I have a system that I have been testing live for almost a year now. It involves setting one or two orders per per day, and is overall profitable system. The risk:reward ratio for each trade is 1:2.5
Most trades ocur before 15h, but I have noticed that for those few that do ocur after this time, the success rate is only 5%. I thought to reverse the orders at 15h if they have not been filled, and then target the what would have been the stop loss.
My question is this. Out of 150 trades, 20 took place after 15h. Is this a large enough population on which I can make a decision on this, as it is spread over almost a year. I am worried because for these trades, my risk to reward ratio would be 2.5:1.
To me it sounds like what you’re attempting to do falls under the realm of ‘curve fitting’ - something which no doubt we have all experienced when trying to maximise the performance of a trading system; it takes a rational approach to avoid this, ensuring you back-up any system adjustments with a [B]valid quantifiable and objectified reason[/B], rather than making adjustments just for the reason of ‘[I]maximising profits[/I]’ by fitting the market data around a certain trading approach.
First of all I would take a ‘top down approach’ into your findings. There may be a valid reason why market orders are less profitable after not being filled for 15h from the time of placing them. One potential reason could be that the market you trade after 15h is not within it’s ‘[I]prime liquidity[/I]’. Lets take GBP/USD for example, any trades placed within the time of European open to US close are more profitable than orders placed outside of these times, such as during the Asian Session. The reason for this, as explained above, is because liquidity is at its highest when the European Markets open and when the US Markets overlap; liquidity is at its lowest during the Asian Session for GBP/USD. The currency pairs that you trade though combined with their respective liquidity sessions will vary to the GBP/USD example.
The above example would be explaining your findings from an objectified approach.
To explain your findings from a quantifiable approach you would, as you suggested, need a larger sample size over a number of years. 20 trades over a year falls foul of both [1] the number of trades in a sample size and [2] the time duration of 1 year.
[1] - Number of trades in a given sample size should really be exceeding 100, preferably multiples of this. The fewer trades you have the smaller the degree of confidence in your findings. Logical really, we’d all want more trades to base system adjustments upon, right?
[2] - The time duration of 1 year is, [B]really[/B], way to small. Markets, not just Forex, have cycles which include boom, busts and ranges - quite often you will see each of these lasting over a year. It’s no good building a trading system over a years worth of data if the market was always in either a boom, bust or ranging period; because when the cycle changes you will not be able to adapt. As an example, look at August 2008 (sticking with GBP/USD), the ‘bust’ part of the market cycle lasted for over 12 months - how would a system built around this data have worked for the past two years where we have been ranging with no significant direction (some may still say we have not recovered since 2008 and are constantly in a bust market cycle)
Thanks Jezzode, I think I really needed to hear that. I had read about overfitting, but I suppose the tempatation was quite strong to try to try and justify it.
With regards to you comments on sampling. I have backtested back to 2003, using a tester that I built myself in VBA and Python. I have also been using this tester in time with my live testing to make sure that it is analysing the data properly and yielding realistic results. I have a 99% correlation between the two, so I am happy that my sample (in the thousands) is as large as I can make it without paying for more data.
As for your comments on prime liquidity, you nailed it, as I am working with GBPUSD. Its interesting how I could ignore facts like this when it came to trying to maximise profits. I suppose the lesson here is that psychology is just as, if not more important when analysing trades as it is when actually placing them.
Thank you once again, I really appreciate the insight.
The data for one year can be quite significant to base your changes in trading system. Of course you can try to place reverse trades on those “unlucky” 5% but I would completely exclude them from your trading system and place only those that have high success ratio (probably trying to increase lot size on them,i.e. taking more risk)