A world champion trader explains the Variance Accelerator Model, which detects when a strategy shifts from positive to negative autocorrelation after about 7-8 losing trades. That reversal signals an imminent win, so he scales position size from 0.5% to 3% of capital using the Kelly criterion to capitalize on the edge.
This is the model that I used in the championship. When I saw that my streak of losing months were not like this, but more like this, I started to go deeper on the autocorrelation and I built a model that I called Variance Accelerator Model. If you analyze your drawdown and your data in different states, you can get multiple autocorrelations in the same strategy. If I'm analyzing my whole equity line, I know in all my data I have the positive autocorrelation and that's it. If I analyze Some kind of data, the streak, for example, I can get totally different number of autocorrelation. And I saw that if I analyze this, my autocorrelation shift from positive to negative. So I started to test when the streaks will most likely comes to an end by using the negative autocorrelation. After a losing streak of 7 to 8 trades, my autocorrelation became strongly negative. What does it mean? So if 7 and 8 are losses, my autocorrelation shifts, so I will probably get a winner. And this I used combined with Kelly for understand how much my streak go deeper in my data. And that's what I apply in this 2 model, the VAM I would call it, because it's a combination of this concept and Kelly for understand how much should I risk. If my autocorrelation is really high, I can go from 0.5% risk that I'm using to 3%.
