[veeenu]

The assumptions underlying Brownian motion of prices have been disputed for quite a while now: the normality hypothesis can be rejected on most if not all historical financial returns series, as it turns out that most returns are actually fat tailed processes with very significant (and variable over time) correlations between distinct assets, which makes research around portfolio theory even harder to conduct.

[inter_netuser]

This is absolutely correct

If the markets were efficient and equivalent to random brownian motion, Jim Simons wouldn't be producing 50%+ returns for decades non-stop in HFT.

His net worth is 25 billion dollars, and there are other countless billionaires, made from the "efficient" markets.

That's how much this fallacy is worth.

[jcrben]

Depends on your timeframe right? Shorter time periods are a bit more random. But there's also clearly some autocorrelation which makes sense given the inflationary / deflationary expectations at play

[veeenu]

Yes, definitely. The big issue there is that intraday intervals vs daily closes vs monthly prices all require very different kinds of analyses and features as they target different scales of behavior. For example, you could exploit order book models for intraday which make little sense for longer time frames. In the same way, portfolio theory on intraday intervals tends to not hold as well as it does on longer timeframes.

[samvega_]

Which assets are that?

[veeenu]

Most if not all of them. Especially during a downturn, all assets of all classes tend to correlate and produce negative returns. During big crises, stocks that before seemed uncorrelated/anti-correlated have a tendency to increase their correlation and go down together; at the extreme, even bonds cease to act as a diversifier against stocks plummeting.