undefined | Better HN

0 pointsjustin_vanw15y ago0 comments

I wouldn't claim that the market generally is efficient, but I strongly suspect that it a lot like Brownian Motion. No observer could possibly predict which direction it will jump next, but the overall parameters are determined in some fundamental way. I call this Brownian because to someone observing the motion of a single atom, it appears random, but it is entirely determined by the motion and position of other atoms that end up colliding with it, and there are fundamental conditions, such as the overall temperature or density of the stuff surrounding the atom being observed, but while knowing those things lets you make predictions, those predictions only hold after averaging over a very large number of atoms or a very long amount of time, and by then, if you aren't in a carefully controlled laboratory environment, the global conditions will have changed enough that your predictions aren't sensible any longer.

However, it's not efficient, since there is a lot you can do to make a better guess than assuming it is a completely random process.

While this will make it more efficient, there are plenty of people making random guesses, which reduces the effect your better guesses have on the overall efficiency of the market.

There are also plenty of fools, and they are investing on all sorts of techniques which ignore any of the global considerations, and of course this is how bubbles expand.

0 comments

4 comments · 1 top-level

dxbydt15y ago· 3 in thread

" I strongly suspect that it a lot like Brownian Motion"

1.Mkt isn't like BM. BM permits -ve values & any market index say SPY must be non-negative.

2. We (quants) model market indices like SPY as a log-normal process like so : dS = rSdt + sigmaSdz where S = underlying ( or market index ) dS = change in S, dt = change in time r = risk free interest rate ( about 0.03 % currently ) dz = normally distributed rv with mean 0, std dev sigma

There is an embedded Brownian component in dz, so also in S. But that doesn't make S itself Brownian. More here: http://en.wikipedia.org/wiki/Wiener_process

"No observer could possibly predict which direction it will jump next"

Nobody is trying to do that in the very next instant. OTOH, you can certainly predict where the market will go, say, in a year. Well known econ result: DP ratio regresses positively with return. ie. Dividend-price ratio predicts stock return with high statistical significance. Source: Cochrance, French/Fama, et al. http://dash.harvard.edu/bitstream/handle/1/3122601/campbells...

"there are plenty of people making random guesses...there are also plenty of fools, and they are investing on all sorts of techniques which ignore any of the global considerations, and of course this is how bubbles expand."

Ha ha ha! Very amusing but very inaccurate. I think this is my general problem with geeks on the outside of finance. They happen to think markets are purely stochastic. X has randomness doesn't mean X is unpredictable. Otherwise why have APT, CAPM, covariance matrices, factor models, Heath Jarrow, Black Litterman...you've written off some 100+ years of economics in one breath !

To summarize, there are tons of observers, who predict where the overall market or some portfolio of equities ( or bonds or agencies or munis or treasuries or whatever else ) will go. To get data points for these predictions, you use a historical time window, compute coVar matrices, invert them, build a CAPM....etc. Anyways, its not rocket science. There are tons of CFAs & Finmath engg ( I'm one of them ) who do this sort of work day in and day out. Its our bread & butter. If I said something like "market is a lot like brownian motion & people are making random guesses", I'd be out of a job.

pash15y ago

> 1.Mkt isn't like BM. BM permits -ve values & any market index say SPY must be non-negative.

That's one reason everybody models log-returns, which map non-negative prices onto the whole real line.

> 2. We (quants) model market indices like SPY as a log-normal process like so : dS = rSdt + sigmaSdz ... But that doesn't make S itself Brownian.

That's a geometric Brownian motion with drift. It is in fact the model developed by Louis Bachelier in his thesis of 1900 that presented the first mathematical description of Brownian motion.

Much of the argument you're trying to make can be made better, in my opinion, simply by pointing people to the Grossman-Stiglitz paradox [1]: markets cannot be fully informationally efficient or no one would have the incentive to provide information to them through trading.

1 - [PDF] http://www.math.ku.dk/kurser/2003-1/invfin/GrossmanStiglitz....

justin_vanwOP15y ago

"Otherwise why have APT, CAPM, covariance matrices, factor models..."

So are you saying you consistently outperform the market? If not, what are you 'modeling'? In 'science', a model is considered useful in how well it predicts the observations of some experiment, especially experiments on effects which were not known when the model was created. What do your bragged about models predict? What are they good for besides talking about with other quants?

meric15y ago

The CAPM tells you how "risky" a company is depending on its historical market returns; E.g. A manufacturing company is likely more risky than a retailer because the former's previous market returns vary more than the latter's.

j / k navigate · click thread line to collapse