The Art of Knowing Nothing

This piece briefly discusses my views into quantitative investment/investment strategies, where opinions outweigh facts, and emotion surpasses rationality. People's understanding may change with experience and knowledge. Perhaps after some time, looking back, I might find my views completely different, but the current understanding isn't necessarily wrong, nor is the future understanding necessarily right. Unlike science, which is progressive, the market is adaptive.

Let’s start with quantitative investment. Terms like quantitative investment, algorithmic trading, high-frequency trading, and quantitative analysis often appear together. For the general public, quantitative investment seems very distant, yet it dominates today's financial markets; it appears that practitioners in this field are all mathematical geniuses, and the trading methods are purely quantitative models, profound but guaranteed to be profitable. Behind crises, quantitative investment is often blamed, such as Black Monday in 1987, LTCM's collapse in 1998, the 2007 quant crisis, the flash crash in 2010, and China's stock market crash in 2015.

But is this really the case? I personally believe that most quantitative investments are merely handing over what humans do to computers, and only a small portion of quantitative investments are entirely data-driven, which I will discuss later. I agree with Rishi Narang's view:

Quantitative investment can be broadly defined as the process of systematising investment strategies, with most of these strategies being quantitative. The difference between quantitative strategies and discretionary investment strategies lies in their development and implementation methods. Quantitative strategies can simultaneously test and trade thousands of investment underlyings, where investment managers often do not know the positions of individual stocks or even whether they are long or short, but they don't care either. Discretionary strategies often rely on some subjective forecasts of the future, with more concentrated bets.

Quantitative investment strategies can be subdivided into theory-driven and data-driven strategies. Theory-driven strategies focus on finance, with typical examples including value, momentum, and low-volatility investment strategies. The quantitative representation of value is to simultaneously buy hundreds of undervalued stocks and short hundreds of overvalued stocks, ensuring market neutrality. For computers, "undervalued" and "overvalued" need strict definitions, and in quantitative investment, there is usually no fundamental analysis of individual companies, but rather using common financial ratios to construct a value index, composed of metrics like price-earnings and price-book ratios. Commodity Trading Advisors (CTAs) are another common user of a theory-driven strategy - time series momentum. They use different technical indicators to construct trend indicators, such as moving average crossovers, Kalman Filters, and HP Filters. In the examples above, quantitative investment is merely implementing proven financial anomalies (value, momentum). These anomalies are also frequently used by non-quantitative investment managers. The difference lies in quantitative investment being highly diversified and disciplined, with no subjective human intervention.

Often, people ask, "Is quantitative investment useful?" The basis is that more quantitative investors in the market might mean the strategies are no longer effective. However, the survival of theory-driven quantitative investment strategies depends only on whether the underlying stories have changed. In non-quantitative hedge funds, stock analysts often look for potential stocks, with the standard being cheap and having a catalyst. Essentially, it's just a combination of two factors: value and momentum. If the value-momentum combined quantitative strategy "is no longer useful," then these non-quantitative strategies wouldn't be useful either. During the internet bubble, many quantitative managers based on value investment were driven to the brink, and more people began to denounce quantitative investment. However, it’s worth noting that Warren Buffett also suffered significant losses during the internet bubble. Claiming that quantitative value strategies are "no longer useful" is equivalent to saying "value investing" is no longer useful. Of course, most discussions about the demise of quantitative investment come from those who don't understand it, ignorant yet feeling their views are valuable. As for whether quantitative investment strategies will disappear due to more market participants, based on strategy crowding metrics (value spread), traditional quantitative strategies have not only not become more crowded but have become cheaper, symbolising higher expected returns.

Another category of quantitative investment strategies is data-driven. These investors take the opposite approach, abandoning all financial theories, and directly mining recurring patterns from financial data, assuming these patterns will continue to appear in the future and investing based on this. These investors often have purely technical backgrounds with no financial knowledge, which isn't a disadvantage but rather a point of pride. The biggest issue is that most patterns found in financial markets are purely spurious. Although these investors do filter patterns, they are more likely to find spurious patterns than theory-driven quantitative investors. An example is Jim Simons' Renaissance Technologies. Jim Simons himself has expressed views consistent with this:

Neither is necessarily superior, but perhaps I inherently believe that markets are mostly efficient, so I lean towards theory-driven models. I personally think market inefficiencies only last for very short periods, and only a few can capture them, like Renaissance Technologies. My guess is that data-driven strategies can only exist in the higher frequency spaces, whereas only theory-driven strategies can exist in low-frequency space.

