During Trump turbulence, value-at-risk may go pop

Trading risk models have been trained in quiet markets, and volatility is now looming

A smooth sea never made a skilled sailor. Franklin D. Roosevelt’s wisdom that adversity forms character could also apply to the models that Europe’s banks use for assessing their trading exposures.

Trained on the calm waves of last year’s markets, today’s market risk models may be unprepared for the potential storm to come during Donald Trump’s second term as US president, bankers fear. The expected surge in volatility could affect a range of asset classes, from equities and rates, through foreign exchange and commodities.

“In 2023 there was large volatility in commodities that came down in 2024,” says a senior trader at a European bank. “The expectation is volatility under Trump is going to be higher, so more volatility on the levels of 2023.”

There are multiple repercussions for banks trying to navigate with faulty risk models through a market that is much changed from the previous year.

For one, inaccurate loss estimates could cause the model to fail a key backtest that is designed to check the validity of the model’s outputs.

21.28

US dollar/Mexican peso exchange rate on February 3, before reaching 20.31 later the same day

“If volatility suddenly blows up because of Trump, then I would expect there will be a lot of backtesting failings down the Street,” says a senior risk modeller at a second European bank.

These failings could result in punitive capital add-ons for trading desks. Depending on the size of the bank’s trading book, the add-ons could amount to tens or hundreds of millions of euros in extra capital requirements.

Traders may also have risk limits that are set too low for current market conditions. This raises a danger of desks breaching these limits and having to rapidly scale back exposures as a consequence.

Delays to the adoption of a new trading book regime in the European Union mean banks currently calculate their market risk capital requirements using the old framework devised by the Basel Committee on Banking Supervision. Under the Basel 2.5 rules, banks opting for the internal models approach (IMA) must use two value-at-risk measures to estimate maximum one-day losses, which are then converted to capital requirements.

One of the VAR measures estimates losses based on observations of market moves from the past year. The other is calibrated on the worst losses that could be inflicted on a bank’s current portfolio from market moves taking place in a one-year period since 2008. The two measures make up the majority of internally modelled capital requirements for market risk.

Three multipliers are added on to both VAR and the stressed version of VAR, or SVAR. A base multiplier is set at three, which then increases with the addition of the other two variable multipliers.

A quantitative multiplier gradually increases the more times losses in a single trading day exceed the results of VAR over the most recent 250 trading days – essentially a calendar year. Breaches are totted up at the end of each calendar quarter and reported in the bank’s quarterly results.

Any time risk goes from low to high, you have limits that are underestimated, because they were estimated last year and they apply this year

Carlo Acerbi, Risknowledge

The backtesting multiplier add-on ranges from zero for banks scoring fewer than five exceptions and increases to a maximum of one for banks finding 10 or more breaches. For example, banks with five breaches have to increase their multiplier by 0.40, which would effectively mean an increase of more than 10% in market risk-weighted assets (RWAs) generated by VAR and SVAR. The bank’s minimum capital ratio is calculated as a percentage of RWAs.

The final qualitative multiplier is based on the supervisor’s assessment of a model for faults not picked up by backtesting.

Four sources fear the benign market conditions of last year will prove irrelevant for forecasting the year ahead, leaving banks vulnerable to VAR breaches and negative supervisory assessments.

“VAR is low now these days, which is not good because it means you can get backtesting exceptions quite easily, and this increases your RWAs,” says a risk modeller at a third European bank. “Any small event in the market might cause a backtesting exception.”

Hoist the colours!

The concern is whether markets will gyrate more as traders react to implications of actions taken or comments made by Trump, who became the forty-seventh US president on January 20. He has already proven a controversial figure, with comments on the territorial integrity of Canada, Greenland and the Panama Canal.

His threats of waging economic war by imposing tariffs on imports to the US is making markets jittery. So far, the US president has threatened Canada, China, Colombia, Denmark, the European Union, Mexico and Panama with tariffs.

The Mexican peso fell by more than 4% against the US dollar between Thursday, January 30 and Monday, February 3, after Trump vowed to follow through on setting tariffs at 25% on imports from the southern neighbour. But the announcement of a month-long pause of the policy later on Monday caused the exchange rate to swerve back to its original level. The US dollar/Canadian dollar exchange rate performed a similar U-turn after Trump also backtracked on Canada tariffs.

European markets are expected to react once more information about his plans for tariffs become known. During Trump’s first term in office (2017–2021), the US imposed tariffs of between 10% and 25% on goods from the EU, which were later relaxed under the Biden administration.

Commodity and foreign exchange markets in particular are expected to become more volatile. Commodities are imported and exported across borders, which makes them a ripe target to be hit by tariffs. Foreign exchange rates can fluctuate significantly from tariffs as demand to buy products in foreign currencies hit by tariffs weakens, whilst demand for US-denominated products from homebred companies strengthens.

Equity markets have proven jittery already, with one of the largest moves of the calendar year so far not being Trump-induced. Following news of China’s DeepSeek developing similar artificial intelligence capabilities as US tech companies – but with less computing power – stock prices for US technology companies plummeted.

The dramatic shift in US policy towards the rest of the world is leading risk managers to fear VAR will be unable to accurately predict maximum losses until it is able to learn from increasing volatility. The danger is that desks will rack up backtesting exceptions before VAR models have adapted to current market conditions. Traders may also burst through limits calibrated to benign markets.

