"When I look at my own performance from back then it's like, 'whoa what happened? Was I just really bad?' I didn't change anything, thankfully, but I was sure scared." - Arbitrage
Richard: We have been thinking about new data: new features and a new target. We talked recently about having a target that's feature neutral in the target. That way you're not paid for any feature exposures. What you would have seen, if we had a feature neutral target, is that those first two months of Kazutsugi would have been kind of normal and people would have made money if they were neutral over that period. But then they wouldn't have made much money in the last month or so where things have gone really well. That's kind of why we gave that feature neutral code away in the analysis and tips doc. You can tweak it yourself, you can say, 'I want to be 50% neutral,' or, '60% neutral,' it's kind of like deciding how optimized you want your portfolio to be. The more neutral, the better to some extent from some perspective of sharpe. However, as many people have noted, there are diminishing returns. It looks like going beyond 50% neutral starts to cut into your profitability.So when we thought about this feature neutral target, which we actually had ready, we decided not to release it and instead tried to make a Kazutsugi-style target without feature neutrality but with a few tweaks that make it a bit better, more performant, and more generalizable. One of the ideas we had is we don't want to make everybody retrain their models if there's only a minor difference in the target, and the other thing is that if we gave out the target, we want models trained on the new target to actually do better than would have even scored against the Kazutsugi target. In a sense, basically, can we make a new target that's strictly better and wouldn't hurt anybody? It feels like we have that, and we're a few days away from feeling like it's ready. It could be within the next few weeks that we put something out as a, "take a look at this," preview. Then everybody will have a chance to play with it to see that it really does help. But, we won't switch to payouts based on it for quite a while. Maybe at the end of June or something, we'll see.That's what we're thinking in terms of new targets. In terms of new features: we're not going to be adding new features any time in the next month or so, but I think maybe a couple new features by the end of the year but we wouldn't want to completely revamp the whole data set. We like what we have now and don't want anybody to think they'll never see Intelligence 9 again - whatever features we have now will stay ****there.
"I would very strongly caution you from even trying to do that." - Arbitrage