Читать книгу Artificial Intelligence for Asset Management and Investment - Al Naqvi - Страница 27

In 2016, when the AI fever was just starting to pick up the pace, I had the opportunity to talk strategy with the co-founder of what he described as the world's first automated hedge fund. In a typical Silicon Valley's passionate and visionary entrepreneurial style, he gave me the analogy of OODA loop, which was a model used to explain the success of the most successful fighter pilots in dog fights in the Korean War. OODA stands for Observe, Orient, Decide, Act. It is a loop where, after you act, you observe your environment again, reorient, decide, and act. Basically, you go through these loops, and you make your trades. Human traders, he argued, are constantly looking, second-guessing, checking their current state, analyzing their profitability, deciding whether they need to make a decision, what decision to make, reading the signals, understating the context, and taking action. As you go through the loop, you adapt and readjust. The AI apparatus he described was composed of random forest and an evolutionary rule-based genetic algorithm (an approach known as Pittsburgh-Style Learning Classifier System, which is used to evolve a solution composed of multiple rules). He claimed that news is a lagging measure in terms of useful signal since news is already reflected in the time series. At best, news can increase confidence in the signal, but it is not signal itself, he contended. Hence, if you genetically evolve the solution, a machine can pick patterns throughout the trading day, and while it can explore millions of options, it may take only a handful of buy-sell actions. He said that his hedge fund was fully automated and required absolutely no human intervention. In his fund, staff set things up, and they do have a big red button to press only if all hell breaks loose—other than that, the system decides everything: what to trade, when to trade it, how to trade it, how long to hold, when to exit. This all sounded like a dream come true for AI. It was perfection at its peak. As long as no one had to push the red button, things seemed in control. Except, in 2018, the hedge fund was suddenly liquidated. Bloomberg reported that the fund had not made any money in 2018, after making a meager 4% in 2017 (Kishan and Barr, 2018). The failure of the fund can be attributed to the approach, or to the technology, or to the lack of human guidance, or to the fact that no single strategy can be viewed as the winning strategy, or that perhaps strategy was not consistent with the market conditions. We may never find out. But what we can learn from this failure are two lessons: (1) the human role in investment management needs to be more than the red button pusher or cheerleader for a strategy; and (2) the role of broader business strategy is critical for investment management. The former because we know that human intervention in the strategy could have observed that the strategy was not working (actually, a machine could have helped point that out) if humans were not emotionally invested in the fully autonomous model and did not view the human intervention as counter to the business model. The latter because we operate in a dynamic system where the environment evolves and a change in strategy becomes necessary. It is likely that if the investment team were surrounded by a diverse set of thinkers, the chances of groupthink could have been minimized. Strategy is not just about investment. It should not be limited to investment strategy. The broader strategy for the organization is just as important, if not more. This chapter is dedicated to strategy building, and its value will become clear as we go through the book.