Most non-factor strategies in the market, without a convincing story, are likely spurious. Among all strategies on Quantopian, the correlation between in-sample Sharpe ratios and out-sample Sharpe ratios is only 0.16, indirectly proving this point.

Why do I value a story so much? Many times, things just exist without explanation. This may often appear in conversations but is logically untenable. Everyone’s ultimate net performance is the market itself; the average value of short-term intervention is zero, excluding transaction costs. If a specific investment strategy can consistently profit, it means someone on the other side of the trade consistently receives below-market returns. To understand why a strategy is profitable, one must understand under what circumstances the counterparty would accept long-term losses relative to the market.

People are willing to sacrifice returns in roughly a few situations:

1. People want to avoid risk, so they are willing to pay a premium to buy insurance. In this case, the strategy's excess returns come from additional risk; you are the insurance provider, losing money when it hurts the most, but earning above-market returns in normal times, which accumulate to outweigh your losses during disasters. The most common strategy is volatility selling.

   

2. Behavioural biases. People systematically do some things wrong. For example, when new information comes out, due to anchoring bias, people underestimate its impact (under-reaction), and then overestimate it as it gets reflected in prices (over-reaction), buying at high prices. Underestimation and overestimation together create the trend phenomenon (one explanation of trends).

3. Some market participants' trades are not profit-driven. The most common investors of this type are central banks and hedgers from the real economy in the commodity market. In this case, speculators' profits come from hedgers' losses; hedgers and central banks don't care about losses. (I never consider speculation a derogatory term; the mad gambling of retail investors is not what I understand as speculation. Speculation should be an intellectual game benefiting the market.)

4. Predictable behaviour of some market participants in the short term; extra returns come from predatory trading. Global macro investors often express their views on specific macro events, like interest rates, through direct risk exposure. But they need to hedge their positions, often choosing the cheapest bonds --- 7-year bonds. This predictable behaviour causes 7-year bonds to be overvalued relative to synthetic 7-year bonds (a combination of 5-year and 10-year bonds). Naturally, a profit method is to go long 5-year and 10-year bonds and short 7-year bonds, maintaining market neutrality.

5. Regulatory/risk control-induced market distortions. Fischer Black's view is that all regulations bring opportunities for extra returns, which is also his market philosophy at Goldman Sachs for many years.

If there’s no story behind a trading or investment behaviour, there might be a problem.

Theory-driven quantitative investors focus on diversification, long-term perspectives, and the story behind the quantitative strategy. Besides this, they know nothing. They don’t know if the market will go up or down tomorrow, the direction of the AUD in the coming months, when the strategy will yield good returns, or when it will be mediocre. Their most commonly used phrase is, in the long run/on average, our strategy creates value. The overlooked subtext is that no one can predict the market, whether quantitative or not. Many claim they can predict the market; I know their success rate, which is 50%. Quantitative investors know nothing, but they believe their diversification and discipline should bring them extra long-term returns.

I “know nothing,” so I can't even be sure if what I've said above will be “correct” in future markets. Markets can take any form. Part of the market's uncertainty comes from its dependence on past markets, making current frequentist methods unsuitable for estimating returns and risks (refer to "Uncertainty"). But this doesn’t mean we’re helpless. Bayesian methods theoretically improve on current methods by considering our existing knowledge. Traditional fields like graph theory, topology, and discrete mathematics, previously deemed unrelated to finance, will far outperform current econometric methods in explaining the complexity and hierarchical structure of financial systems. The advent of quantum computing also brings the possibility of solving traditionally unsolvable problems of undecidability and computational irreducibility, with implications far beyond financial markets, possibly touching the essence of world rules.

Before Fischer Black and Myron Scholes proposed the stock option pricing theory, option pricing was based on conventional rules. After the Black-Scholes Model appeared, actual option pricing increasingly aligned with the model, transitioning from intuition to mathematics within two years. This is the norm for markets; previous markets weren't necessarily “wrong,” current markets aren’t necessarily “right,” markets just change. However, physicists’ understanding of things deepens unidirectionally because the physical world itself isn’t altered by humans.

The above are my reflections, which may not even be helpful for investment, but the thought process might be my greatest gain as an small guy in front of one of the oldest superintelligence - the financial market.

Disclaimer: The data and information mentioned are from third-party sources, and accuracy is not guaranteed. This article shares information and views, not professional investment advice. Consult professional advice before making investment decisions.

This article was originally written in Chinese and posted on my WeChat platform in 2016. The Chinese link can be found here