As the world becomes more digitised, so too do the markets. With huge swathes of market data becoming more readily available each day, major traders including hedge funds and investment banks are relying more heavily on statistical models derived from this data to conduct autonomous trades and speculate on future market movement. This practice, often referred to as algorithmic or ‘quant’ trading, has become a dominant force in the financial world with 80% of the daily trade volume of some markets being conducted entirely by autonomous trading systems.
Sophisticated systems of this nature often rely on programmatic techniques such as neural networks, back propagation, machine learning, and sentiment analysis to forecast how they expect markets to behave, and then conduct trades accordingly. However, despite their sophistication, these systems share a common characteristic with their rudimentary counterparts; their reliance on historical data to forecast expected market behaviour. For this reason, most automated trading systems in their current form are essentially incapable of predicting how a market is likely to behave in a time of crisis.
The proliferation of trading systems that exhibit this flaw is an issue which often presents itself implicitly in the form of price distortion and nonsensical market movement. In this piece we’ll briefly discuss several examples of such events, and the market inefficiencies that persist as a result of automated trading.
JPY during 2017 North Korean Crisis
A prime example of how algorithmic trading can lead to nonsensical movement in the markets, occurred on August 29, 2017. On this date, North Korea launched a ballistic missile over the Japanese island of Hokkaido. The launch followed months of escalating tensions resulting from North Korea’s successful test of both a missile capable of striking the mainland United States, and a nuclear warhead small enough to fit on it.
Soon after the launch, North Korea’s media appeared to threaten the US territory of Guam with a preemptive military strike; a threat which brought humankind the closest to a nuclear war that it had come since the Cuban Missile Crisis. Such a war that would likely devastate the Japanese homeland due to their proximity to North Korea, and wreak havoc on their industrial economy for decades to come.
Given the strategic and economic gravity of the situation, a fundamental investor might expect the Japanese Yen to depreciate relative to a currency more isolated from these incidents such as the Euro. However, on the five days following the launch, in spite of the fact that Japan was facing possible economic devastation, the Yen actually strengthened by 0.9% on the Euro. This is part of a broader trend; as tensions with North Korea ratchet up and indices fall, the Yen rallies.
There are two primary reasons for this; the first is that Japan is a NET overseas investor, meaning that when tensions are heightened, Japanese investors seek to repatriate their money to avoid FOREX risk, increasing demand for the currency and superficially inflating its value. The second, however, is that the Japanese Yen has historically been considered as a ‘safe haven’ currency due to its resilience in times of economic crisis.
As a result of this precedent, and the statistical inference that derives from it, many models used in automated trading systems would purchase the Yen during times of crisis due to its negative historical correlation with bullish indicators such as the MSCI, DJIA and Nikkei225, and positive correlation with bearish indicators such as AUDUSD. However, these models are relying on historical datasets that clearly do not capture the effect that a nuclear war would have on the Yen’s fundamental value.
An interesting metric to look at when considering ‘safe-haven’ assets such as the Yen, is the rolling correlation that the return’s of these assets have with those of other indicators such as the MSCI World Index. Safe-haven assets often have a negative correlation with these indicators, meaning that when one goes down, the other goes up (and vice versa).
By analysing the 36-month rolling correlation between a safe-haven ‘asset’ such as the JPYEUR and an index such as the MSCI World, it is possible to see how the correlation and ‘safe-haveness’ of the asset changed during the crisis. As can be seen from the chart above, the negative correlation between the two weakened during the 2017-2018 North Korea Crisis, and strengthened again once the crisis abated. This suggests that while the ‘safe-haveness’ of the Yen declined during the crisis, it was still perceived by the market as being an effective hedge against wider market volatility.
The Stay-at-Home Sharemarket
A more recent example of the statistical models used in automated trading systems nonsensically affecting market prices, occurred in March of 2020 as a consequence of the COVID-19 pandemic. In the initial four weeks of the outbreak, the S&P500 declined by almost 35% from its peak; the biggest collapse in the US stock market since 1987. The adoption of nation-wide lockdowns, stay-at-home orders and strict curfews meant that many people were unable to work, socialise, or travel. It also meant that people would on average be spending far more time at home.
A fundamental investor may have been inclined to believe that this would result in people spending a larger portion of their disposable income and time on entertainment platforms accessible from their homes. However, during the initial outbreak in March, listed home entertainment companies such as Netflix, Activision, Nintendo, Sony and Pinterest declined in parallel with broader market expectations:
*SHIX is a pseudo-index of ATVI, NFLX, SNE, and NTDOY weighted by their market capitalization on 19th August 2020
As can be seen from the above charts, the returns of ‘stay-at-home’ shares such as NFLX, ATVI and SNE only began to diverge from those of the broader market almost three weeks into the outbreak. In the case of Netflix, the cause of this divergence was not the release of a positive equity research report by a prominent analyst, or a change in buy/sell recommendation by a reputable asset management firm, but rather it was the news that Netflix was lowering the bitrate of its streaming service to cope with a spike in active users. A rational and fundamental investor would have been able to preempt this spike and capitalize on it, however the automated trading systems that dominate the modern markets were not capable of doing so.
Market Inefficiency & Alpha
The key take-away from these two case studies is that the market can often behave nonsensically as a result of the statistical models and algorithms which underpin the vast majority of trades. These models rely on technical indicators and economic data, but in doing so adopt a technical trading approach as opposed to a fundamental one. They do not consider the drivers of value for these assets, and fail to preempt how crisis events such as a conflict in Asia or a pandemic such as COVID-19 will affect the fundamental value of these assets.
Instead, they use statistics and technical indicators as heuristics or “rules of thumb” to project future market behaviour, based on what has happened previously. The issue with this is that many crisis events, and the downside risks which stem from them, are unprecedented and so cannot be modelled using historical precedent. Consequently, many of the autonomous trading systems and algorithms that underpin our economy are ineffective in modelling crisis events and their effect on the market.
So long as automated trading systems remain incapable of modelling market behaviour during times of crisis, the markets will remain inefficient. This inefficiency is an opportunity for fundamental investors to capture the alpha that these trading systems cannot. However, it is also an opportunity and a cue for algorithmic and quantitative trading firms to develop methods that will allow them to more accurately model the economic implications of crisis events. This could potentially be achieved by using a combination of deep learning and alternative data to track an asset’s estimated value in real-time.
1: Bigiotti, Alessandro; Navarra, Alfredo (October 19, 2018), “Optimizing Automated Trading Systems”, Advances in Intelligent Systems and Computing, Springer International Publishing, pp. 254–261, doi:10.1007/978-3-030-02351-5_30, ISBN 978-3-030-02350-8
Edited by Victor Yan, Dominic Holden and Gary Palar
Disclaimer: The views expressed in this article are solely that of the author’s, and do not necessarily reflect the position of UNIT nor the University of Melbourne. The advice given is general in nature and does not consider an individual’s personal financial circumstance. Transacting off this information is done so at one’s own risk, and individuals are encouraged to consult a finance professional before making investment decisions based off of this article.