Rob is doing better than most of us, at least this year. Lots of dispersion out there. For example, today alone, I see DBMF down 1.19% while 40in20out is up 2.61%.
Portfolio thinking - this is the way! I deliberately keep equities futures to be only ~15% of the risk in my system as I already own a bunch of passive equities my portfolio. Since equities have been trending up for a while it's easy to be 'overweight' those in a managed futures system depending on what other investments you have on the side.
You likely know all this and it is out of the context of the paper you referred to but.... I think you can squeeze a lot more out of mean reversion trades by anchoring on to some fundamental that is relevant to the price in that particular asset class. E.g: Equities - some sort of smooth earnings/forward earnings estimates FX - some idea of PPP/fancy Big Mac index with adjustments for income (Balassa-Samuleson paper) Bonds - neutral interest rate (Laubach-Williams style model) or some proxy of this like future labour force growth Comms - demand/supply proxies The challenge obviously is in getting clean, timely, non restated data! Even when you can find the data the risk is the cycles can be long and all the markets within the asset class tend to revert together making it a risky low breadth trade. At least in theory, fundamental mean reversion 'elegantly balanced' with trend and carry could be magic. Some pure price mean reversion things that I have seen kinda work is buy-the-dip style trades in equities - we know equities mostly go up so short sharp down moves likely revert fast. Have you ever looked at relative mean reversion (stat arb-ish) within related commodities - soybeans vs its derivatives. Seed Oils etc... I find that the unadjusted prices have strong linkages but the carry complicates the matter massively. As always thanks for the insightful write-ups!
"think you can squeeze a lot more out of mean reversion trades by anchoring on to some fundamental that is relevant to the price in that particular asset class" yeah sure, and indeed the 'value and momentum everywhere' paper does this to a degree. Someone commented on the blogpost I would have probably got better results with spot rather than futures prices - I could use 'synthetic spot' for sure, and it would be easier than collecting a whole bunch of other data. Combining with carry would probably make sense. Ultimately this stuff is too low value for me to waste more time on it! I'm more intruiged by fast mean reversion. "equities mostly go up so short sharp down moves likely revert fast" yes the linked paper discusses this, and I have the same results in AFTS. I discuss and explain in the book why I don't personally trade these effects. "Have you ever looked at relative mean reversion (stat arb-ish) within related commodities" yes this is covered in AFTS Rob
I tried to check if this has been discussed previously in this thread but I couldn't find it. As most here are trading futures and using IB, I thought it can be a relevant question. IB doesn't pay any interest in cash that is held in the "commodities segment". With rates now at non-negligible levels, it does represent a reasonable drag. IB does facilitate an auto-sweep that seems to minimize the amount of cash sitting in the non-interest-paying commodities segment versus the interest-paying securities segment. Apparently, when one owns T-Bills or T-Bonds (short term?) in the securities segment, and the securities segment cash balance goes negative because it is was used to fund the commodities segment margin, it technically does not generate an interest charge. Has anyone here given proper thought on how to optimize interest income?
Thanks for updating this. I'm (almost exactly) at the $250000 system level, and it is interesting to see how the systematic selection compares to my own (39 vs. 41 instruments; with an overlap of 21), and will certainly think about including some of the instruments in the systematic list but not my own, especially on the commodity and currency side which I've always struggled in selecting an optimal set as the tiny contracts seem to be more correlated with each other tahn the bigger ones. Some quick intersting observation: Only two bonds (KR10 and OAT); but lots of equities (7 country level and 5 sector level), including doubles of some countries (2 korea, 2 japan, 2 china if you count hong kong as seperate).
Equities generally have smaller contract sizes, and more volatility so lower minimum capital. This obviously overcomes the diversification penalty from having too many equities. [A second order effect might be that I just have more equities in my data set] Rob
I'm getting confused about handcrafting. In PST, the full_handcrafting.py script can be used standalone to derive instrument weights as well as trading rule weights. It adjusts the weights of instruments based on volatility, correlation and optional Sharpe Ratio, and volatility targeting. Rob covers this in a series of blog posts on his blog. Where my confusion comes in is how these vol adjusted instrument weights flow through to generating unbuffered optimal position sizing. In AFTS and Systematic Trading - position sizing is equal to: Capped Forecast * Capital * IDM * Weight * Target Risk / (10 * Multiplier * Price * FX * Volatility) We are dividing by volatility again? This seems to have already been addressed in the instrument weight. I'm having trouble tracing through all the stage wiring in PST to see how this works. It seems as if the volatility scalar in class PositionSizing is in fact scaling the each sub system position by the volatility scalar for each instrument. Again, hasn't this already been achieved via handcrafted weights?
I'm guessing your filter to exclude instruments with low volatility and/or high trading costs lead to the exclusion of the shorter term bond contracts (e.g. KR3, BTP3, SHATZ, CAD2) all of which have vol between 1-2%?