Consider a system with a 40% win rate and a 2:1 average win:loss ratio. The higher you try to go, eg 50%, 60%, 70% win rate, normally by adding filters and/or picking markets where your signals worked well in the past, the more you risk falling into the curve fitting trap. Even when it seems obvious the market will go higher it often still goes down takes out your stop before blasting higher like your signal suggested. That is why maintaining a win rate above 50% is hard. Now if you want to be profitable 10 out 12 months. You need find around 40 signals per month (assuming a 40% and 2:1 system). Finding that many signals per month, with even a 40% win rate, is not easy either.
Good Morning themickey, I'm telling that to the OP not me. I could not find sucess with algo trading and loss $25,000 last year doing it. The OP may find a profitable algo, all he has to do is keep trying
My brain tells me in order to win in trading, one must be focused, systematic, stick to a plan. Also the brain must be capable of absorbing and remembering detail. Imo, its nigh on impossible doing this via discretionary trading.
Hello themickey, Of course, whatever it takes to make money. A plan may work, discretionary may work, or systematic trading may work. I prefer discretionary trading, because I am unable to get systematic trading to make money for me. Discretionary works good enough. I don't need perfection, I need money. It's about money to me, I don't need all the extra stuff in this trading business Systematic trading may works for other traders, but I does not work for me.
I do apologize for the useful information. https://www.algomatictrading.com/post/8-tips-to-avoid-curve-fitting Not the norm around'ere
Hi Drawdown Addict, No this is absolutely fine I don't have a definite answer and the purpose of this thread is to invite discussion! I read your link several times in detail it's really helpful and I agree with most parts. Below is a trade I just came up with and now testing it live with paper money. What's your thoughts on it? Do you consider it curve-fitting? Entry criteria: 1. time is between 940 am to 1035am 2. Exponential MA 20 is below Exponential MA 60 3. Close prices for the last 10 bars are all below VWAP 4. The previous candle exhibits the following: High>VWAP, Close<VWAP, Close<Open Enter a long position if all the above conditions are met.
These are really useful guides! Thanks so much. I once made a 70% win rate system with a 1.8:1 win-loss ratio on a 1-minute chart with avg holding period of 50 minutes. Failed miserably when the market microstructure does change.
The concept of curve fitting applies to back testing. It is essentially a way to cheat in your backtest to get the best results, knowing in advance what the result would be. In you example you are just setting up an entry based on some criteria that you like. It is a perfectly valid way to enter a trade. As good as any other. If you have chosen that criteria based on historical results you might be fitting it to the curve. Essentially the well know sentence : "historical records are not indicative of future results" applies here. If you base your trades on what happened in the past you will be bound to failure.
Thank you!! But isn't this the essence of backtesting? - to test whether your hypothesis would yield positive results. This hypothesis could come from observing what happened in the past. In this particular case, I saw quite a few instances of a nice reversal developing after the above conditions were met. So I coded it and tested it over a longer time frame to see if it still holds.