the system not working will result in a relatively large drawdown.

A system with a win:loss ratio of 1:1 and hit rate of 0.75 will have a much smaller standard deviation and because of this it'll become statistically obvious that the system no longer works before large losses are made.

Based on this I believe that I'll have a better chance of success by looking at much smaller time frames. Back to the drawing board...

 

[grantx]

Vrothdar,

Statistically, even an 80% win-rate needs be treated with caution (as LTCM discovered). Out of a hundred trades, the statistic would still be valid if the first twenty were all losers. But then it may be too late. I suppose the solution is very tight stops to preserve capital for the eventual winners and/or improve the system.

Grant.

Grant.

 

[Vrothdar]

It'd only be valid if you were testing against a binomial model. Binomials are a good way of testing the results of a week, month, year etc. as it's easy to pick up a change in your hitrate (everything else being equal) and determining that your model is no longer accurate.

You'd also need to use a geometric model to test your losing streaks. If you tested your above example on a geometric model you'd find that the chances of having a losing streak of 20 trades before the next success is 0.0000000000008%. You'd be statistically confident long before this that your model was no longer accurate.

:edit: I've added an attachment showing the likelihood of different losing streaks for p = 0.8. I stopped at 10 decimal places as it was getting hard to read if I added enough to see the first significant figure for the longer losing streaks.

 

[rathcoole_exile]

bioric what"s ??
grant can you translate into Manc so I can understand ?

 

[Vrothdar]

I've done some more work recently and have worked out a system that a 50-55% hit rate with r:r 1:1 on a daily timeframe. The entry points are pretty solid however my exit strategy leaves a lot to be desired. I'm going to play about with sliding stops to see how that affects the results.

After looking at how that fares I'm going to try using support and resistance levels as another possible exit strategy.

 

[Vrothdar]

I'm putting my work on testing different strategies on hold for the time being. I've realised that even when using more extensive historical data I am effectively looking for a needle in a haystack by throwing together various criteria on which to base trades in a fairly arbitrary manner and looking at the results.

I feel as though I can spot a bad system and recognise one that has potential however I need to learn a lot more so that my approach is a bit more focussed.

Trader_dante spoke about hunting with a rifle and not a shotgun in his "Making money trading" thread. That seems particularly relevant to me at the moment.

Time to go find a rifle...

 

[Vrothdar]

Well I've found a rifle now so it's time to resurrect my journal.

What I've done since the last post:

1) Downloaded data from disktrading to give me a huge amount of historical to test against.
2) Downloaded data from gaincapital to give me a second set of data to run my tests against and confirm the results. This data will require some work to clean up which I wont be doing until I've achieved results using the disktrading data that I'd like to get confirmation of.
3) Determined which statistics and visuals to use to measure system performance.
4) Created a template spreadsheet to paste trades into which will generate aforementioned statistics and visuals.
5) Tried out several strategies using MA envelopes loosely based on bollinger bands with no significantly interesting results.
6) Tried out several stratagies using breakouts with an aroon filter with interesting results. Generally they highlighted how the markets change behaviour over time. I couldn't find a set of parameters which worked consistently enough.

An idea borrowed from rnicoll was inspiration for my next attempt so if I make millions from it I'll give him a fiver or something

It's going to be (as far as possible) self optimising and will take far more indicators than I've previously used into consideration in a slightly more intelligent way than "when x crosses y do z".

I'll post back when I've done some practical work on it.

 

[Vrothdar]

First attempt at my new system

I've just finished a quick test with my new system idea using 4 different EMA crossovers giving each one a different consideration depending on it's performance over the last 50 bars.

The light blue line is the cumulative total pips won if all 4 EMAs were traded independently and the dark blue line is the cumulative total of pips won using my proposed system.

Needs some work I think

 

[Vrothdar]

Turning it on it's head look very fun! Will have to apply some money management to it to see how it performs.

:edit: Forgot to mention this earlier but that's the graph for GBP/USD. I'll be testing it against other pairs this evening and going over the formulae in the (20mb!) spreadsheet with a fine tooth comb.

Hopefully I'll also have time to try mucking up the parameters for the EMAs and introducing other indicators to give it a real hard robustness test.

 

[Vrothdar]

Looks like I'm going to need to do some serious work optimising my spreadsheet before I can do much of the above. It's slow as hell to work with.

 

[north5]

Maybe you should look at taking the next step into something like metatrader? Happy to lend a hand when I can if you do.

 

[Vrothdar]

I managed to optimise my spreadsheet enough in the end. Don't think it'd be much use (on this PC at least) for any time frames shorter than daily. At least not going as far back as my daily data does.

Unfortunately I noticed an error in the way weights were calculated for the voting of each EMA. When the weights were calculated for day X is was taking into consideration the results of day X (ie my system was using the future to decide what to do in the present). This shouldn't have had too big an effect on the results as the weights of the votes are calculated based on the performance over the last 50 days. As you'll be able to see in the following posts that wasn't the case.

In all the following screens I've unflipped the voting strategy as I realised that it shouldn't need to be flipped. Based on the fact that sometimes it performs better when flipped and the huge difference the error had I think I need to take a closer look at the way the voting works - There's probably a mistake in there somewhere or it's working in a way I hadn't anticipated.

Either way, I'm pretty pleased with the results when the error is in place. It'll be a big step forward if I can work out why that one day makes such a big difference to the performance.

I might get a little work done tomorrow, otherwise it'll be slow going for a little while as I'm going away this weekend for a couple of days with my family and next weekend is reading festival.

 

[rnicoll]

Unfortunately I noticed an error in the way weights were calculated for the voting of each EMA. When the weights were calculated for day X is was taking into consideration the results of day X (ie my system was using the future to decide what to do in the present). This shouldn't have had too big an effect on the results as the weights of the votes are calculated based on the performance over the last 50 days. As you'll be able to see in the following posts that wasn't the case.

Thing is, EMA's are weighted on time, so the current day is the most significant, then the previous day, then the day after that. By 50 days ago the influence is actually tiny. That's probably what's happened.

Pity, though, I really liked the voting

 

[Vrothdar]

 

[north5]

Interestingly I am also taking a tangent looking into a few voting ideas with indictors. I think that if you can understand the indicators enough then find some way to factor the weight they should have in the current market decision it could really lead somewhere.