We've discussed in considerable detail the impact and causes of profit decay for existing trading signals and importance of continuously searching for new sources of trading edge/advantage in terms of researching and building new trading signals. As mentioned, a lot of market participants have entire teams of quantitative researchers implementing and validating new trading signals full-time to achieve this. Searching for new trading signals, or alpha, is an extremely difficult task, and is not a well-structured or well-known process.

Trading signal ideas are brainstormed from live trading analytics, by inspecting periods of losses, or by inspecting market data and interactions between market data, market participants, trading signals, and trading strategies during those times. Based on what is observed and understood from this inspection/analysis, new trading signals are conceptualized based on what appears like it would have helped avoid losing positions, decrease the magnitude of losing positions, help produce more winning positions, or increase the magnitude of winning positions. At this point, the new trading signal is just an idea with no quantitative research or proof to back it up. The next step is to implement the trading signal, then the values output by the trading signals are tweaked and validated to understand its predictive abilities similar to what we discussed in the section on trading signals database.

If the newly-developed trading signal seems to show some potential/predictive abilities, it passes the stage of prototyping and is forwarded to the trading signal optimization pipeline. Most trading signals never make it past the prototype stage, which is part of what makes developing new trading signals extremely challenging. That is often because what makes intuitive sense does not necessarily translate into useful/predictive trading signals. Or the newly-conceptualized trading signal turns out to be quite similar in predictive abilities to already-developed signals, in which case it is dropped since it doesn't offer any new predictive abilities. If it makes it to the optimizing step, then we find the best variants of the newly-developed trading signal and they are forwarded to the step of being added to predictive models. Here, it interacts with other pre-existing trading signals. It might take some time and many iterations before we find the correct method to combine the new trading signal with other predictive trading signals to find a final predictive model that is better than any other. After that step, the new trading signal is used in a final trading strategy with strategy parameters that undergo another round of evaluation and optimization before the final evaluation, where we try to determine whether the addition of the new trading signal improves the profitability of our trading strategies.

We saw how much time and resources need to be invested from brainstorming a new trading signal all the way to making it into a final trading strategy where it can improve profitability. It should be obvious that new trading signals have to pass many intermediate validation and optimization stages, compete with other pre-existing and well-known trading signals, and then interact with other trading signals in a way that improves PnLs by adding new value to the trading strategy's ability to trade profitability. In many ways, new trading signals have to go through a survival pipeline very similar to evolution and natural selection – only the best and fittest trading signals survive to make it to live trading strategies, and others die out. This is what makes developing new trading signals so difficult and a task with a very low probability of success. However, researching new trading signals is mandatory for all algorithmic trading business to compete and stay profitable, making the best quants the most sought-after employees in all algorithmic/quantitative trading businesses in the industry.