How Wall Street Lied to Its Computers | Hacker News Reader

How Wall Street Lied to Its Computers

bits.blogs.nytimes.com

34 pointsrobg22 comments

Well, that explains that. Now explain this: If these MBSes were so complex that Wall Street MIT-trained quants and Wharton-trained traders and their computers didn't see it coming...how exactly did I see it coming? Me.

I mean...liar loans and ARMs starting at record-low rates and housing prices that looked like the NASDAQ right around AD 2000? I mean, the reason they didn't see it was...because they did see it, but the hand that picked up the commissions had a mind of its own.

In the short run, these markets selected for people who didn't properly analyze the risk. If the average CDO makes you 7% a year until one day it loses 20%, it's not a prudent investment -- but if you somehow estimate that the maximum loss is 2% instead, you will be willing to buy many more of them, and your return on capital will look a whole lot better. If you're not the only one doing this, the flow of all that money into risky products will give you returns even higher than you expected -- and if you're like the average investor, you will adjust your reasoning in retrospect to make yourself as brilliant as possible.

So it's actually unsurprising that people outside of the finance business were disproportionately aware that the whole real estate bubble was crazy. Someone with your views working for a big bank would be the equivalent of a peacock who realizes who impractical those feathers are.

Not a web connection or mag subscription among the lot of them, eh?

http://www.marketoracle.co.uk/index.php?name=News&file=a... http://www.forbes.com/2005/02/21/cz_0221oxan_default.html http://money.cnn.com/magazines/fortune/fortune_archive/2006/...

It's pretty easy to ignore what you read on the web when it doesn't fit your preconceptions.

Yes, and to everyone who now claims he was really sure of the bubble back then: why didn't you make a fortune in short selling?

Because even if you know (in general) what is going to happen, it is quite hard to time it. Plus I think you need balls of steel in certain situations (everyone telling you are wrong).

Shorting is generally a bad idea because even if you're right, if your timing is wrong, you can still lose everything. Say that you call "overvalued" and short the market. You're right, but the market gets more overvalued before it corrects. You can get short-squeezed out of your positions because with the new, bubble-inflated prices, you don't meet margin requirements. Then the market corrects, you're vindicated, but you've lost everything already.

Same goes for buying on margin when the market undershoots. You can be right, but if the price continues to drop past rationality (as it often does), you can lose everything.

And I did make a pretty good return, both in 2001 and 2007, simply by holding money in cash when people were greedy and investing when they were fearful. Not a fortune (I was a poor college student in 01/02, and only had 2 years of accumulated work savings in 07), but enough to fund my year-long startup adventure, and to fund a few more years of it if I had a decent idea.

Yes, timing is a problem. The market can stay irrational for longer than you can stay solvent.

Good luck with your ventures!

This is really rather simple to understand: Part of the motivation for inventing complex new derivatives is to create things that look conservative to the model but are actually risky.

The personal incentives for traders are to get big returns, which implies making risky bets. But your risk management system won't let you make risky bets directly. So instead, you make risky bets indirectly through instruments specifically engineered to game the risk management system.

It's nothing new. Michael Lewis' excellent book "Liar's Poker" describes how Solomon Brothers invented ways to obtain triple-A credit ratings for incredibly risky forex trades so that they could be sold to S&L's with strict rules about the credit-worthiness of their investments. And that was at least twenty years ago.

To paraphrase Orwell: "Trust a snake before you trust an anecdote, trust an anecdote before a statistic, but do not trust a model."

Part of the motivation for inventing complex new derivatives is to create things that look conservative to the model but are actually risky.

Nailed it. Things like FX, commodities and equities are closed systems. These markets make their money by volume and it's not particularly interesting.

Take something like a CDS where you don't really know what the risks are and it becomes easier to set prices fairly arbitrarily. Maybe some cpty has a stellar rating. Does that mean they'll always be stellar? Who knows?!

But this is still missing the real killer. Leverage. Everyone forgot the lesson of LTCM.

It was a bitter irony that the only bank to tell LTCM "fah q!" was the first to go. It's not so ironic now that Lehman, Merrill and possibly Morgan are gone now too.

