Hi all
Recently I met with a relative who worked in the quant world. I was really interested to know how he made it in the quant world and, especially how his strategies were built and if he has any tips for success.
One tip gained from the conversations is to create a model that manages multiple algorithms. Kind of like what Quantopian is doing, but on a smaller scale. He advised instead of trying to find the mythical 3.0 sharpe algorithm, just keep coding every idea you have, make a dozen uncorrelated 0.5-1.5 sharpe algos and combine the good ones together to get 3.0 sharpe. (Combine smartly, not by equal weighting methods).
Here's what I intend to do in the backtester. Let's say we have 10 strategies we wish to combine in one.
1) Code something that'll track the performance of each individual strategy. Like sharpe ratio, returns and some kind of Bayesian analysis.
2) Use a ranking system to dynamically rebalance the strategies every month, or even disable a strategy.
It would be simple if we have a higher level view that treats each strategy as an individual stock, unfortunately we don't. Has anyone coded something similar? Any suggestions how this structure can be coded in a memory efficient way?
Thanks!
