Thanks for the clarification. I am not sure how I should calculate that for my account. I started in October 2016. But since then have I added and withdrawn funds a few times, which made parameters such as "high water mark" and "result since inception" a bit meaningless.
Fully automated futures trading
- Thread starter globalarbtrader
- Start date
I have the same problem because I changed my risk target after the first few months of trading, and I'm always taking money out of my account.
The basis for my calculation is adding up the % profit I have made each day over the notional capital employed at that time. That gives me an account curve I can check the HWM for.
Strictly speaking it should be a cumulative product of (1+r%) but I don't keep profits in my account to compound.
GAT
Thank you, this is a good suggestion.
I listened to a replay of your recent presentation for ReSolve, titled "Portfolio optimization when you don't know the future (or the past)". I found it very educational, as it turned a "unknown unknown" into a "known unknown" for me. I had not considered about the impact the uncertainty of historical data has. Your presentation made this very insightful.
However, it did made me wonder: in your first book, Systematic Trading, you explain the handicraft method of first spreading the available account volatility over several asset classes, and then spreading it over several instruments within each class (e.g. chapter eleven). If one uses only a few instruments (e.g. you use three instruments in the presentation and the book) I can understand this approach. However, in "real life" you use much more diversification and probably around 30 ~ 40 instruments. In that case is the cascaded approach converging towards an (1/n) approach, with n being the number of instruments. Each of the instruments will get approximately the same weight percentage, most likely about 3% each.
Add to this the uncertainty of the historical data, and its influence, and it becomes less clear what approach would be the better one: the more complex cascaded approach using correlations and such, or the much simpler (1/n) approach. In your presentation you referred to having had seemingly pointless discussions on whether a portfolio should have "one percent more, or less, of a certain instrument".
Does portfolio optimization still serve a purpose when the number of instruments is large?
It's a good question, and the answer is - it depends on how different the relative correlations are, and the distribution of assets in groups.
Let's take my portfolio; to make the numbers easier pretend I have 40 instruments over 8 asset classes (bonds, stir, equities, fx, vol, metals, energies, ags). For the sake of argument let's suppose we give each 1/8 or 12.5%. Now the equities asset class has 10 instruments in it, so each would get 12.5/6 = 1.25%, about half the 1/n weight of 1/40 = 2.5%. In constrast the STIR asset class has only one: Eurodollar. So under 1/n it would get a 2.5% weight, and under handcrafting which considers the correlation structure it would get 12.5%. That's a substantial difference between lowest and highest weight: a ratio of 10.
If we consider S&P 500 stocks instead, then the 1/n weights would be 0.2% and the highest weight on any single stock from a top down sector weighting would be 0.38% for materials (11 sectors, each gets 9.1%; 24 stocks in sector) and the lowest 0.13% for industrials [another way of thinking about is that under 1/n the weight to the materials sector would be 14.1% and 4.7% to materials, versus both getting 9.1% under top down handcrafting].
With 500 shares the difference between 0.13% and 0.38% isn't going to move the needle very much. Plus the correlations across S&P 500 sectors aren't all that different, whereas across a diversified futures portfolio their is much more difference.
GAT
I must say it is hard to keep the faith having started trend following 12 months ago. I'm down 25% although the lesson is to stick to the programme as large losses came from thinking I could do better than the system when trading NQ. Ha!
This is going to disappear when I am fully automated, still currently acting on the daily email my system sends. Need to remove the meat from the loop
@GAT, what kinds of action do you take (or have coded up) for extreme intraday moves? I can't help but feel I could have saved some pain by using stop loss orders or having some kind of intraday risk manager system in the loop. It felt awful to watch NQ melt away yesterday.
This is going to disappear when I am fully automated, still currently acting on the daily email my system sends. Need to remove the meat from the loop
@GAT, what kinds of action do you take (or have coded up) for extreme intraday moves? I can't help but feel I could have saved some pain by using stop loss orders or having some kind of intraday risk manager system in the loop. It felt awful to watch NQ melt away yesterday.
I must say it is hard to keep the faith having started trend following 12 months ago. I'm down 25% although the lesson is to stick to the programme as large losses came from thinking I could do better than the system when trading NQ. Ha!
This is going to disappear when I am fully automated, still currently acting on the daily email my system sends. Need to remove the meat from the loop
@GAT, what kinds of action do you take (or have coded up) for extreme intraday moves? I can't help but feel I could have saved some pain by using stop loss orders or having some kind of intraday risk manager system in the loop. It felt awful to watch NQ melt away yesterday.
What you have to bear in mind that there are an almost equal number of times when you get filled at a better price than if you'd traded immediately - but of course you don't notice them. If I test this, by delaying the trading in my system by 1,2,3... days, I find that on average I have to delay my trades by about a week before I notice the difference in the second decimal place of a Sharpe Ratio (eg from 0.99 to 0.98).
GAT
Where is this presentation? Could you please share the link, if you have it?
Thanks
Yes, I agree with you on this one. If the asset classes do not each have a similar amount of instruments then you get a larger difference. Same as you does my STIR class only have Eurodollar. I did look to others to add to this class but those contracts were either too large for my account, or the market data subscription was very expensive. However, in other classes (e.g. bonds, forex, equity, ags) I noticed that I'm having more or less the same number of instruments per class and the weight to each instrument became more or less equal. In fact, this made me even doubt: the weight assigned to Eurodollar is now so much larger than any other instrument that it really sticks out. It made me think: is this right, or should I reduce it to make it stick out less? Anyway, I decided to not change it and leave it as is.
In practice I don't put the full 12.5% in eurodollar.
I can't remember if I put this in my book or posted on it, but basically you adjust your weights according to the inverse of the diversification multiplier for that asset class. So an asset class with loads of members, and/or a lot of internal diversification, will get a higher weight than something like Eurodollar.
There is also the fact that STIR is quite correlated with interest rates, so really you should lump them into a 'super sector' when you do your top down allocation.
My sector weights look like this if you're interested:
Ags 20% (highly diversifying)
Bonds 14% (quite correlated with STIR)
STIR 4% (only one instrument, quite correlated with bonds)
Vol 5% (only 2 instruments, quite correlated with equities)
equities 13% (quite correlated with vol)
FX 20%
Metals 16%
Energies 8% (only 2 instruments, but quite diversifying)
GAT