If you’re trading the EUR/USD on the 15m scale and are using moving averages you might consider plotting 275, 625 and 925 period moving averages. Out of a thousand MA’s in our data-set these three had the most forecasting value over a 24 hour period (and beyond). They’re not going to give you perfect signals but might be useful in your broader analysis.
I will update this thread with more time series and changes as they occur. - JP
They don’t always work this well, but the optimized MA’s provided pretty accurate S&R today. - JP
Hi JP,
Is that sort of today’s lucky numbers?
Cheers
Everything is lucky with hindsight 
Those levels should be good ones to follow for a while, they were part of a pretty large regression and stood way out over the other levels for impact. - JP
Why those, many people like 20,50, 100,200 period MA’s why are they better than these mentioned?
I think a better question here is, why do all those people use those MA’s? What is [B]their [/B]basis for it (I’ve already explained mine but will try to clarify better)?
Part of the trading process for me has always been why? Who do individual traders do this, why do professional traders do that and so on. What is the basis for it and does it actually have meaning?
One of the long-standing components of my trading has been indicator optimization, which also satisfies why one MA might be used over another. The latest batch of MA optimization used 1000 moving averages over the most recent 100,000 15 minute time periods. The system evaluated 30,000,000 different models over the course of nearly 3000 processing hours.
Out of the top 100 models those three MA’s had the most impact on the models ability to forecast the direction of the EUR/USD in the 15 minute time series over those 60,000 periods of time. It may not be perfect basis but it might be as meaningful (hopefully more so) than some of the other reasons. - JP
Taking a look at the price action this morning, with the EUR/USD breaking below the 925 MA, we’ve seen some good momentum build. If you are short here, trailing your stop with either the 275 or the 625 MA’s could be useful if you don’t have a different exit strategy. - JP
How did you perform the tests, since you know that all MA’s are good yes?
All of them are accurate, it is more important on what scale are you looking for.For example EMA 4 & EMA 8 cross will signal 10-15 pip trades while SMA 50 & SMA 100 cross will signal 200-300 pip trades, its just the scale of the trade which differs, the accuracy is not since the market is fractalized, it will repeat its direction sooner or later, even if the next candle after the cross will not show that.
It’s good if you find some Ma’s which work for you, as everyone has their own which are unique to their system. Some people like using MA’s for that one time frame but others like to use longer term. I tend to find a 100, 50 on a 30 minute chart and a 21 on H4 and 8 on a daily give a good bias for the trend.
My preference and experience for modeling and the application used for this regression and classification project is Low Level Linear Genetic Programming over a massive distributive grid. Here is a good intro to that LGP. - JP
Ok i understand that you try to search for the optimal MA but still an MA is only an average of the price, the higher the period the bigger the lagg is, except if a clear trend is emerging.If you found the 625 MA and the 275 good for entries, that doesnt mean that the 4 & 8 wont be the same good.
This is your system.It really works fine if you enter and exit at the crosses, but here is an MA 10 & 20 cross
Why would be this one any worse than the 625 /275 cross?
Both signal the same accuracy the entry exit points? It may be that some trades are quick, thats why you go higher timeframe to not get destroyed by the high transaction cost/spread.
Here is a some insight into Indicator Optimization with Genetic Algorithms/Programming. Indicator Optimization
Here is a paper on Distributive GP.
These will help give thread readers some insight and basis into the “how” of the modeling process. It will also likely bore you to tears if you’re not inclinded to this type of reading and research… - JP
Ok i will read it when i will have time, but i still think that the best method to test it is by writing an EA which:
ENTERS BUY when QUICKER MA is above SLOWER MA, EXITS BUY when QUICKER MA is below SLOWER MA and
ENTERS SELL when QUICKER MA is below SLOWER MA, EXITS SELL when QUICKER MA is above SLOWER MA
Then run the EA on atleast 10.000 trades, and then compare profit,drawdown and winrate, most importantly win rate 
I just did this, and the results were mostly inconclusive, but i also found out that higher period MA crosses are more accuratye since at higher periods the market noise is 90% filtered out, but one mistake will give you a huge drawdown, so i believe MA cross methods should be done only on intraday trading.
Not for the winrate but for the safety and proper moneymanagement rules.100 EMA /200 EMA or 50EMA /100 EMA on M15 is very nice ,although it varies on all pairs.Ranging pairs need quicker period crosses while trending ones need slower ones.I dont think you can just generalize it, since if you look at crosses above 600 period i think you should do atleast a backtest of 20-30 years which i dont think you have data for.
For intraday trade on 10.000 trade sample, 5-10 years is enough for statistical accuracy
You seem to have it all figured out. I’m happy for you, most of us never get there. Like I said in my intro post I’m just an old dog and maybe all of you can teach me some new tricks. - JP
Dont get me wrong ,i analyze the market mathematically just like you ,so there is nothing wrong with your work, but there is really no need for complex formulas to analyze it.The key is to keep it simple, the most important part is to have a standard, to compare to.You dont need to calculate all combinations of the parameters, since it will never work out, and i will probably take you many years unless you have a supecomputer with like 1 terrahertz processor 
You can get accurate statistical data more easier than that, and even if its not that accurate, doesnt matter the market will change too.I dont find the MA cross strategy very reliable by itself, if you use quick period cross you will have small lagg but many fake signals,if you use long periods you will have big trends and probably higher winrate but one position will be so big that it will be bigger than your preferred account risk.So the best is to use intraday MA settings with other tools to filter fake signals.
I backtest my methods very easily.And this works for any strategy not just for the MA crosses.Backtest a couple trades with a preset configuration of parameters, i find the local maximum (of winrate,DD, or profit), and then retest the others with different settings compared to the local maximum values.If they are better then that will be the new standard, if not then search for other configurations.
For sample i use atleast 10.000 trades no matter what TF are we in, of course 10.000 trades on H1 will probably need like 10 years of price data.Thats a pretty accurate statistical sample and when i`m done i compare the results and mainly look only at the Winrate and the DD, the profit is irrelevant because its market dependant and will vary always, the key is really the increase if the absolut average winrate (without doing tricks with the TP and SL, like setting the TP to 1 pip or so).
The other method i use is with fixed TP and variable SL, for TP i set 1 pip and this will be the new standard.By this i will maximize the winrate of the strategy, and if any improvement can help it make bigger winrate than, that will pass the test also and that can be said to be an improvement of the strategy.The 1 pip method i described here more accurately when i studied the ATR indicator:
And note this is not a strategy by itself, it would be crazy to trade like this, this is just for research purpose.
Could you elaborate what you mean by forecast direction?
Cheers
Prox, I’m starting to feel relieved that we disagree on the “how”. This thread was a bad idea. Not sure what I was thinking. - JP
What do you mean by that?
I suppose he means what he says, no sarcasm.
Cheers
Kas, I’ve sent you a pm. - JP