I'm going to draw the obvious conclusion that most of the models were built on the premises of mediocristan, when in fact the world they were modeling lived in extremistan.

The thing is that quants who come up with models showing large amounts of risk to otherwise profitable investments aren't as popular as ones who say everything is just fine. So there is a clear incentive for models to be optimistic.

I thought this was a great analogy to express your conclusion:

"It was like a weather forecaster in Houston last weekend talking about the onset of Hurricane Ike by giving the average wind speed for the previous month."

The impression I got was not that they lied to their own computers, but rather those who programmed the alarms did so inadequately and could not handle the complexity of the transactions.

His doctor analogy is good, but not exactly perfect. Something more along the lines of hiring a med student for your physical instead of a grey-haired professional.

Not so.

"So some trading desks took the most arcane security, made of slices of mortgages, and entered it into the computer if it were a simple bond with a set interest rate and duration."

I find the lying to your doctor analogy very good to describe this kind of behavior.

The models should penalize investments that are complex, hard to understand and infrequently traded, she said.

...by how much?

It seems to me that the root cause to the computer models failing was that the government stepped in to "help."

Once government requirements were put in place, I can only imagine the level of red tape required to make changes to what was a dynamic system before.

None of this was mentioned in the article. Does anyone have any facts on this?

Nothing I have read indicates that was the case. I believe Raganwald above is mostly correct. The major roll that government played, is that by setting the precedent of "too big too fail" starting twenty years ago, and by setting interest rates too low, it enabled Wall St to play with far more risk than was wise.

It was precisely because regulation was removed that the current conditions were possible...

http://en.wikipedia.org/wiki/Glass-Steagall_Act#Repeal_of_th...

Do you mean capital adequacy requirements?

Because there is discussion that the biggest failures were where the SEC relaxed it's requirement. http://www.nysun.com/business/ex-sec-official-blames-agency-...

100 year flood. Brilliant, quants. Quick investment tip: Never, ever, invest in a prospectus that has the word "Gaussian" in it.

Traders and bankers knew, philosophically, that the risk in these products was much higher than the models would admit. "25 sigma", to use Goldman's excuse, means nothing when the distribution is not normal. For a 20-year-old to die (death being a 0/1 event) within a year is "30 sigma", and yet it is not uncommon for a college student to die.

My lay explanation of the risk-management failure is as follows. Let's say that you're of average means, with a net worth of $25,500. You've decided that "fuck you" money is $10 million, and you want to get there by (wait for it) betting on coin flips, pursuing the Martingale betting strategy. You'll stop flipping when either (1) you lose everything, or (2) you get to the fuck-you mark of $10m.

Martingale works as follows: start with a small bet (say, $100). If you win, bet again at the small size. If you lose, bet again, doubling your size. You'll win almost all of the time, losing only on an improbable string of losses. When you win, you'll be up exactly $100.

Obviously, this is an extremely stupid strategy. On the first go, you lose everything on a string of 8 failures (1/256). You win $100, 255/256 of the time. Expectancy is still zero, and although a blow-out loss is unlikely on a single round, you're going to progress to $10m so slowly that you'll almost certainly fail out beforehand. You have, roughly, a 0.255% chance of getting the "win" outcome of $10m. This doesn't improve if you change the size of the bet.

If you're able to borrow $1 billion, in addition to your meager $25.5k, this strategy makes perfect sense. Let's assume that the coin-flips are instantaneous, and interest is agreed-upon to be a flat 1%/$10m, meaning that you need to win $20 million to have your "fuck you" money. Now your blow-out probability is extremely low. The probability of getting to +20m before -1000m is about 98%. So, you have a very high chance of reaching your goal, and a low chance of losing your few-months'-salary bankroll (plus a lot of someone else's money). Of course, the billionaire is getting screwed.

This is a toy example, but it's not far off from what actually happens. Much of the money made in finance has been obtained by borrowing others' money to bet against "black swan" events, so infrequent that no one can accurately model their likelihood and impact. This is what "rock star traders" try to do their banks, and what banks try to do to their customers, and we've now seen the resulting clusterfuck.

Reply on news.ycombinator.com