Here’s just one famous LTCM strategy: The algorithm identified situations where the spread had widened between the yield on the most recently issued 30-year U.S. Treasury bond and that of an “off-the-run” bond, such as one issued a year earlier (essentially, a 29-year bond). Historically, investors pay a modest liquidity premium for “on-the-run” bonds (which means their yields are slightly lower than the off-the-runs) but, given that both investments represented the same “risk-free” exposure to the U.S. government, the idea was that the spread between the two shouldn’t deviate too far for too long. So, at times when the liquidity premium rose and the yield spread widened, the LTCM algorithm would place a short position (sell) against the on-the-run bond and a long position (buy) on the off-the-run. Historical mean reversion would do the rest.
