Risk limits and the loss your model said was impossible

Graduate · enrolled learners

This lesson opens with FXCM and the Swiss franc, 15 January 2015.

What happened

The Swiss National Bank had capped how strong the franc could get against the euro, at 1.20 francs per euro since September 2011, and on 15 January 2015 it abruptly abandoned that cap, so the franc rose by more than 30 percent against the euro in chaotic early trading, with almost no liquidity in between, before settling at a gain of about 15 percent on the day. FXCM's retail clients held leveraged foreign-exchange positions, and, on this lesson's reading of what followed, the broker's risk system was built on the assumption that a client's loss was bounded, that a stop-loss order would fill near its level and a margin call would close a position before the client's balance went below zero. In a move that large and that fast, with no liquidity to trade against, stops did not fill near their levels and positions blew straight through margin into deeply negative balances, leaving clients owing FXCM negative balances it first put at about 225 million dollars and later, in its filings, at 275.1 million. FXCM said it might be in breach of regulatory capital requirements and survived on a 300 million dollar emergency loan from Leucadia the next day. Nothing was hacked; the risk system had modeled a world where losses were bounded because the market always offered a price to exit at, and the SNB removed the floor from under exactly that assumption.

The decision point

A risk system encodes assumptions about what can happen, and when it is built on the premise that a loss is bounded, that a stop will fill near its level, that a position can always be closed before the balance goes negative, it fails precisely when a gap or a liquidity vacuum makes the bounded loss unbounded. FXCM is the case: its risk system assumed clients could not go meaningfully negative because stops and margin calls would close positions in time, and the Swiss franc's 30 percent gap blew straight through that assumption, leaving clients owing about 275 million dollars and nearly destroying the firm. The lesson is that the losses a risk system most needs to survive are the ones it assumes cannot happen, because those are the tails that arrive with no liquidity, where a stop-loss is only a request that may go unfilled and a market move can be larger than any move in the data. So risk limits cannot be built only for normal conditions; they must account for gaps, where the price jumps with nothing to trade in between, for slippage, where an exit fills far from its intended level, and for the tail, where a move exceeds anything the model has seen. A risk system that treats a stop as a guarantee and a loss as bounded is a system that will be solvent right up until the day the market gaps, and then discover it was never protected at all. So the discipline is to design risk assuming the exit may not be there when it is needed: size positions to survive a gap, treat stops as best-effort rather than guaranteed, and stress the system against moves larger than history, because FXCM is what happens when the loss the model called impossible turns out to be merely rare.

Recorded loss

$275,100,000

Orientation and Year One are open: anyone can read them without an account. From Year Two onward the lessons are for enrolled learners, because progress through the later years only means anything if it is tracked against a record.

It is free. We do not sell the list and there is nothing to buy at the end of it.