The fundamental tenet of all system trading is that the strategy should work across a broad range of time ranges and a wide swath of products. For example, a “robust” algo in FX should be able to trade all the eight majors currency pairs and make money for 10 years back.
Otherwise, you are just cherry-picking and curve-fitting your data and all the serious data scientists will go tsk tsk in disapproval.
Yes, I may be committing my greatest trading apostasy to date, but I am here to tell you that the ONLY way to make money from algo trading is to cherry-pick away.
First, let’s agree that all trading systems fail 100% of the time. It’s just a matter of time before they start to bleed money. Indeed very often the best-tested systems fail the worst, sometimes at an alarmingly rapid rate when they are put into production. If the laws of data science really applied that would not be the case.
The laws of data science, of course, do NOT apply at all which is why the whole philosophical foundation for determining what is or is not a “valid” trading system is incorrect.
The statistical method implicitly assumes that it is observing the truth. And when it comes to the physical world that assumption is generally correct. The laws of gravity do not change and the flip of a coin over a very large sample size will always end up to be a fifty-fifty bet. But the psychological world is not at all like the physical world. One of our most distinguishing characteristics as human beings is that we lie.
Statisticians in social sciences found out just how much we lie the hard way in the 2016 election. But elections are child’s play compared to financial markets. Financial markets are the absolute apotheosis of human lying. Whether on day trading time frame or investment time frame the function of the market is to sucker as many people as possible into making a false bet.
That’s why data scientists constantly talk about “noisy” data in the financial markets -- which is just a polite way of saying that everybody lies and you can’t draw any conclusion from past price action no matter how far back it goes in time.
So what’s the answer for the retail trader who wants to use algos? Stop looking for Mr. Right and go with Mr. Right Now. The single best way to have confidence that the system will work is to see if it’s been working in the past six months. The success of any trading system, in my opinion, is really a function of it being in sync with the current market regime and whatever unique exploitable patterns you may have found in an individual instrument. So yes, it is very possible for a system to make money in CADCHF and in no other pair and keep doing it for much longer than you think.
That’s why the only practical way to make money algo trading to cherry-pick away. Design the system, test in many pairs and only trade the absolute best most recent results. And then do it again with another system and another system and another system, because the key to making money from algo trading is to run a portfolio of systems that are working and then remove those that start to fail.
Next week, I’ll discuss exactly what I think “failure” means in this context, but in the meantime you still need to backest everything you try and as we know that can be unbelievably tedious, so my friend Daniel Sinnig created an Auto backtesting software that can let you run hundreds of tests while you sleep.
Here is his info here.
Save time and money with the MT4 Backtest bundle. 50% for a limited time at https://mt4.tradingexperience.com