Thanks for taking the time to perform this comparison. Here is the result I get for E/U with data from dukascopy. (shall I say the former data were from fxcm?) image

Machine Learning with algoTraderJo

All dates are GMT, I let your framework adjust.

Machine Learning with algoTraderJo

url

Machine Learning with algoTraderJo

They are very similar but of course different since the data isn't the same. The sign of MLE-MSE is the same for only 78.3% of the records. What is the format of the 1 min data you use? I may generate a file so you can try with my dataset.

Machine Learning with algoTraderJo

The results highly depend on the start time. If I simulate from 2002 I get a DD of 75% that the system never recovers (final=-32%). If I start the simulation in July 2007 the final result is +178%. July 2007 is the lowest of the DD for the system ...

Machine Learning with algoTraderJo

I can't reproduce with E/U; neither with the parameters post #408 nor with thoses post #420. But the overall shape of the two equity curves is the same, except mine has no up trend! The peaks and valleys seem to match in time. Jo's DD are always ...

Machine Learning with algoTraderJo

OK for the recurrent events (NFP, FOMC...) but how to code the punctual events like bankrupty of a big bank, earthquake, explosion in a Japanese nuclear power plant, war in whatever country, embargo because of a previous event...? And how do you ...

Machine Learning with algoTraderJo

How do you quantify the fundamentals over the historical data?

Machine Learning with algoTraderJo

This is what I get with EUR/JPY and this new set of parameters. (didn't try E/U yet)

Machine Learning with algoTraderJo

Can you reproduce the results posted by Jo with the Asirikuy framework?

Machine Learning with algoTraderJo

I don't want to be mean but the GA you used to get A, B, C and D before backtesting over the whole set is already data snooping.

Machine Learning with algoTraderJo

Certainly just a wording misunderstanding. When you say "the current bar" I think you mean the candle which is currently forming and not yet closed. It is now 17:54 my time so you mean the bar that started at 17:00 and will close in 6 minutes. To me ...

Machine Learning with algoTraderJo

Not so much of a difference actually. FX doesn't have that many gaps. The bars usually open very near of the close of the previous one. I use the close price simply because I cannot use the open of the bar the strategy receives. Per definition the ...

Machine Learning with algoTraderJo

I simulated using JPY as the account currency to prevent the rate fluctuations to distort the result. I completely rewrote the strategy to make sure it is correct. I still get negative results. MLE and MSE ARE capped to zero as seen in post 356: the ...

Machine Learning with algoTraderJo

Is anybody able to reproduce Jo's curve? None of mine remotely look like his. Using A=1, B=80, C=6 and D=12 on USD/JPY from 2007 to today, I get breakeven with a 15% max DD. 2.5 years to recover. With E/J A=0, B=130, C=9, D=12, I got a profit factor ...

Machine Learning with algoTraderJo

Ok so you cap the values to 0 in the training target. But nothing prevents the forecast to be negative, and it often is. Do you cap the forecast as well?

Machine Learning with algoTraderJo

It happens twice with my dataset from 2007 to 2015. 29/10/2008 MLE=-9.003751139075197E-4 MSE=3.8783981800940843E-4 09/05/2010 MLE=0.002561915803222095 MSE=-0.004921393412012904 Note: It is possible that the MLE is negative (when the D future prices ...

Machine Learning with algoTraderJo

I was not really able to reproduce the results. I could obtain a positive return but nothing as great as the equity curve posted. Jo, what are you using for the spread? the real ask-bid or some fixed value? If you use 3 pips like in the previous ...

Machine Learning with algoTraderJo

No GA is needed. This has got a closed form solution. In matrix notation: image Where q is the query vector (the last 9 returns), S is the samples matrix (130x10), T is the target matrix(130x2), ŷ is the forecast. EDIT: q and S must have their ...

Machine Learning with algoTraderJo

Happy to see Joe is back. You don't regress the samples to get a line. Each sample is made of C=9 returns. It is a 9 dimension vector. You have B=130 such vectors. The target is made of two values (mle and mse). It is a 2 dimension vector. You also ...

Machine Learning with algoTraderJo