sleytus wrote:Regarding pips -- I understand and will consider it. But as with everything in life, there are always trade-offs. I'm not an experienced trader so my perspective may be off, but the problem with using pips is I would no longer be able to compare one strategy with another. The purpose of the app is to help compare strategies to one another and exclude the poor performers. Pips have different values depending on currency pair (e.g. EURUSD vs USDJPY, lot size). So, for example, if you were comparing two strategies -- one that won 10 pips and another that won 50 pips -- you couldn't say which was the better performer. But, if I had one strategy that won $10 and a second that won $50, it's pretty clear you'd chose the latter. Also, I like using the Return/DD metric -- I find that better than the the Win Ratio -- and I need to use $ to calculate that.
A clear improvement for comparison is "R-Expectancy". With this metric, you can compare ANY two strategies, across any pairs, and with any lot sizing. R-Expectancy is basically your profit for every dollar you risk. So if you risk a total of $100 and your profit is $10, your R-Expectancy is 0.10. It is also 0.10 if you risk $2,500 and your profit is $250. It seems simple (and it is), but in this way, you can directly compare different trades of different lot sizes, regardless of pair.
****
Read the following at your own risk:
Getting a bit more complex, I would highly suggest implementing System Quality Number. To do this with MT4Tracker, it would be extremely helpful to also add a "margin of error" function using Bayesian probability. The reason for this is that SQN doesn't become 100% dependable until a system has made 40 trades. (Actually, there are some issues with it, but it would be the best reference point available.) Unfortunately, I would be of little use in the specifics of structuring the function. Perhaps somebody else is handy with advanced statistics.
How would a Bayesian function help? This is best illustrated by an example. For this example, I'm using profit instead of SQN; it is easier to conceptualize.
Let's say that your EA has made only five trades so far. The largest loss has been -$200, the largest gain +$150, and the average return -$30. Without more information, you may already be starting to write this strategy off. The Bayesian probability might tell something like the following: In the long run, you can expect with 68% certainy for your average return to be between -$150 and +$90. And with 95% certainy your average return will be between -$190 and +$130. (These numbers are purely made up, but the general idea is there!).
Now, let's say that your EA made more trades...20 total trades now. The largest loss has been -$280 and the largest win $300. Your average return has increased to -$10. Now, you may see the following: 68% certainty that long-term average return is between -$25 and +$5; and 95% certainty that it will be between -$35 and +$15. You can probably think about throwing this one out now.
Perhaps that wasn't the best illustration, but this measure can make you understand when a system that looks quite profitable (based on a small number of trades) actually still contains a large amount of possible Variance.
Doing the same for SQN would be a bit more complicated, but even more useful, as it's hard to understand how SQN might fluctuate with small samples.
That certainly wasn't my clearest post (but then none of them seem to be!), so if you're interested in discussion, please let me know.
ThanX!
John
