Pyramid, good or not? Despite the pyramid trade management (aka add-in) sounds appealing I always had some doubt about it. So far, I couldn't find the trick to prove it right or wrong. Actually it turns out to be quite easy when you realize that to ...

Expectancy Management

I have very different prices than yours. My source is FXCM for this BT. When I'm back at home I try again with Dukascopy data. I think this is the reason. When I feed my linear regression algo with your data (post #99) I get exactly the same result ...

Machine Learning with algoTraderJo

Hi ATjo I'm still trying to reproduce the linear regression results. My parameters are: Currency of the account: USD Starting balance: $100000 Money Management: 1% Timezone: Europe/Paris,Berlin Price used: open price, (bid+ask)/2 Stop loss: 0.6 x ...

Machine Learning with algoTraderJo

The 1 microlot with 100€ was only to have the graph scaled properly ;-) (this is on my "to be fixed" list) It is true that a poor MM can deteriorate and even kill a strategy. But it is more importantly true that no MM in the world can turn an ...

Machine Learning with algoTraderJo

"GMT +1 non-DST, GMT +2 DST" I'm always confused with this GMT offsets... To me this is Paris/Berlin timezone. I corrected my strategy to use local time taking DST into account. MM = fixed lot size 1 microlot. Starting with 100€. When I use London ...

Machine Learning with algoTraderJo

What is the timezone used in the file detailed_trade_log.txt? I can't match with the prices I have. The SL can get hit at any time. Why are all the trades in the file stopped at an o'clock hour? This is what I get. I assumed the file contains GMT ...

Machine Learning with algoTraderJo

We use a daily sampling therefore, as long as we sample at the exact same time (forgetting the DST days), we still have a 24 hour period. Each period of the trading session is represented. I mean if I sample every 1:37am GMT, the asian session, the ...

Machine Learning with algoTraderJo

I didn't at all though about this "details". I knew I would someday have to bother coding this GMT -> any TZ+DST.... :-( As you're from the USA I suppose you refer to the US DST, right? Europe and US don't shift on the same week. I'll try again take ...

Machine Learning with algoTraderJo

Has somebody been able to reproduce ATJo result on E/U daily? My result is always a loser strategy. Of course I checked that if I feed the data in the excel file (post #99) I get the same predictions. I noticed that the week-end bars are absent from ...

Machine Learning with algoTraderJo

Yes except for the top right one. Here the problem is the model: the line tries to capture a linear relationship between X2 and Y2 where the relation is actually quadratic. Another example would be when the samples are packed in a small region of ...

Machine Learning with algoTraderJo

algoTraderJo posted a method using linear regression. It is perhaps interesting that I post some facts about this method people should be aware of. They aren't obvious when you just use off the shelf black boxes software. To keep the explanations ...

Machine Learning with algoTraderJo

At least it would be built-in multi TF :-)

Machine Learning with algoTraderJo

Following the same idea. What would happen if we used various lags like 1, 2, 4, 8, 16, 32 bars? My idea is that it may regress a rough estimate of the Hurst exponent. When the market is in persistent mode the probability that the next bar is the ...

Machine Learning with algoTraderJo

I mean that 0.5 and 20 should perhaps be themselves part of the solution of the regression. At least the 0.5. For example, let's say the target candle is UP 65% of the ATR (close-open). This doesn't tell about the intraday DD. If the lower wick is ...

Machine Learning with algoTraderJo

This was also a question I had for you: if it is possible to measure the quality of an indicator independentely of the trading system that uses it. The answer is clearly NO from what I read. Therefore shouldn't the model contains the trading rules ...

Machine Learning with algoTraderJo

Thanks. There is a plethora of libs for machine learning in various languages: url I personally use java. Weka seems good and features a standalone GUI for quick and dirty testings. There is no need for a full blown ML library anyway for a trivial ...

Machine Learning with algoTraderJo

U/CHF and E/U are strongly (anti-)correlated. It is no wonder that their results are similar. Why do you conclude there is a fundamental difference between E/U and G/U instead of questioning the validity of the strategy? Isn't it simply by random ...

Machine Learning with algoTraderJo

Hi algoTraderJo I'd like to reproduce your experiment post#10. I don't understand the approach. From what I understand you take the last 200 triplets of days. The first 2 define the input and the 3rd is the target. I took E/U dataset I get these ...

Machine Learning with algoTraderJo

No. You don't want to maximize your profits. If it were the case you would bet 100% for each bet because this is the bet which maximize your profit and a string of winner maximizes the sum of the profits. Of course the first loser loses it all. What ...

Money management - What is the secret?

AlgoTrader how do you deal with multimodal predictions? image

Machine Learning with algoTraderJo