Another lesson learned over the years is the effect bad data has on your testing results. This can be somewhat mitigated by statements made in my previous post regarding mixing up your watchlist. However, when reviewing your results you may notice some wildly profitable positions. A clear signature of this is a windfall profit after a hold time of only one bar. This is further established when you have a profit target and the position closes far above that target. Typically in a trading system with profit targets based on daily data, your PT order will be submitted the next bar (day) after the system gets confirmation that the order was filled. Of course, you can get a finer granularity of data and make this happen much quicker, but let's keep it simple because it can apply across the board. Typically what happens is a rogue tick or just a plain wrong low/high print will trigger the backtesting engine to open a trade. This will cause your position, when analyzed at the close, to show a massive, but erroneous profit.
It's a bit unrealistic to comb through all of the data and make sure that it's perfect, but you'll want to analyze your backtesting results and identify these trades. If possible, remove them from your results by editing the data for that particular bar (assuming you don't know the proper high or low print). What you do need to be aware of is bogus prints on current data. The signature of this can be an unusual entry price. What I mean by this is imagine a simple band violation system where all of a sudden the bands some to widen substantially. This may be the result of some false increased volatility. This will potentially throw off your algorithms and backtesting results will not match reality.
Finally, bogus trades with wildly incorrect profits seem to be more apparent in limit order based systems. Again, typically due to the misprinted high or low. A market order based system can cure some of this, but based on my previous statements, you're not 100% in the clear with market orders either.
Subscribe