I once did some consulting for a very well known person in the financial services industry, a recognized leader in the field. As part of that work I got to see his financial "models". They took the form of the most incredibly complicated Excel spreadsheets I have ever encountered. They were real monsters with hundreds of thousands of rows, dozens of workbooks.
Being a software engineer, I wrote some code to reverse-engineer them. I wrote some VB to dump all the formulas as text, and then wrote a parser and code analyzer. The result was a real eye-opener. 90% of the formulas were dead code that didn't actually play into the results at all. The rest was still a mess, but when I started combing through it I discovered that even the live code didn't really matter to the results because at the end there was a "fudge factor" that the guy just set according to his gut feel about the market, and this was the main thing that determined the final result.
Financial analysts use one of three approaches. There's the "whatever my gut tells me" approach, which is surprisingly common. There is modern portfolio theory. And nowadays there's machine learning. And it's all bullshit. MPT is bullshit because it uses volatility as its model of risk, which is IMHO the wrong model. They use it because it's something they can measure and so write academic papers about without sounding like they're just making shit up, but volatility does not actually correspond with what most people consider the word "risk" to mean. It also relies on the assumption that past volatility is a reliable predictor of future volatility for a particular security, which is often not true. And ML is bullshit because if you throw enough data at a regression algorithm it will find something that looks like a signal but will in fact almost certainly just be a coincidence if you actually do the statistical analysis properly.
So yes, I'm probably being a bit harsh. There are probably analysts who actually know something, and MPF is actually not totally worthless. But "it's all BS" is a pretty good first-order approximation.
The thing is, it's impossible to tell the difference between clever and lucky even in retrospect. It's entirely possible that his models were BS. Even a broken clock is right twice per day.
It's not impossible to tell the difference between a lucky guess and a well-designed model, although in finance you can often be right for the wrong reasons or wrong for the right reasons.
The issue is that the things you're trying to model are usually extremely complex and dynamic, in a way that's fundamentally unlike most sciences where the objects of analysis obey more or less immutable laws.
So it's less like trying to model where a kicked ball will land in a football game, and more like trying to model where the ball will be after 30 seconds of passing and running.
But Michael Burry wasn't doing that, and his models weren't wrong. He correctly saw that the characteristics of the mortgage contracts that were being put into MBSs invalidated the assumption of uncorrelated defaults that were being used in the banks' and ratings agencies' models.
Well, bonds are incredibly macro-sensitive, so when the course of the world economy can be upturned by one Trump tweet (or whatever) then yeah, a super complicated model that you throw away and replace with a fudge factor sounds like about the right way to model it.
100% I've seen the same thing. But they think they're all brilliant masters of the universe. Frightening when you think on how common place that is in the industry. That said, I do believe that the larger firms have begun to evolve away from that.
Firms in finance run the gamut from being breathtakingly stupid to scarily smart. Since the financial crash things have changed a lot (less stupid guys) though.
I mean surely you don’t think Goldman Sachs, Morgan Stanley, JP Morgon, etc are still around because of survivorship bias? That is, everyone is just flipping coins and some people got lucky.
If you think that’s true, the natural correlary is that all companies are just flipping coins.
I once did some consulting for a very well known person in the financial services industry, a recognized leader in the field. As part of that work I got to see his financial "models". They took the form of the most incredibly complicated Excel spreadsheets I have ever encountered. They were real monsters with hundreds of thousands of rows, dozens of workbooks.
Being a software engineer, I wrote some code to reverse-engineer them. I wrote some VB to dump all the formulas as text, and then wrote a parser and code analyzer. The result was a real eye-opener. 90% of the formulas were dead code that didn't actually play into the results at all. The rest was still a mess, but when I started combing through it I discovered that even the live code didn't really matter to the results because at the end there was a "fudge factor" that the guy just set according to his gut feel about the market, and this was the main thing that determined the final result.
I didn't last long in that job.