There's no reason why in theory you shouldn't use the same methods you'd use for a backtest (bootstrapping, bayesian...), but of course your return history is much shorter (unless you've been live trading for donkeys years with an unchanged system), and hence you're extremely unlikely to be able to distinguish between different strategies (unless you are a high frequency trader with a high single digit or low double digit Sharpe). This means in practice you're likely to recover your prior. GAT
Well I'm trying to build an HFT market making algo based on the paper here https://arxiv.org/pdf/1105.3115.pdf (using the results at the bottom of page 13). The issue here is that it's tough to paramaterise the equation. I can do a reinforcement type learning and get it to learn from trading, however I'm concerned that it will overfit.
Congrats to GAT et al. January has been a huge month for the trendies. Name-brand shops up 5- 10% as of last week.
Don't jinx it! I've already given almost half of it back in the last couple of days (7.2% DD from HWM @ 25% vol).
I feel your pain! BTW @isotope1 you mentioned a while back that you scale out of positions when loosing - by this do you mean you do this in addition to the natural scaling out of positions which would occur in a decreasing forecast strength? If so do you do this programmatically and if so under what rules?
I do the two normal things: Forecasts do their thing, and obviously reduce a position if the forecast is weakening. If I loose money, I recalculate the portfolio for the new, reduced portfolio (which basically means getting rid of some contracts). (Kelly). These two things are the basic stuff; I don't do anything fancy. Moreover, anything fancy I tried reduced drawdown but at the cost of return. The only exception to this is that I keep a forecast of -10 on VIX & VSTOXX.
Today / yesterday aren't large enough moves to really damage the trendies, unless they had a thumb on the scale by sprinkling in some "risk premia" plays (perma-selling Vix because they naïvely believe the academic literature and/or believe backtests that don't include market crashes).