Michael J. Mauboussin excerpts (PDF) from a presentation the Greenwich Roundtable:
LTCM used statistical arbitrage to “vacuum up nickels”, as one of the founders, Myron Scholes, described it. One essential component of their portfolio was that it was highly diversified: the correlations between the positions were historically quite low—10 percent or less. To be conservative, in their value at risk models LTCM assumed correlations could jump to 30 percent, vastly higher than anything they saw in their historical data. However, the summer of 1998 saw a real contagion—a diversity breakdown of epic proportions. Notwithstanding the substantial arbitrage opportunities, there were no arbitrageurs to be found, and correlations rocketed higher, to about 70 percent. Add high correlations, leverage, and declining asset prices, and you have the story.
Listen to Michael on the Trend Following podcast.