Howard Marks, Oaktree Capital — 2022-09-08
-
Japanese
-
Korean
-
Simplified Chinese
-
Traditional Chinese
Memos from Howard Marks 2022-09-08T07:00:00.0000000Z" pubdate title="Time posted: >9/8/2022 7:00:00 AM (UTC)">Sep 8, 2022
- PDF (English)
- PDF (Translations)
- Listen to Memo
- Archived Memos
Subscribe
I've been expressing my disregard for forecasts for almost as long as I've been writing my memos, starting with The Value of Predictions, or Where'd All This Rain Come From in February 1993. Over the years since then, I've explained at length why I'm not interested in forecasts – a few of my favorite quotes echoing my disdain head the sections below – but I've never devoted a memo to explaining why making helpful macro forecasts is so difficult. So here it is.
Food for Thought
There are two kinds of forecasters: those who don't know, and those who don't know they don't know.
– John Kenneth Galbraith
Shortly after putting the finishing touches on I Beg to Differ in July, I attended a lunch with a number of experienced investors, plus a few people from outside the investment industry. It wasn't organized as a social occasion but rather an opportunity for those present to exchange views regarding the investment environment.
At one point, the host posed a series of questions: What's your expectation regarding inflation? Will there be a recession, and if so, how bad? How will the war in Ukraine end? What do you think is going to happen in Taiwan? What's likely to be the impact of the 2022 and '24 U.S. elections? I listened as a variety of opinions were expressed.
Regular readers of my memos can imagine what went through my mind: "Not one person in this room is an expert on foreign affairs or politics. No one present has particular knowledge of these topics, and certainly not more than the average intelligent person who read this morning's news." None of the thoughts expressed, even on economic matters, seemed much more persuasive than the others, and I was absolutely convinced that none were capable of improving investment results. And that's the point.
It was that lunch that started me thinking about writing yet another memo on the futility of macro forecasting. Soon thereafter a few additional inputs arrived – a book, a piece in Bloomberg Opinion , and a newspaper article – all of which supported my thesis (or perhaps played to my "confirmation bias" – i.e., the tendency to embrace and interpret new information in a manner that confirms one's preexisting views). Together, the lunch and these items inspired this memo's theme: the reasons why forecasts are rarely helpful .
In order to produce something useful – be it in manufacturing, academia, or even the arts – you must have a reliable process capable of converting the required inputs into the desired output . The problem, in short, is that I don't think there can be a process capable of consistently turning the large number of variables associated with economies and financial markets (the inputs) into a useful macro forecast (the output).
The Machine
The greatest enemy of knowledge is not ignorance, it is the illusion of knowledge.
– Daniel J. Boorstin
In my first decade or so working at First National City Bank, a word was in vogue that I haven't heard in a long time: econometrics. This is the practice of looking for relationships within economic data that can lead to valid forecasts. Or, to simplify, I'd say econometrics is concerned with building a mathematical model of an economy. Econometricians were heard from a great deal in the 1970s, but I don't believe they are any longer. I take that to mean their models didn't work.
Forecasters have no choice but to base their judgments on models, be they complex or informal, mathematical or intuitive. Models, by definition, consist of assumptions: "If A happens, then B will happen." In other words, relationships and responses. But for us to willingly employ a model's output, we have to believe the model is reliable. When I think about modeling an economy, my first reaction is to think about how incredibly complicated it is.
The U.S., for example, has a population of around 330 million. All but the very youngest and perhaps the very oldest are participants in the economy. Thus, there are hundreds of millions of consumers, plus millions of workers, producers, and intermediaries (many people fall into more than one category). To predict the path of the economy, you have to forecast the behavior of these people – if not for every participant, then at least for group aggregates.
A real simulation of the U.S. economy would have to deal with billions of interactions or nodes, including interactions with suppliers, customers, and other market participants around the globe. Is it possible to do this? Is it possible, for example, to predict how consumers will behave (a) if they receive an additional dollar of income (what will be the "marginal propensity to consume"?); (b) if energy prices rise, squeezing other household budget categories; (c) if the price for one good rises relative to others (will there be a "substitution effect"?); or (d) if the geopolitical arena is roiled by events continents away?
Clearly, this level of complexity necessitates the frequent use of simplifying assumptions. For example, it would make modeling easier to be able to assume that consumers won't buy B in place of A if B isn't either better or cheaper (or both). And it would help to assume that producers won't price X below Y if it doesn't cost less to produce X than Y. But what if consumers are attracted to the prestige of B despite (or even because of) its higher price? And what if X has been developed by an entrepreneur who's willing to lose money for a few years to gain market share? Is it possible for a model to anticipate the consumer's decision to pay up and the entrepreneur's decision to make less (or even lose) money?
Further, a model will have to predict how each group of participants in the economy will behave in a variety of environments. But the vagaries are manifold. For example, consumers may behave one way at one moment and a different way at another similar moment. Given the large number of variables involved, it seems impossible that two "similar" moments will play out exactly the same way, and thus that we'll witness the same behavior on the part of participants in the economy. Among other things, participants' behavior will be influenced by their psychology (or should I say their emotions?), and their psychology can be affected by qualitative, non-economic developments. How can those be modeled?
How can a model of an economy be comprehensive enough to deal with things that haven't been seen before, or haven't been seen in modern times (meaning under comparable circumstances)? This is yet another example of why a model simply can't replicate something as complex as an economy.
Of course, a prime example of this is the Covid-19 pandemic. It caused much of the world's economy to be shut down, turned consumer behavior on its head, and inspired massive government largesse. What aspect of a pre-existing model would have enabled it to anticipate the pandemic's impact? Yes, we had a pandemic in 1918, but the circumstances were so different (no iPhones, Zoom calls, etc. ad infinitum) as to render economic events during that time of little or no relevance to 2020.
In addition to the matter of complexity and the difficulty of capturing psychological fluctuations and dynamic processes, think about the limitations that bear on an attempt to predict something that can't be expected t