Similar to why it's important to continuously research and generate new trading signals to stay competitive and build an algorithmic trading business that stays profitable for a long period of time, effort must be made to build new trading strategies that add value to the trading strategies that currently exist and are being run in live markets. The idea here is that since trading strategy profitability is affected by a lot of factors, ranging from trading signal and trading strategy decay, improvements made by competing market participants and changes in market conditions that affect underlying assumptions for certain strategies may no longer hold true. In addition to continuously optimizing existing trading strategy signals and execution parameters, it's also necessary to invest resources in adding new, uncorrelated trading strategies that make money in the long run but perform differently. These new strategies should counteract the possibility that some trading strategies will go through periods of reduced profitability or diminishing profitability due to market conditions or seasonal aspects.
Similar to researching and building new trading signals that interact with other trading signals to add non-overlapping predictive powers, we need to build new trading strategies that interact with other pre-existing trading strategies to add non-overlapping sources of profit. It is important that newly-developed trading strategies make money during periods where other trading strategies might be losing money, and that newly-developed trading strategies don't also lose money when other trading strategies are losing money. This helps us to build up a diverse pool of trading strategies that rely on different trading signals, market conditions, market participants, relationships between trading instruments, and seasonal aspects. The key is to build a diverse pool of available trading strategies that can be deployed to live markets in parallel with the objective of having enough intelligent trading strategies running. This helps in dealing with changing market participants/conditions, which can be handled better under the assumption that since the trading strategies are based on different signals/conditions/assumptions, it is unlikely for all of them to decay simultaneously, thus reducing the probability of significant profit decay and of complete shut-down of the algorithmic trading business.
Trading strategies that we've covered in previous chapters that complement each other include trend-following strategies combined with mean-reversion strategies, since they often have opposing views on markets that are trending/breaking out. A slightly less intuitive pair would be pairs-trading and stat-arb trading strategies, since one relies on a co-linear relationship between different trading instruments holding and the other relies on a co-related lead-lag relationship holding between different trading instruments. For event-based trading strategies, it is better to deploy them simultaneously with their trend following as well as mean reversion bets. The more sophisticated market participants usually have combinations of all of these trading strategies with different trading signals and parameters deployed to different asset classes over multiple trading exchanges.
Thus, they maintain an extremely diverse range of trading exposure at all times. This helps to deal with issues of profit decay in trading signals and trading strategies, and optimize risk versus reward, but we will explore that more in the next section.