Search results

That is interesting, and there are many possible explanations. Leave it with me GAT

It's possible to test this by plotting the average forecast over time GAT

Not any more. But yes before when it did very occasionally GAT

About a third of the largest drawdown (in 2018/19) is caused by VIX; in particular VIX is short and gets smashed in early 2018. I have a 5% allocation to VIX (and another to V2X), so reducing these would be an easy way of improving the skew/drawdown whilst of course hurting the SR...

There is now a live vs backtest plot at the end of the blog post, for those who are curious. GAT

Variable IDM Doesn't make much difference after the first few years, since the IDM goes up to the maximum of 2.5 fairly quickly, but definitely much more even risk. GAT

I run a 20 year backtest to save time/speed/memory; 2 years is probably enough frankly. There is no path dependence since I only do a single dynamic optimisation based on what my positions actually are in the database. GAT

I'm struggling to see why you'd have correlations for instrument returns but not substrategies... some kind of warm up?? As a heuristic in ST I suggest using 0.7*instrument correlations GAT

Just for fun, I'm going to run a sim with an estimated IDM so you can see the effect that has. I also want to dig a little deeper into the significant drawdown that shows up last year. Finally, just for fun, I thought it might be interesting to compare my live trading record with the backtest...

The higher vol in the 1970's is because I'm using a fixed IDM - nothing to do with the vol of the underlying markets. There are fewer markets back then, so the IDM of 2.5 is way too high. With a properly estimated variable IDM it would be on target in every decade. That would also even out the...

Yes it's on my to do list for this year as I'm also going to be writing about spreads in the book. Yes GAT

Try it now GAT

1 = 1% GAT

At HFT frequency the correlation between stocks at tick level will be effectively zero, and not informative (since the price changes will be dominated by the noise of the bid/ask bounce). Probably five minute data is the lowest frequency at which correlations make sense. Actually correlations...

A very subtle question, I like it! Well, there's an implicit long bias anyway, because most markets have gone up over time I will always have more +20 and -20. And the effect you describe is probably strongest in clear risky assets like equities (or risk off assets like vol). There is some...

I wish... but I'm skating a bit on thin ice with the data I supply in pysystemtrade as it is. Although it's 'user generated' (not raw contracts prices, only backadjusted) and I gain no commercial benefit, ultimately it's probably a data licensing grey area. I think if I put up the adjusted...

Just checked. It's actually 1.05! (I gues the last few days hurt the SR more than you'd think) GAT

2011 to now Always with costs GAT

Sharpe is .... well let's round it up: 1.0 perc.stats() [[('min', '-9.323'), ('max', '5.493'), ('median', '0.1334'), ('mean', '0.07394'), ('std', '1.185'), ('skew', '-1.052'), ('ann_mean', '18.91'), ('ann_std', '18.97'), ('sharpe', '0.9967'), ('sortino', '1.17'), ('avg_drawdown', '-13.53')...

I was eyeballing on daily returns, but you're right the monthly SR is 5.5 (assuming zero risk free). Which is the equivalent of about SR 1.23; which by coincide is roughly what my own backtested system has and I have ~50 years of data and ~114 different markets. Anyway the point still stands...