Interesting Stanford paper: http://cs229.stanford.edu/proj2011/BrownMundkowskyShiu -PredictingIntradayPriceMovementsInTheForeignExchangeMarket.pdf Basically, the authors backtested a forex strategy using only the bid and ask sizes at 1 second intervals and achieved a very high predictive ability. They claim a profit of $2800 per day over the 45 period of the test after transaction costs and slippage. I'm wondering, however, where a trader would get forex market depth, since it's a decentralized market. Do any brokers offer this by API? I don't know about forex, but this probably won't work on stocks since "level 2" is no longer accurate given HFT, dark pools and all the tricks MMs use to hide orders.
80% Accuracy in Predicting Intraday Price Moves Remain focused on..., managing the other 20% - that is what will ultimately make / break a trader Even with a 99.9% hit rate - the other .1% is the potential killer Once one learns to manage the killer (whatever % it actually is) - a "high" win rate simply no longer matters.., and the chasing BS ceases RN
I could not have said better, once you learn that risk doesn't have to be scary. I use to think long term trading carried all the big risk and day trading offered smallest risk, when it is really other way around for me. Day trading has all the huge risk and compared to long term has very little when you hedge. In any given day I am risking in day trading what I would risk in 16 commodity markets for long term. Somebody come with fat finger in ES...there she blows.
There is no such thing in spot fx, which is probably why the paper is vague about the source of the data. Bank fx quotes are client-specific as to both width and depth. Also 'the data spans from 1 February 2008 to 13 March 2008'...
I remember back in the early 90's if I had positions in currency futures, desk would put protective stops in the "Exchange for physicals" after day session closed and recall it was before forex as we see it today. They had forex back then, but mainly banks and large entities used it. And if I was filled at night, it was another way for brokerage to tack on another fee. I often thought I was in the dark.
This was an assignment for an undergrad machine learning course, not a published academic paper. No peer review and as pointed out who knows what dataset they used. Given the quality of what I churned out for similar courses back then, I'm not sure I'd put much faith in it.
Data mining result. This article shows what happens to 1-min forex traders that do not have substantial edge (no one does, only MMs)