14.12 Case studies

14.12.1LTCM, 1998: correlation failure in a levered portfolio (brief; see M03)#

Context Long-Term Capital Management ran highly levered "convergence" trades: long the cheap and short the rich version of related bonds and spreads, across many markets.

Timeline In 1998, after the Asian crisis (1997) and Russia's default (August 1998), the fund lost about US$4.6 bn in under four months. On 23 September 1998, 14 financial institutions agreed a recapitalisation of about US$3.6 bn (some sources: US$3.65 bn), brokered by the Federal Reserve Bank of New York. The fund was wound down by early 2000 (Wikipedia and CFA Institute summaries).

Mechanism The positions were diversified by market and instrument, so measured correlations in normal times were low. But they shared one exposure: each was short liquidity, paid to wait for spreads to converge. When liquidity was withdrawn, all spreads widened together. By 14.6, a book of ten unit-risk positions has standard deviation 4.4 at $\rho = 0.1$ and 9.1 at $\rho = 0.8$; leverage chosen at the former is twice as risky at the latter. Proposition 14.14 fails too: drawdown control by reducing exposure needs continuous paths and market liquidity to sell into, and LTCM's positions were large relative to their markets (M05). M03 covers the leverage and Kelly side: estimated edge, true edge and overbetting.

14.12.2A disposition-effect example#

What prospect theory predicts Relative to purchase price, A and B are gains, C and D losses. By Proposition 14.15 he sells winners: selling A (₹1,25,000) covers the need. Odean's measures on this day: realised gains 1 position, paper gains 1 (B), realised losses 0, paper losses 2. PGR = 1/2, PLR = 0/2.

What a rational process does The purchase price is irrelevant to the future. Ask, for each stock, whether you would buy it today at today's price; sell the one you would least want to own. Tax adds a direct incentive the other way. Suppose all four were held under 12 months. Selling A realises a ₹25,000 short-term gain taxed at 20% (₹5,000). Selling D realises a ₹30,000 short-term loss, which can be set off against other gains (it saves up to ₹6,000 if gains exist to offset, at the 20% STCG rate; rates as of FY 2026-27, verify). Selling D raises only ₹70,000, so the remaining ₹30,000 must come from another position. If the investor's views on A and D are equal, selling D and raising the remaining ₹30,000 from whichever other holding he would least want to own dominates selling A.

14.12.3Jhunjhunwala: separating the trading book from the investing book#

Context See 14.11 for the verified quotes and M06 · 6.12.1 for the record.

Analysis with this module's tools

1. Two return streams with different distributions. His trading (as he described it) was short-horizon with quick loss-taking: many draws of an R-multiple distribution, judged by expectancy. His investing was a handful of concentrated positions held for years: few draws, each with a wide value distribution, judged by thesis. Applying one set of rules to both breaks one of them: trading rules applied to Titan would have stopped him out in the drawdowns he said did not impair value; investing rules applied to trades become "hold the loser because I believe in it", which is the disposition effect.
2. The separation rule as a constraint on reference points. Refusing to move a losing trade into the investment book fixes each position's reference frame at entry. It is a pre-commitment device against Proposition 14.15.
3. Funding. Trading profits funded investments early on. In portfolio terms, the trading book's cash flows were reinvested into a long-horizon book with (he believed) higher expected return. That works only if the trading book has positive expectancy after costs, which, given SEBI's findings, is the exception.
4. Where the rules did not save him. DHFL was an investment that he added to while forensic red flags were public (M06). Separation of books protects against the disposition effect, not against an investing thesis that ignores evidence.