In the previous section, we discussed the advantages of having a diverse set of trading strategies that rely on different trading signals. Here, each trading strategy is profitable by itself, but each one's performance is slightly different depending on market conditions, market participants, asset class, and time periods, which are, to a large extent, un-correlated to each other. To recap, the benefits are greater adaptability to changing market conditions/participants and a better risk-versus-reward profile for the entire portfolio. This is because all strategies do not lose money simultaneously, which would lead to very large drawdown across the entire portfolio of trading strategies deployed to live trading markets. Say we have a diverse set of trading strategies, how do we decide how much risk to allocate to each trading strategy? That is a field of study known as portfolio optimization, which has entire books dedicated to understanding the different methods involved.

Portfolio optimization is an advanced technique for algorithmic/quantitative trading, so we won't cover it in too much detail here. Portfolio optimization is the technique of combining different trading strategies with different risk-reward profiles together to form portfolios of trading strategies which, when run together, provide optimal risk-reward for the entire portfolio. By optimal risk-reward, we mean it delivers maximum returns while trying to minimize the amount of risk taken. Obviously, risk versus reward is inversely proportional, so we try to find the optimal reward for the amount of risk we are willing to take and then use the portfolio allocation that maximizes the total reward of the portfolio while respecting the maximum risk we are willing to take across the portfolio. Let's look at some common portfolio-optimization methods, and observe how allocation varies for different allocation methods.

Note that implementation details of different portfolio allocation techniques have been omitted here for brevity's sake, but if you are interested then you should check out https://github.com/sghoshusc/stratandport for a project that implements and compares these different methods. It uses mean-reversion, trend-following, stat-arb, and pairs-trading strategies applied to 12 different futures contracts and then builds optimal portfolios using the methods we've discussed here. It uses python3 with the cvxopt package to perform convex optimization for markowitz allocation and scikit learn for the regime predictive allocation and matplotlib for visualization purpose.