March 2023: I was long gamma into a banking crisis. I still lost money.

The setting

The spring of 2023 was not a normal quarter for anyone with European bank names on their book.

Silicon Valley Bank failed on 10 March, Signature Bank two days later, and First Republic was seized and sold in May. On 19 March, Credit Suisse — not technically a bankruptcy, but functionally the end of a 167-year-old institution — was absorbed by UBS over a weekend, with roughly CHF 16bn of AT1 capital written to zero while equity holders received something. That inversion of the capital structure did more damage to sentiment than the merger itself. Five days later, on 24 March, Deutsche Bank's CDS spiked and the stock fell sharply intraday on no news anyone could point to. In parallel, and largely unrelated in cause but not in effect, Nordic real estate was unravelling: SBB lost its investment grade rating a few weeks later and the sector was already being repriced.

For a structured products market maker, the interesting part is not the narrative. It is that a book which had been quietly carrying a particular shape of risk for two years suddenly had that shape tested.

What sits in a barrier book

A market maker in this space does not choose positions. The positions are the residual of what has been issued.

Two years of low volatility and yield-hungry retail demand had filled the book with capped bonus certificates and the barrier structures around them. Strip one down and the payoff is a delta-one leg plus a barrier put plus a short cap. Hedge the vanilla components and what remains — what you actually carry overnight — is barrier risk.

Barrier risk has an unpleasant property. Its greeks are functions of distance to the barrier, not of anything you decided. Far from the barrier, the position is dull. Approach it, and gamma and vega inflate together, because the entire value of the structure now hinges on whether a line gets touched, and both realized moves and implied volatility govern that probability.

So in a two-year calm market, the book is quiet. And then in three weeks in March 2023, half of it is suddenly close to a barrier.

What actually happened

On the Deutsche Bank position, three things happened simultaneously, and I read two of them correctly.

One: I was long gamma, and I traded it properly. As the stock fell, I bought stock. That is the mechanical consequence of long gamma and there is nothing clever about it — but through a fast, gapping, wide-spread tape, executing that discipline without hesitating is most of the job. That part worked.

Two: my vega grew as I approached the barrier, and implied volatility exploded. Three-month at-the-money implied volatility reached 50%. On a large European bank. My position was long vega and getting longer with every point the stock fell. On paper, the mark-to-market was excellent.

Three: I did not sell enough options. This is the one that mattered.

At 50 volatility on a three-month tenor, the listed market was offering to pay me, in cash, for exactly the exposure my book was accumulating for free as a byproduct of issuance. That is the entire point of running a hedging book against a structured products franchise: you are long a shape of risk you did not choose, and when the market pays an extraordinary price for that shape, you recycle it. I sold some. I should have sold far more.

I did not, and the reason was not analytical. It was that the position was making money, and there is a strong pull toward not interfering with a position that is making money in the middle of a crisis. Adding a short volatility leg while headlines were still deteriorating felt like standing in front of the move. Every risk framework I know describes that instinct correctly as the thing you are paid to override.

What the rest of the year cost

The crisis ended, as they usually do, without a resolution anyone could date.

Deutsche Bank did not break. It also did not do very much. The stock spent the rest of 2023 in a range, realizing volatility nowhere near the 50% that three-month implied had reached. And so the book paid theta — a genuinely punishing theta, because a long gamma position near a barrier struck at crisis-level implied volatility is expensive to carry.

Read the gamma P&L equation and it is all there:

PnL = ½ · ∫ Γ_t · S_t² · (σ²_realized,t − σ²_implied) · dt

I was long the left side of the parenthesis at 50. The market delivered maybe 20 to 25. Every day of that gap, weighted by a dollar gamma that was large precisely because I was near a barrier, was a payment out.

Then the second leg. Implied volatility mean-reverted through the summer and autumn, and the vega gain that had looked so good in March — the gain I had declined to crystallize — reversed. Not partially. Entirely, and then some, because as the stock drifted and time passed, the barrier proximity that had inflated my vega in the first place decayed, and with it the position's ability to benefit from any subsequent vol move.

The position closed the year with a five-figure loss. Nothing that threatens a franchise. But a loss on a trade where I had been right about the crisis, right about the volatility spike, and right about the hedging mechanics.

Four things I took from it

1. Being long convexity into a crisis is not a trade. Monetizing it is the trade. Long gamma and long vega going into March 2023 was a fortunate accident of what had been issued. The only decision that carried alpha was what to do at 50 volatility, and I made it too slowly and too small.

2. A vol spike is a mark-to-market event, not a P&L event, until you act. Vega gains have no natural mechanism for turning into cash. They convert only in two ways: by selling options into the bid, or by the underlying actually realizing the volatility you are marked at. Assuming the second is what I did, and it is not a decision so much as a hope.

3. Barrier proximity is the most path-dependent greek profile in the book. My dollar gamma and my vega were both large only because spot sat in a specific narrow region. Any drift away from it — even a favourable one — decayed the position. A correct volatility view expressed through barrier risk still needs the spot to cooperate on where it sits, not just on how much it moves.

4. The hardest short volatility trades to put on are the ones the market is begging you to do. Selling three-month options at 50 volatility on a bank, five days after a systemic institution had been resolved over a weekend, felt reckless. It was the single best risk-adjusted trade available to me that quarter. The discomfort was the signal, not the warning.

The general lesson is the one every gamma trader eventually learns by paying for it: predicting volatility correctly and being paid for it are two different problems, and the second one is harder.

Then price the other side of the trade: the variance-swap pricershows what selling that 50-vol was worth, and the delta-hedged gamma P&L behind the whole story is in the greeks reasoning primer.