1 The Search for a Unifying Theory It is the dream of many researchers and practitioners, whatever their field of study, to come up with a construct that explains all observed behavior and a template for forecasting future behavior. In the physical sciences, that search is abetted by nature, which imposes its order on observed phenomena, allowing for cleaner tests of any theory. In the social sciences, the search has been less focused, partly because human behavior does not always follow predictable patterns. It is easy to understand why we search for universal theories that explain everything, since they offer the promise of restoring order to chaos, but that search comes with risk. The most significant risk is overreach: sensible theories get pushed to their breaking point and beyond in order to explain phenomena that they were never meant to cover. Once a theory becomes prevailing wisdom in a discipline, the temptation to use it to explain everything becomes overwhelming. The second significant risk is bias, which takes shape as a theory''s most ardent supporters become selective in their assessment of evidence, choosing to see only what they want to in the data, focusing on supportive evidence and denying evidence that contradicts their theory. Eventually, if a theory has weak links or is wrong, the weight of data or evidence contradicting it will lead to its modification or abandonment-but not before its pursuit by single-minded supporters creates damage.

The Search in Finance and Investing Economics is a social science, but what sets it apart from the other social sciences is the easy access that its theorists have to rich economic data and, especially, market data. Researchers and many practitioners have tried, over time, to come up with economic theories or models that explain everything from how businesses make investments to financing and dividend decisions and how investors price companies. In this section, I will lay out some of the attempts over the last seventy years to build an overarching theory of finance-and explain why they have all fallen short. Economic Theories To the extent that finance is an offshoot of economics, it stands to reason that many of the early theories in finance came from economics, with economists'' work on risk aversion and utility functions animating the search for financial theories that would explain market pricing and investor return. It can be argued that modern finance had its beginnings when Harry Markowitz, with an assist from the field of statistics, put forth his work on modern portfolio theory. In effect, Markowitz drew on the law of large numbers to argue that investing across multiple risky assets that do not move together yields better return payoffs, for any given level of risk, than investing in an individual asset. The Markowitz efficient frontier provided an elegant way of compressing the investment process into a search for higher returns, with risk operating as a constraint. Figure 1.

1 The Markowitz Efficient Portfolio The power of Markowitz''s theory went well beyond the optimized portfolios that it could be used to generate, since it upended the very notion of risk in markets, supplanting the old idea that investors should assess risk on an investment on a stand-alone basis with the idea that the risk of an individual investment comes from the risk it adds to a portfolio of investments. By introducing a riskless asset into the Markowitz universe, John Lintner and William F. Sharpe changed and simplified the efficient frontier. They showed that for all investors, no matter their level of risk aversion, a combination of a riskless asset and a supremely diversified portfolio (labeled the market portfolio because it includes every traded asset in the market), held in proportion to each asset''s market value, would generate a better risk/return trade-off than any portfolio composed purely of risky investments. Figure 1.2 illustrates the effect. Figure 1.2 The Capital Asset Pricing Model Lintner and Sharpe''s capital asset pricing model ("the CAPM," as finance geeks refer to it) also had a reach that well exceeded the core application, since it allowed for a linear equation that could be used to both explain past returns and predict future ones on risky assets: E(Return on Investment j) = Risk-free Rate + Betai (Expected Return on the Market Portfolio − Risk-free Rate) The reach of this equation, extending from businesses using it to determine their hurdle rates (for accepting or rejecting investments) to investors using it to estimate the expected returns on individual stocks and portfolios, made it one of the most widely used and studied economic models in history.

Those studies, though, uncovered a painful truth: the model lacked predictive power with regard to large segments of the market. The benefit of the grounding in theory that characterizes theory-based models-wherein you start with economic first principles and build up to models-is that the development process constrains you from fitting the data that you observe to preconceptions that you may have about how the world operates. The drawback of these models is that for them to be useful, you have to make simplifying assumptions about human behavior (ranging from how human beings derive utility to what comprises rationality), and to the extent that these assumptions are erroneous, you will end up with models that are elegant in theory but have little real-world explanatory power. Data-Driven Models/Theories Just as the Markowitz portfolio theory and the capital asset pricing model were being advanced as offering answers to all finance-related questions, a group of researchers centered at the University of Chicago were advancing a different approach, grounded in the belief that markets were efficient and that market prices were therefore the ultimate signals of truth. In the efficient-market world, the market response provides the tiebreaking answer to the question of whether a business decision was good or bad, with good decisions provoking positive market reactions and bad decisions resulting in negative ones. The efficient-market view of active investing, wherein investors attempt to time markets and pick the best stocks, was that it was pointless, since market prices already reflected all available information. With an abundance of data (at both market- and company-level) that has been easy to access for decades, you could argue that finance discovered big data well before the rest of the world caught on to its allure. In fact, the first model to seriously challenge the capital asset pricing model was the arbitrage pricing model, in which researchers using observed data on asset prices and related expected returns to statistical (and unnamed) factors.

In effect, in the arbitrage pricing model, you assume that if risky assets are priced in the market to prevent riskless profits (arbitrage), you can back out the risk factors from the pricing. These data-driven pricing attempts, which began in the late 1970s, picked up steam in the following years as access to macro- and microeconomic data widened and deepened, resulting in factor pricing models. In 1992, Eugene Fama and Kenneth French looked at returns on all US stocks between 1962 and 1990 and noted that a significant portion of the variation in annual returns across stocks during this period could be explained by two characteristics: their market capitalizations and their book-to-market equity ratios. Specifically, they found that small-market-cap and high book-to-market (low price-to-book) stocks earned higher annual returns than large-market-cap and low book-to-market (high price-to-book stocks. They attributed the higher returns to the risks in small-market-cap and low price-to-book stocks. In the years since, with access to more and richer data, researchers have added to the list of characteristics that explain differences in market returns, in what can broadly be categorized as factor pricing models. By 2019, there were more than 400 factors that had been identified as explaining price movements and differences in returns in major finance journals, leading some researchers to talk of a "factor zoo" and argue that most of these "market-explaining" factors are more attributable to data mining than to market behavior. While academics were attracted to data-driven pricing models because of their capacity to explain investor and market behavior, practitioners were drawn to these models for a much more prosaic reason: to the extent that these models can uncover market mispricing, they offer the potential for profits to those who can find those market mistakes and benefit from their correction.

Jim Simons was an early adopter, and his mathematical and statistical skills allowed him to earn market-beating returns for decades. In more recent years, quantitative investing has drawn more players into this game and, with powerful computing added to the mix, driven down the returns available from using data to find investment opportunities. Put simply, using powerful computers to find moneymaking opportunities, as high-frequency traders did in the early part of the last decade, comes with a countdown clock for that profit making as new investors enter the market with their own computing power. Data-driven pricing models do have an advantage over theoretical models in their capacity to explain observed behavior, but you can argue that this is an unfair test, since a data-driven model preserves the ability to add more or different variables to improve explanatory power, unconstrained.