57
Financial Fallacies Igor Rivin (Temple U/Brown U/Meteque Holdings)

Talk at the quantopian Boston meetup

Embed Size (px)

DESCRIPTION

Slides from a finance talk I gave in Boston; the original is keynote: https://www.dropbox.com/sh/6j1swb2w6v490mw/Vlie9SCQ0a

Citation preview

Page 1: Talk at the quantopian Boston meetup

Financial FallaciesIgor Rivin (Temple U/Brown U/Meteque Holdings)

Page 2: Talk at the quantopian Boston meetup

Fallacy I

“A dollar is a dollar”

Page 3: Talk at the quantopian Boston meetup

A dollar is a dollar?

Page 4: Talk at the quantopian Boston meetup

A dollar is a dollar?

Depends to whom. The question was first analyzed by Daniel Bernoulli back in 1735, prompted by his cousin Nicholas Bernoulli’s St Peterburg game paradox.

Page 5: Talk at the quantopian Boston meetup

St Petersburg Game

I offer you a game: I flip a fair coin. If it lands heads, I pay you $1. If it lands tails, we flip again, and if it lands heads, I pay you $2. If it lands tails, flip again, and if it lands heads, I pay you $4, and so on.

Now, how big an entry fee will you pay to play this game?

Page 6: Talk at the quantopian Boston meetup

St Petersburg game

Easy to compute that the expectation is infinite!

Will you give me all your money?

Page 7: Talk at the quantopian Boston meetup

St Petersburg Game

Empirically, the answer is a resounding NO! And it is unlikely that anyone will pay more than $10 or so to play.

Why? Bernoulli’s thesis is that if Peter has $100000, and Paul has $1MM, then $10000 is worth about the same to Peter as $100000 is to Paul.

Page 8: Talk at the quantopian Boston meetup

A dollar is a dollar?

An example: a 10% per year return is viewed as a steady return. A $100000 per year return is strange.

Bernoulli’s example: insurance. If you insure your house for a bit more than the expectation of loss, good for you and the insurance company.

Page 9: Talk at the quantopian Boston meetup

A dollar is a dollar

A bit of analysis leads to the conclusion that the UTILITY of N dollars is proportional to log N.

Page 10: Talk at the quantopian Boston meetup

A dollar is a dollar

So, giving the guy money is good for society (that way be socialism…)

Page 11: Talk at the quantopian Boston meetup

St Petersburg game

And of course, the logarithmic expectation of the St Petersburg game is finite.

Page 12: Talk at the quantopian Boston meetup

Fallacy II

log (1+x) = x

prevalent in the finance community (because pension fund managers do not understand logs?)

Page 13: Talk at the quantopian Boston meetup

log(1+x) = x

leads to complete confusion, not the least manifestation of which is the Sharpe ratio.

And its cousins, the CAPM and Markowitz portfolio optimization.

Page 14: Talk at the quantopian Boston meetup

Sharpe Ratio

Defined as the mean of portfolio excess returns divided by the standard deviation of excess return.

Should be the mean of excess log returns divided by their variance.

(call this the KT-ratio, for Kelly-Thorp, more on this below)

Page 15: Talk at the quantopian Boston meetup

Three confused Nobelists

Harry Markowitz

Paul Samuelson Bill Sharpe

Page 16: Talk at the quantopian Boston meetup

And their intellectual offspring

Page 17: Talk at the quantopian Boston meetup

Nassim Taleb

There are two methods to consider in a risky strategy:

The first is to know all parameters about the future, and engage in optimized portfolio

construction, a lunacy, unless one has a god-like knowledge of the future. Let’s call it

Markowitz-style.

Page 18: Talk at the quantopian Boston meetup

What if you do know about logs?

Page 19: Talk at the quantopian Boston meetup

Optimal betting strategies

Problem: you have an edge on the house (say, counting cards, or doing statistical analysis of stock returns). How much should you bet?

Page 20: Talk at the quantopian Boston meetup

Optimal betting strategies

ANSWER: depends on what you want. If you only have one shot, you should bet all of your money (if you are really, really sure you have an edge).

Page 21: Talk at the quantopian Boston meetup

Optimal betting strategies

But if you can play as long as you want, betting all your money every time is a terrible idea, since you will LOSE all your money quickly, even if you have an edge.

Page 22: Talk at the quantopian Boston meetup

Optimal betting strategies

But, the answer has been figured out (at Bell Labs! — not as surprising as it seems, since the phone company made a lot of money from bookies back in the day).

