The FICC Market Standards Board Ltd. (FMSB) is a London-based standards body for participants in the wholesale fixed income, currencies, and commodities (FICC) markets. It has been looking into the root causes of market misconduct, and pursuant to that research it recently published a report on the “themes and challenges” in algorithmic trading and machine learning.
The author of the report is Rupak Ghose, former head of corporate strategy for ICAP/NEX and a former equity research analyst at Credit Suisse.
Ghose writes that for a long time “algorithmic trading has been most prominent in highly liquid markets, which have significant amounts of high-quality data,” but as of late the application of algos has expanded to less liquid products, increasing the challenge of securing the quality and consistency of data. Highly sophisticated machine learning may be of negative value if the data is misleading.
A related problem with markets in illiquid products is that they are likely to have very few non-bank market makers willing to extend liquidity in crisis conditions. Such conditions raise the ante on model validation and the threat posed by tail risks.
Managing Model Risk
For such reasons it is vital that institutions manage their increased model risk, Ghose says. Model risk (he reminds us, citing a US Federal Reserve System report), can mean that a model has fundamental errors and produces inaccurate outputs, or it can mean that the model is being used incorrectly or inappropriately.
Model risk is a factor in human-centered trading as well, but it becomes increasingly dangerous as humans withdraw from the scene. One problem is that there are often “very simple model assumptions made within an algorithm, for instance the use of moving averages in price computation.” The simplicity of the assumptions and the inflexibility with which they may be embedded should be considered in connection with the governance framework.
Due to such issues, machine learning in wholesale FICC markets will likely be limited, for an extended period yet, to being a tool for the performance of “specific minor functions only and as a relatively small part of the overall trading and reporting process with tight controls in place.”
If the use of algos and machine learning is taken too far too fast, “market abuse or stability risks” seem likely to emerge, and there is increasing discussion within the industry about the best practices that may mitigate these risks.
The FMSB report quite sensibly distinguishes between pricing and risk models on the one hand, and algo trading models on the other. Given the differences, it says, “different model validation approaches may need to be developed.”
Especially in the latter case, that of algo trading models, stress testing may be needed. An investing institution needs to develop “scenario analyses that stress test data inputs and their impact on algorithmic models.” These analyses should involve negative stress testing in order to find the conditions under which the model assumptions would break down. The algo models of different firms may be similar, and that fact may create interdependencies, which in turn generates an unintended risk.
The report recommends that algorithmic governance ensure that the algorithms used for trading are fair and that they do not encourage market abuse or market stability risks.
Some of the established liquidity risks perpetrated through the new technologies, such as spoofing and collusion, are “easier to perpetrate in conditions where public reference prices are harder to establish, as may be the case in these less liquid products.”
As for machine learning, the report suggests ethical governors for the algos (science fiction readers will recognize the idea—Asimov’s robotics rules for financial markets). Ghose cautions that “a machine optimising on its own will ‘discover’ that unethical, manipulative trading practices are more profitable than ethical trading” unless ethical benchmarks are programmed in that will reject trading tactics that violate certain standards.
More broadly, the report proposes that the carbon-based life forms of the FICC world will benefit from guidelines that “make the traditional model validation process more suitable for algorithmic trading.” This could make the trading more efficient, as well as reducing the risk of market abuse and lessening the threat that a rapid expansion of the new technologies to illiquid markets could otherwise pose to market stability.