Why Banks Sold the Bonds They Could Sell Without Showing the Loss

This chapter examines the Philadelphia Fed’s most direct evidence about U.S. bank trading during the 2022–23 rate surge: were banks less willing to sell a bond when its market price was below its accounting value? The answer matters because it turns an institutional explanation into an observable behavior. Readers can use the underwater-versus-above-water pattern to distinguish a reluctance to realize losses from a general lack of liquidity, while keeping clear that a strong association does not reveal every manager’s private motive.

The sale pattern was sharply asymmetric

The relevant comparison is between a security’s market value and its amortized cost. The Philadelphia Fed paper calls a security “underwater” when its market value is below 99% of amortized cost, treats a value between 99% and 101% as around par, and calls a security above 101% above-water. The analysis focuses on AFS portfolios because HTM securities are subject to restrictions on sale. It then asks whether the fraction of a position sold in the following quarter varies with that market-value-to-amortized-cost ratio.

The raw pattern is striking. During 2022–23, the quarterly sale propensity was about 1% for underwater securities and about 9% for above-water securities. The difference is not a small change around an already active market. It means that a bond trading at a premium to its accounting value was several times more likely to be sold than a bond trading below that value, even though the latter was often the bond whose sale would have done more to reduce duration or raise cash.

The logit estimates express the same result as odds ratios. In the preferred specification, the odds of a full sale were about 3.2 times higher for securities around par than for underwater securities, and about 8.5 times higher for above-water securities. These are multiplicative effects on the odds of sale, not claims that 8.5% of the bonds were sold. The underlying probability for the underwater group was still below 1% per quarter in the preferred model. Keeping the denominator and the metric in view prevents the result from being exaggerated while preserving its economic significance.

The design also makes the comparison more informative than a simple chart of total bank sales. The regressions can include bank-by-time fixed effects, which compare securities held by the same bank at the same time. That holds constant the bank’s overall appetite or need to sell in that quarter. They can also control for security type, floating-rate status, time since purchase, duration, convexity, and remaining maturity. In the most demanding specification, the researchers compare the same bond across banks and the same quarter, where different purchase prices or hedging arrangements can leave the bond with different market-value-to-amortized-cost ratios.

Those comparisons address several plausible alternative explanations. Perhaps underwater securities were simply less liquid. But the main portfolios consisted largely of Treasuries and agency MBS, both highly liquid fixed-income instruments, and the relationship survives security controls and a Treasury-only robustness check.