I use the same weights as for everything else, because I'm not an overfitting monkey.... I don't use just momentum and carry, so a better way to answer your question is to tell you what I have in 'divergent' (like momentum) and 'convergent' (like carry): Code: groups: trendy: weight: 0.6 speed1: weight: 0.16667 relmomentum10: 0.2 breakout10: 0.2 assettrend2: 0.2 normmom2: 0.2 momentum4: 0.2 speed2: weight: 0.16667 accel16: 0.1667 relmomentum20: 0.16667 breakout20: 0.16667 assettrend4: 0.16667 normmom4: 0.16667 momentum8: 0.16667 speed3: weight: 0.16667 accel32: 0.1667 relmomentum40: 0.16667 breakout40: 0.16667 assettrend8: 0.16667 normmom8: 0.16667 momentum16: 0.16667 speed4: weight: 0.16667 accel64: 0.1667 relmomentum80: 0.16667 breakout80: 0.16667 assettrend16: 0.16667 normmom16: 0.16667 momentum32: 0.16667 speed5: weight: 0.16667 breakout160: 0.25 assettrend32: 0.25 normmom32: 0.25 momentum64: 0.25 speed6: weight: 0.16667 breakout320: 0.3333 assettrend64: 0.3333 normmom64: 0.3333 not_trendy: weight: 0.4 mean_reversion: weight: 0.333 mrinasset1000: 1.0 skew: weight: 0.333 skewabs365: 0.25 skewabs180: 0.25 skewrv365: 0.25 skewrv180: 0.25 carry: weight: 0.333 abs_carry: weight: 0.6 carry10: 0.25 carry30: 0.25 carry60: 0.25 carry125: 0.25 rel_carry: weight: 0.4 relcarry: 1.0 Exact weights will depend on the market trading costs, but the top level there is 'trendy': 0.6 and not trendy 0.4; so the same 60:40 weights in AFTS. No, I wouldn't advise it. Unlike trend, carry doesn't cut your position automatically when it moves against you. Rob
@globalarbtrader Hi Rob, I have a question regarding IDM and Weights in Chapter 4 of AFTS. From Appendix B - Use weekly returns for the instrument sub-strategy, assuming that the given instrument has all the capital available, i.e. an instrument weight and IDM of 1. Recalculate the correlation matrix every year, using all available historical data up to that point. Floor negative correlations at zero. My understanding of it - This like like running strategy 3 separately on all instruments using complete capital, combining the strategy's returns and then calculating the correlation matrix every year based on all available history. So let's say run strat3 on S&P, US 10-year bond, WTI CRDUE OIL and GOLD MICRO separately using all capital and all get instrument weight 1 and all get IDM 1. Then using substrategy returns from all 4 backtests, calculate correlation matrix each year based on all available history, cluster the correlation matrix using hierarchical clustering to get weights every year and then calculate IDM every year. Then use those weights and IDM we have at each year when backtesting Strat 4 for those 4 instruments. From Footnote 90 of Strat4 - Actually the method I used is slightly different from what I presented earlier in the chapter, and uses sub-strategy return correlations to group similar instruments together, rather than doing this manually. This makes it suitable for use in backtesting where it can be applied on a rolling basis, using only backward looking data and accounting for additional instruments as they appear in the data set. The IDM was also calculated on a rolling basis, using only historic information; 2.47 is the nal value of the IDM. My understanding of it. So does this mean that when starting the backtest, initially we have just 1 instrument in the dataset, so give it the complete weight of 1 and IDM 1. Then when the next instrument is available, split weights to 0.5 each (This will create unintended PnL for instrument 1). Then recalculate IDM when both have sufficient substrategy return history available. Then when more instruments are available, use hierarchical clustering to get instruments weights and then update IDM when sufficient history is available. I am confused between the 2 ways. I think that for backtesting we should go ahead with appendix B. Could you please clarify? Thank you for your time and your wonderful book. Best Regards, CryptoCaptainX3.
I've read this twice and I think I'm describing exactly the same method with slightly different words? Rob
My confusion comes from the returns that will be calculated. So in the first example(4 separate backtest for 4 symbols) let's say we do those 4 backtests with x million capital for each symbol. And in the second example(portfolio backtest with those 4 symbols) let's say we start with y million. In the second example, when the backtest reaches the second symbol's start point, the starting capital can be different because of the first symbols' substrategy returns. So the denominator used to calculate returns will be different for the second symbol, that will give us a different percentage return and hence correlations will be different. Let me know if I need to be more clear. Do I have to use fixed notional capital to calculate percentage returns like you had done from strategy 2 onwards? Thanks!.
One follow up question, In the backtest, at any point if equity becomes less than starting equity, then for sub strategy returns calculation and position sizing both should use minimum(starting equity,current equity) (half compounding approach)? If not, then can we use any arbitrary large value which is constant as denominator for substrategy returns calculation and half compounding approach for position sizing? If not, then what to use for as the capital value each day?