Page 23: Talk at the quantopian Boston meetup

Optimal betting strategies

The answer is “the Kelly Criterion”

Page 24: Talk at the quantopian Boston meetup

The Kelly Criterion

Page 25: Talk at the quantopian Boston meetup

The Kelly Criterion

Interesting properties: betting LESS than the Kelly criterion reduces your returns, but also decreases the volatility

Betting more ALSO decreases returns, while INCREASING the volatility

Page 26: Talk at the quantopian Boston meetup

The Kelly Criterion($1000 initial bankroll, p=0.51, 2000

bets)($1000 initial bankroll, p=0.51, 2000

bets)

Page 27: Talk at the quantopian Boston meetup

The half-Kelly ($1000 initial bankroll, p=0.51, 2000 bets)

Page 28: Talk at the quantopian Boston meetup

The Kelly Criterion ($1000 initial bankroll, p=0.54, 2000 bets)

Page 29: Talk at the quantopian Boston meetup

The half-Kelly ($1000 initial bankrol, p=0.54, 2000 bets)

Page 30: Talk at the quantopian Boston meetup

Some Kellyists

Ed Thorp

Jim Simons

Ray Dalio

No Nobel prizes, and no regrets

Page 31: Talk at the quantopian Boston meetup

Back to fallacies

Page 32: Talk at the quantopian Boston meetup

Madoff Returns

Impossible! Or is it?

Page 33: Talk at the quantopian Boston meetup

Take a look:

See any similarity?

Page 34: Talk at the quantopian Boston meetup

The martingale strategy

You go to the casino, and bet $1 on red. If you win, you walk away.

If you lose, you bet $2 on red. If you win, you are up $1, and walk away.

If you lose, you bet $4, and so on.

Page 35: Talk at the quantopian Boston meetup

The martingale strategy

It’s amazing! You win $1 with 100% probability.

Well, except for running out of money (in a casino, there are table limits, but in the stock market, there aren’t…)

Page 36: Talk at the quantopian Boston meetup

The martingale strategy

So, we found a hedge fund — MG partners, which has $1000 under management to begin, and every day we run the martingale. What happens?

Page 37: Talk at the quantopian Boston meetup

The martingale for fair coin toss (2000 bets, $1000 starting bankroll), betting $1

every time.

Page 38: Talk at the quantopian Boston meetup

The martingale for fair coin toss, betting 0.05% every time ($1000 starting capital,

2000 bets)

Page 39: Talk at the quantopian Boston meetup

The martingale with an edge ($1000 starting capital, p=0.54, $1 bet, 2000

bets)

Page 40: Talk at the quantopian Boston meetup

Still looks a little variable…

The “standard trick” (or so people tell me…) is to found a few funds, and only tell people about the good one(s), just to be on the safe side.

As the amount of investment goes up, the risk of catastrophic failure goes down.

Page 41: Talk at the quantopian Boston meetup

When things crash

The investors are screwed, but the manager walks away with the 2/20.

Of course, this is not nice to the investors, so…

Page 42: Talk at the quantopian Boston meetup

We borrow money from the bank

At the end, the manager has done well, the investors have done well, the bank, umm…

Not so nice to the bank. But now there is a really simple solution.

Page 43: Talk at the quantopian Boston meetup

Instead of $1000, use $10Bn

Then, the manager does (very) well.

The investors do quite well

The bank earns nice interest income

And when things blow up, the bank is too big to fail; the Fed bails it out with money it prints, so

Page 44: Talk at the quantopian Boston meetup

Everybody wins!

Page 45: Talk at the quantopian Boston meetup

Quantopian exchange

Page 46: Talk at the quantopian Boston meetup

How many “trading systems” are the martingale in disguise? People like winning

every day…

Page 47: Talk at the quantopian Boston meetup

Of course, once the fund gets big enough

Hire enough physics PhDs to have SOME alpha.

Then inertia is your friend.

Page 48: Talk at the quantopian Boston meetup
Page 49: Talk at the quantopian Boston meetup
Page 50: Talk at the quantopian Boston meetup
Page 51: Talk at the quantopian Boston meetup
Page 52: Talk at the quantopian Boston meetup
Page 53: Talk at the quantopian Boston meetup

(that was another fallacy):

You need to have good performance to raise money.

Page 54: Talk at the quantopian Boston meetup

Another fallacy (on a smaller scale):

Ride your winners, dump your losers

Page 55: Talk at the quantopian Boston meetup

Ride your winners, dump your losers

Infinitely many backtests show that mean reversion is quite noticeable in the market.

Which means that this is exactly the wrong thing to do.

Page 56: Talk at the quantopian Boston meetup

Thank you for your attention!

Page 57: Talk at the quantopian Boston meetup

Thanks to Quantopian for organizing!