An Engine, Not A Camera (Donald Mackenzie)
An Engine Not A Camera is either an extremely readable academic textbook, or a heavy going popular finance / economics book. Whilst certainly technical and thorough, the two case studies contain enough narrative to be an interesting read, and formulas / jargon are confined to the appendix. Those wanting to read entertaining stories about Wall Street should read Michael Lewis' Liar's Poker or When Genius Fails by Roger Lowenstein, if you want to go a little deeper, this is the book to read. Like those two, this book is about the Black Monday stock market crash of 1987 (Liar's Poker) and the collapse in 1998 of the massive hedge fund Long Term Capital Management (When Genius Fails); the common thread of the two incidents are the individuals involved, and the common theme of overconfidence in models of economic finance.
Mackenzie sets out to answer the question of whether academic models of finance are (more or less) accurate representations of the world in which we live, or if they are in fact shaping that world themselves (he calls this Barnesian Performativity). Specifically he examines the Efficient Market Hypothesis, the Capital Asset Pricing Model, and the Black-Scholes-Merton model for pricing bonds and options. Structured much like a thesis dissertation (I found this book when looking at potential part-time Masters Degrees, this is on the reading list for Economic History at LSE), the book is well argued and balanced - and light on any unprovable opinions. Therefore his conclusion to the question posed is predictably, a qualified yes - the theory does influence the outcome, sometimes, but not always.
For me, there are two interesting takes from the book.
The 1987 stock market crash came without any real warning, and was unprecedented in its magnitude - a larger single-day drop than the Wall Street Crash of 1929. However unlike 1929, it didn't cause a depression; the recovery was swift, and the impact on the wider world was minimal by comparison. It has been argued that a major cause of the crash was the emergence of portfolio insurance. The new financial models showed that one could quantify risk, and then insure against large losses using derivatives - for example, if an investor's portfolio was reflective of the S&P500, one could buy futures in the S&P at a level where if shares dropped the investor's losses were mitigated. Option trading was more restricted and less sophisticated than it is today, so instead most portfolio insurance consisted of a 'portfolio' of the actual portfolio and cash - and if shares began to drop, automatically the proportion of this portfolio that was cash was increased. If shares fell far enough, the entire portfolio would be converted to cash (sold), to protect against large losses. Of course, the obvious problem here is that if enough people have this mechanism in place, a small drop in share prices leads them all to sell, which leads to a larger drop, which leads to more sales...and eventually a crash. Efficient Market Theory posits that prices reflect all information known by traders, which begs the question as to why companies could suddenly be worth so much less the day after Black Monday than before? One answer is that the 'mechanism' of portfolio insurance isn't acting rationally, it is automatic - so what should look like a bargain to traders (and lead to buying shares) isn't happening.
Where Mackenzie shows real insight is to move beyond the initial consequences of those insurance mechanisms to what is happening to everyone else who is trading shares. It was not the mechanism of portfolio insurance causing the crash (the quantity of these sales was too little to account for such a great fall) -rather, it was the fact the traders didn't know why stocks were falling, and so assumed there was bad news they were unaware of. If traders knew there was a mechanism irrationally selling cheap shares, they would have bought them, seeing the potential bargain before their eyes. Therefore the models fail in two distinct ways. Firstly, portfolio insurance is a perfectly sensible idea unless everyone is doing it - the feedback loop it creates is not predicted by the model. Secondly, an assumption of rationally behaving actors doesn't sufficiently factor in the human element - fear of the unknown is not necessarily irrational, so if the stock market is falling and I don't know why, it is perhaps better to get out of harms way.
The second great insight from the book regards the failure of Long Term Capital Management in 1998. This was a hedge fund set up by a group of superstar Wall Street traders from investment bank Soloman Brothers, and Nobel Laureates Fischer Black and Myron Scholes (of the aforementioned Black Scholes formula). The fund grew massively prior to 1998, until it's losses were so great as to threaten the entire worldwide financial system (the big Wall Street banks reluctantly joined forces to bail out the fund). Their trading strategy can be simplified thus. The firm sought minor discrepancies in prices as predicted by their models, and used gargantuan leverage to take equally gargantuan positions in such trades, turning minuscule pricing 'errors' into extraordinary profits.
The popular narrative as to why this occurred takes two forms. Either the arrogant traders & geniuses were victims of hubris, wrongly believing their own ability to beat the market and so leveraging beyond what could be sustained; or the models themselves were wrong, underrating the true risks involved. Mackenzie doubts both explanations in their most basic form. His explanation is that the well publicised profits from arbitrage trading led to many other actors in the market mimicking LTCM, taking similar or identical trades. This meant that when conditions changed, multiple imitators had to each liquidate similar positions, creating a 'super portfolio', and with it unexpected cascades and 'irrational' prices. The resulting situation was that genuine arbitrage opportunities discovered by traders could persist for months or years before returning to supposed equilibrium. So when a systemic shock (in this case Russian debt default) affected many seemingly uncorrelated parts of LTCMs portfolio, the market could defy the predictions of the models for longer than LTCM could remain solvent.
Both these insights are instructive. The predictive models worked (to some degree) before the world knew about them - then the assumptions breakdown with second order effects. Second, third and higher order effects are impossible to predict, as feedback loops and cascades create unknowable consequences. This might explain why predictive text is still terrible - AI can start to 'learn' our systematic mistakes, but human beings are all the time learning how and where text will correct, adjusting accordingly, and so on. When we look at graphs comparing rates of infection of Covid-19 between countries, it's easy to think comparisons are balanced if adjusted for population, and therefore government A is doing a great job, whilst government B is failing. Why then, at time of writing, does New York have 513 deaths per million relative to 17 deaths per million in California? It could certainly be the result of public policy, but also of population density, super-spreaders, international travel disparities and so on. Higher-order effects are unknowable in advance, as argued by the mathematician Benoit Mandelbrot (one of the fathers of modern chaos theory), featured in this book as an alternative voice to the financial modellers.
This book is a reminder that randomness is far wilder than we can comprehend in our day to day lives. The value of this book, beyond any interest in finance and economics, is the illustration of two real-world examples of why and how this is so.