Well. trading is a bit of an art, not only science, so it’s hard to come up with fast and hard rules. But in my humble opinion, this whole in-sample / out-of-sample idea is bollocks. If out of sample doesn’t work, what are you going to do? Pretend you can co back in time try to find something else? And then repeat the process until out of sample is ok? Why not just require a system to be profitable in both halves?
I very much like your thought process. I have just written a chapter for a new edition of Professional Investor Rules which has given me much pause for thought.
In particular my backtests of simple systems on UK and US stock and bond markets going back as far as 1709 make me realise that almost all fund management and trading as currently practised is bullshit at heart. Skill or survivorship bias as regards those heroes of the financial press? I strongly suspect the latter.
Frankly for all the effort and agony of applying the momentum system I document (with mere quarterly re-allocations) it’s so much simpler to apply a 60/40 split equity / short term bonds, re-balance once a year and forget about it.
My strong suspicion is that in the long term this would probably leave the Masters of the Universe at the starting gate. And no real bother or agonising about curve fitting or the efficacy of momentum.
Trading is only worthwhile if you can reap BIG profits. Pissing around for a couple of points over buy and hold for a reduced DD and Vol is so tedious I can no longer be bothered.
So I’m currently trading the VIX which shows large profits and a positive MAR ratio in backtesting to 2004 using futures. In back testing! Ha!
Now this project Genotick does seriously interest me. So does the idea in general of applying machine learning to investment. I have only just started looking through the code and its going to be a struggle but in principal if any mode of investment is going to beat B&H in the future it’s going to be machine generated. In the long term I would bet large money on that.
It may be we need infinitely more data than is available at present and quantum computers. It may be that complex systems are entirely unpredictable and that we live in a random universe.
But hats off to Lukasz – a valiant effort and I look forward to learning more over the coming weeks.