School of Quants. Python project 3. Top-performers continuation of performance strategy. CVAR, Max Sharpe portfolio optimizaiton.
My strategy is based on trend continuation, best performers will likely remain top-performers for the next couple of weeks and worst will remain the worst. I long a portfolio maximizing Sharpe ratio over the best stocks.
Max-Sharpe with best performers and Max-CVAR
Whatever I tried I saw nothing better that a coinflip 50-50, this is understandable given that we deal with financial data most of which is generated by unit root processes. I think ML could be used when there is actually something to detect, like manipulation in pumps and dumps in cryptocurrency markets (my current project), there I see how one can use ML/DL for classification and therefore use the logits for some sort of portfolio weights. We could have trained a model prediciting returns in 14 days, but I bet it would have been total garbage.
