Market knee-jerks keep VAR models on their toes

With a return to volatility, increased backtesting exceptions show banks’ algos are stretched

For a moment there, it felt like the end of history in financial markets. Between the third quarters of 2020 and ’21, the only way was up – for equities – down for interest rates.

But history wasn’t done. It elbowed its way back into markets and the wider world with renewed muscularity. For the next three years, monetary dynamics meshed with geopolitics to produce, arguably, the most sustained volatility in trading since the eurozone sovereign debt crisis.

In this environment, value-at-risk engines faced their toughest challenge in the era of the internal models approach. And some were found wanting.

Between Q4 2021 and Q1 2023, US banks’ market risk disclosures – the most frequent and granular of their kind globally – show that among three dozen or so of their IMA users, the quarterly count of backtesting exceptions ranged between 10 and 30. A backtesting exception occurs when a trading day’s loss surpasses the maximum forecast by VAR.

Between Q2 2020 and Q3 2021, the quarterly count had never gone above seven.

Things seemingly normalised from Q2 2023 onwards, as the tally reduced again to single digits. But then in Q4 2024, the number shot up again to 15.

 

VAR overshoots have then – on the whole – become more frequent. But not for everyone in the sample.

Last year’s 15 Q4 breaches were recorded across 12 banks – an average of 1.4 backtesting exceptions per dealer. This figure is lower than even several pre-Covid quarters.

But in Q2 2022, 30 overshoots were reported by just 14 dealers – so, 2.3 per firm. And in Q2 2023, the rate shot up to 3.3 – 10 across just four firms.

Size matters

The small size of the sample – 34 US- and foreign-owned banks as of end-2024 – means the calculation is somewhat crude. But it does suggest that, where some gusts of volatility manage to confound models across numerous banks, others may catch only a restricted number of dealers off balance – raising the question of whether these models carry idiosyncratic deficiencies that need to be fixed.

Related

‘Spectacular’ vol disconnect ‘ominous’ for risk assets 

Who blew up gas prices? (It wasn’t just Russia) 

Fragile liquidity puts markets in the ‘danger zone’ 

Citi, JP Morgan incurred record VAR overshoots in Q4 

Sure enough, in Risk Quantum’s quarterly round-up of VAR backtesting performance, a few names have been recurring: Huntington Bancshares and HSBC North America have been repeat overshooters. But names like JP Morgan have also popped up more than once in the past year.

Of course, in the eyes of supervisors, the size of any miscalculation is a major factor. While overshoots at JP Morgan and Huntington may be the same proportionally, the sheer mass of a trading loss at the former poses higher risk of systemic reverberations. As of Q4, Huntington’s consolidated VAR averaged all of $586,000.

And this should all, by now, be moot. Under global regulatory plans, VAR-based capital requirement models were due for retirement, as the Fundamental Review of Trading Book – the market and counterparty credit risk component of the Basel III reforms – replaced them with a framework based on expected shortfalls. But, thanks to delays and outright dithering in implementing FRTB, the models’ useful lifetime has been forcibly prolonged.

With volatility once more throwing its weight around, dealers will need to use every VAR overshoot as a chance to refine and repair their algorithms.

Editing by Louise Marshall