I’m not against the weak form of the EMH and I still think the EMH and all of the things that come with it are pretty useful(Black-Scholes, VaR, etc). You just have to be careful how it’s used (unlike the guys that ran LTCM).

The only problem that I have with it is the random walk part, but if you had modeled prices as a random walk with a stochastic volatility, that could give you correlations in the data(I’m actually working on a project where I’m trying to do this). So correlated data still might not be completely inconsistent with the EMH.

]]>Efficient markets doesn’t necessarily imply a random walk in prices. That is 1950s – 1970s finance. Modern finance takes into account that risk aversion changes over time.

If you were willing to buy in 2008-2009 when no one wanted to hold risk, you could expect a higher return. The price of risk itself changes over time.

That’s modern finance. It makes sense when you think about it.

Check out this thoroughly EMH consistent paper called “Predictability, The Dog that Didn’t Bark: A Defense of Return Predictability” by Cochrane

]]>https://editorialexpress.com/cgi-bin/conference/download.cgi?db_name=CEF2011&paper_id=128

It’s very difficult to express the extreme deviations from the data without having some sort of correlation between price movements. There are periods where H=.5, but overall; H does not have to be .5.

By the way, I can’t actually read that article. I don’t have a subscription. Let me see if I can do it from my University page.

]]>Here’s a paper which calculates the Hurst exponent over different time scales. What you’ll notice is that the Hurst exponent depends on the time scales.

http://www.princeton.edu/~sircar/Public/ARTICLES/bps.pdf

http://en.wikipedia.org/wiki/Hurst_Exponent

Here’s another paper that uses the Hurst exponent and compares the EMH to the Fractal Market Hypothesis in the case of the financial crisis.

http://arxiv.org/pdf/1203.4979.pdf

The Hurst Exponent is a very robust method of measuring correlations in the data. H=.5 means a random walk, H>.5 means the data is correlated, H.5 for most. H<.5 for something like volatility because it is antipersistent(volatility can't increase forever).

]]>