I took a look at this paper. I was thinking about how to implement this in pysystemtrade. I decided to replace spot with price contract, near term future contract with price contract, and distant future contract with carry contract. I could certainly be wrong, so test for yourself, but when I did that the formula pretty much reduced to how Rob calculates carry.
If I recall correctly (haven’t looked in a bit) not much. I may not be thinking about it correctly, but I consider the carry signal as an “offset” to the trend signal, that is does carry give a headwind (how much do I have to overcome for the trend signal) or tailwind to trend.
Thanks, similar - shorts just don't work so well for any longer term strategy except in specific short periods of stress for particular markets (and in those spots we're glad to have them). I recently tested a few more strategies from AFTS. "Acceleration" and "Skew", similar experience there so much long attribution to performance. Skew has a good reason though, unless it's for trading Vol Indices which is very negatively skewed. You probably are thinking fine - it depends on how you built your system. Can be thought of the other way as well, using absolute carry also has trend signal in it unless you intentionally use techniques to get rid of it.
Do you observe the 'longs typically work better' phenomena outside of equities and bonds as well? To use @Plex134 term - these asset classes have a clear risk premium 'tailwind' - which carry has to time correctly in order to over come?
Yes, I don't have clear answers but I think it's not due to risk premium tailwind (probably both directions have risk premium tailwind at times, taking on risk that others don't want). My guess comes from looking at the distribution of winners, longs have ranges from 1% to values much higher than 100%. Percentages don't really make much sense with futures but you get the idea.. In other words, longs are more asymmetric as bets. Shorts seem to have ranges much less because normally prices rarely get close to zero, though Contango can keep pushing things down for a long time it's weaker than the effect of a bull market in longs. E.g. the max favourable move in Cocoa is many multiples of 100% to use an extreme example, that kind of thing doesn't really happen in shorts. If anything, I would've thought Bonds would be even better on short side than commodities because they are less volatile so have more room. E.g. STIRs are just an artificial price based on 100 minus the rate. Maybe the reason we haven't seen that yet is because of the regime of rates lowering over so many years in most of our backtests more recent periods. Currencies I think the same (and Currencies are the one thing in my backtest that does better on short side than long side) but they aren't a huge driver - most are very similar/correlated to each other. All this depends on how long term your trades are and how much you allow trading to adjust sizing, or even if you decide not to adjust existing positions at all. Second theory is that it's related to the era we have lived through and backtest on. If we go through 30 years of deflation due to AI in the future, maybe we have long term price declines (though I think the central bankers or politicians will not let that happen). ps my last reply had a typo, I meant vol indexes like VIX are in the positive skew side (short) of the Skew Strategy not negative skewed.
Convexity. Longs can go up an infinite amount but shorts can only go down 100% (in most cases, see front-month crude oil in 2020). The more long-term you trade and the less you change position sizes, the more this convexity matters. (If you have a holding period of one-day, the upside and downside potential are about equal, ignoring skew effects but that's a different subject. And if you adjust position sizes to volatility, you'll reduce, but not eliminate, the amount of convexity.)