In the world of algo trading, it seems that curve-fitting is our number one enemy!! tackling that could save us a lot of money. Just curious - what's everybody's opinion on curve-fitting? Have you come across anybody that has curve-fit their way to success i.e. consistently profitable? By consistently profitable, I am gonna give it a loose definition, say 10 out of 12 months in a year is in the green.
I suspect you mean over fitting. If that is what you meant, then no, you cannot over fit your way to success. Over fitting pollutes your thinking with delusions. You cannot successfully navigate reality based on delusional thinking in a highly competitive setting where your competitors thinking is based on reality and its truth.
What exactly are you doing? Curve fitting what, and to what data? The order of the regression matters. The assumptions about parameters and estimators need to be carefully considered. What inferences will be made from the analysis of said estimators? Are you trained in the use of statistics/probability theory/parameter estimation?
By no means am I a highly trained statistician. I do know how to interpret statistics. I am asking about curve-fitting because of what I read in this blog post where the author suggests that optimization should be done on the whole data set essentially rejecting the idea using out-of-sample walk-forward testing. https://mechanicalforex.com/2016/01...ll-available-historical-data-makes-sense.html
It is meaningless to use OOS periods, cause you optimize IS as long as OOS is good, so basically it is all then In-Sample. Backtesting at all you should not do, the past is not indicative for the future. There might be by far too many different aspects at the present which are totally different from past. Try a qualitative approach where you need to judge not on numbers but on your common sense, that is much better. I was also an algo guy too for over 10 years, just to realize you cannot calculate the markets for high-sharpe strategies. You might have a luck for some months going forward after testing, but that is usually a much too short period and also it happens very rare that you have something working in the future from now. Just avoid and skip all backtesting is my credo now. Be discretionary in all aspects as much as you can.
The website seems pretty naive. The important things in FX trading are like VWAP, momentum and volatility. The FX fundamentals stuff is really hard to learn. This market constantly trades against itself because dealers mark up the quotes they get from ibanks and futures exchanges (brokers pay market rate for their hedges, even while they fill you outside the market (no commision/dealing desk)) which leads to traders chasing ghost liquidity. VWAP alone will improve your trading in FX by like 50%. If you're on top of momentum and vol, you will do even better. Regressions, sampling, etc., can help you figure things out about the market, but you will learn obvious/well known stuff that way. You need to focus on the stuff that matters. Random sampling, curve fitting, parameter estimation are actually quite specific and surprisingly limited tools. They also require the problem to be simple, relationships to be constant/unchanging, and even that the thing being estimated either cannot change, or will not change. Good luck.