Settlement latency is a risk input, not an operations problem

The Quantitative Strategies division runs positions in markets whose settlement conventions differ by two full business days. That variation is treated in most institutions as a matter for operations, reported as a fail rate and forgotten. We treat it as an input to sizing. Cash committed to a trade that has not settled cannot fund another trade, and cash borrowed to bridge that gap carries the funding rate of the day. In a book that turns over four times a year, the aggregate of those bridges is large enough to change which strategies are worth running.

The measurement is straightforward and rarely done. For every market the desk trades, we record the median interval from execution to good value and the ninety-fifth percentile of that interval, measured on our own trades rather than on the published convention. The two numbers diverge more than expected. In the three thinnest markets the desk uses, the published convention is two days and our observed ninety-fifth percentile is five. That tail is where funding cost lives, because the tail events cluster around month ends and around the days a strategy most wants to trade.

Once measured, latency enters the return calculation as a drag rather than as a footnote. Applying the observed intervals and the current cost of short-term funding, the drag on the strategies with the widest geographic spread is 84 basis points a year. The same calculation two years ago produced 19 basis points, and nothing about the strategies changed. The cost of money changed. A strategy earning a gross 11 per cent is not much affected. A market-neutral strategy earning a gross 4 per cent has lost a fifth of its return to a variable its authors never modelled.

Time zones compound the arithmetic. A position executed in an Asia-Pacific market and booked into a Caribbean entity crosses a value date boundary, and the boundary adds a day of funding whenever the trade falls late in the local session. The division measured this across a full year and found 61 trades where the booking route rather than the market convention added the day. Changing the booking entity for those flows removed the cost without any change to execution. The point is not that the fix was clever. The point is that nobody had priced the routing until it was measured.

The wrong response to this analysis is to spend on faster technology. Latency between execution and good value is set by market infrastructure and by the operating hours of correspondents, and no internal system compresses it. The right response is to size positions in slower markets smaller, to hold a working cash buffer calibrated to the ninety-fifth percentile rather than to the median, and to decline strategies whose gross return does not clear the measured drag by a stated margin. Two strategies were declined this year on that test alone.

This is what proportionate quantitative work looks like in a firm of this size. The models that matter are not the ones that predict prices. They are the ones that describe the plumbing between a decision and the cash, because the plumbing is where a mid-sized institution operating across many jurisdictions either keeps its edge or gives it away. The division reports the drag figure to the Risk and Valuation Committee each quarter beside the performance numbers, in the same table rather than in an appendix.

Published 2022-10-18 by the Quantitative Strategies division. Research is prepared for eligible counterparties and does not constitute advice.