Stress, buffers, and the price of scale

The portfolio simulation answers a medium-horizon question about return and downside; it does not answer whether a participant can obtain foreign currency during a market slump. This chapter tests the Synthetic Reserve Deposit (SRD) against that harder problem. It compares the first-tier cash buffer with collateralized borrowing, then follows participation caps from country-level coverage to market depth and possible global-financial-cycle effects. The useful judgment is not that one liquidity tool dominates, but that each tool absorbs a different part of the stress sequence.

What the buffer and collateral route can cover

The first line of defense is the settlement or liquidity buffer inside the pool. In the IMF paper’s liability-side counterfactual, the buffer is set at 10 percent of total SRD assets. The exercise starts in 2007, treats observed quarterly increases in countries’ reserves as deposits and declines as withdrawals, and asks whether the pooled balance sheet could have met the resulting use of the accounts. It is a deliberately partial test: asset returns and reserve flows are taken from history, and the sample excludes the United States, the euro area, and China.

The paper compares three participation rules. In a full-transfer benchmark, countries place all of their starting reserves into the pool. In an initial-cap design, the first deposit is limited but later account growth can continue. In an all-dates account-cap design, each account remains subject to a continuing ceiling. These rules matter because the buffer is a percentage of the pool: a larger pool produces more first-tier cash, but also creates a larger common balance sheet and potentially more concentration. A tight cap contains the pool and its market footprint, yet leaves less pooled liquidity available relative to the withdrawals that participants may make.

Under the two capped designs, the simulated 10 percent buffer is reported as large enough to cover the observed withdrawals over the sample, including the global financial crisis period. That result supports a specific claim: pooling can make the first-tier buffer large relative to the realized, notional use of participating accounts. It does not support the stronger claim that 10 percent is enough for every crisis. The historical flows are one path through the system, not a stress test in which every participant demands cash simultaneously at the same moment.

The paper pushes the test beyond observed withdrawals by constructing sudden-stop episodes from historical gross-inflow data. The 10 percent ratio provides substantial coverage for events comparable to the global financial crisis in the calibration, but coverage falls when the shock is scaled up. This is the correct way to read the result. A buffer can be judged reasonable against a specified set of historical stress magnitudes, but no finite buffer can cover an arbitrarily large synchronized withdrawal without another source of cash or a sale of assets.

Collateralized borrowing addresses the gap between a temporary valuation loss and a permanent need to exit. In the paper’s 30-day simulation, a participant pledges SRD units and receives an advance below the collateral’s market value. A haircut creates the initial cushion. Daily valuation then checks whether the remaining collateral still covers the advance. If coverage weakens, the participant can add eligible collateral, repay part of the loan, or reduce or substitute the pledged position.