“Any time risk goes from low to high, you have limits that are underestimated, because they were estimated last year and they apply this year,” says Carlo Acerbi, mathematical finance academic, and founder of consultancy Risknowledge.

Some European banks have relatively low VARs, according to Risk Quantum’s database. But tracking the level of VAR isn’t the only – nor the perfect – way of gauging volatility accounted for in VAR given it can be influenced by the composition of the bank’s book. VAR may rise if the bank’s trading book is growing, or traders are allowed to take on greater risks. A further way to indicate the level of volatility embedded in VAR is by the number of exceptions clocked on recent backtests, which are currently low across the board.

No battening down the hatches

Some argue that to penalise VAR models in backtesting for not predicting a change in market dynamics is unreasonable, as they aren’t designed to do so. Fair reasons to fail a VAR model would be if it had missing risk factors or inaccurate data, for example.

“A model does what it can do without crystal balls, and what models can do is look at the past,” says Acerbi. “If we all understand that a model will fail this year because simply the world has changed, that is clearly proof that it’s not the model’s fault. The very idea of rejecting models, in this circumstance, should be rethought.”

The EU’s version of the old IMA isn’t forgiving of such failings though. VAR is backtested against two views of profit and loss. The first is the value of a bank’s portfolio at the end of a trading day; the second is the bank’s portfolio from the end of the previous trading day as measured by using market data from the end of the current trading day.

The multiplier is determined by the highest number of exceptions caused by either two views, respectively, known as actual and hypothetical P&L. The EU’s Capital Requirements Regulation allows competent authorities to discount actual P&L from backtesting, if some of the actual P&L exceptions were not caused by model deficiencies. But it does not allow the exclusion of the hypothetical P&L that would also reflect market gyrations.

EU legislators recognised the unfairness in 2020, after the initial spread of Covid-19 across Europe rocked markets. Legislators quickly passed amendments to the CRR in 2020 allowing supervisors to discount overshooting in backtesting provided it wasn’t a result of deficiencies in the model. The solution was temporary, however, with supervisors only permitted to discount exceptions occurring between January 2020 and December 2021.

President Trump has paused tariffs on Mexico after the country agreed to bolster border security

There are ways for banks to speed up the pace at which VAR learns the market’s new conditions. Risk modellers could place more emphasis on observations in more recent data, known as exponential weightings. A further approach is multiplying historical risk estimates by a scaling factor determined by current market conditions. These aren’t perfect solutions though.

“There is a double-edged sword with using exponential weights,” says Acerbi. “The distribution is effectively based on fewer days and the statistical error with these models ends up being higher. So yes, they are more reactive, but they have more statistical error in their predictions.”

There will also still be a lag, albeit a shorter one, until VAR is able to retrain itself on current market conditions. As a result, volatility spikes would still cause backtesting exceptions until the models can be adjusted.

Similarly, using these model adjustments would not solve the problem of traders breaching tight risk limits when volatility strikes, and before risk managers have had an opportunity to review the limits.

The danger in such circumstances is that traders might be forced to sell positions when markets are moving against them.

“You will actually force the trading organisation to sell out risk at the worst possible moment,” says the senior risk modeller at the second European bank.

FRTB travels on a slow wind

In the upcoming market risk rules, known as the Fundamental Review of the Trading Book, banks using the IMA will still need to carry out VAR backtesting. However, one of the key features of the regime is the introduction of a new methodology known as expected shortfall. This works differently from traditional VAR as it better captures tail risk losses beyond VAR’s maximum loss estimates.

Expected shortfall averages all the returns in the distribution that are worse than the VAR of the portfolio at a given confidence level. FRTB sets the confidence level at 97.5%, which means expected shortfall is the average of returns in the worst 2.5% of cases.

At the time FRTB was being drafted, studies had concluded that expected shortfall lacked a mathematical property known as elicitability, which allows statisticians to develop a consistent scoring system for the reliability of the risk measure. As a result, regulators instead fell back on VAR backtesting as a way of indirectly testing the quality of the inputs into a bank’s expected shortfall.

The EU’s rules will allow supervisors to overlook exceptions within the FRTB’s backtesting of VAR. In any case, only three banks – BNP Paribas, Deutsche Bank and Intesa Sanpaolo – are set to run the regulatory test, as they are the only banks seeking to adopt the FRTB’s IMA. The rest of the EU’s banking population will use the one-size-fits-all standardised approach due to it being cheaper to run. The standardised approach doesn’t have a backtest as its risk weights are fixed.

Related

Fortunes of VAR: dealers decry effect of war on risk models 

European banks with large Russian derivatives exposures face risk of backtesting exceptions – and higher capital requirements

Acerbi, however, proposes an alternative to VAR backtesting that doesn’t need supervisors to expressly waive false negatives. Along with fellow researcher Balazs Szekely, the two have developed a way of backtesting expected shortfall, which was published on Risk.net in 2019.

Their expected shortfall backtest compares the realised value of a test variable – a function of the 250-day recorded sequence of expected shortfall, VAR predictions and actual losses – against the distribution predicted by the expected shortfall model for the same test variable. It represents a direct comparison between ex-ante risk predictions and ex-post risk observations – a way to measure discrepancies in expected shortfall predictions.

The authors say an advantage of their backtesting method is that it doesn’t just indicate whether the model is off but also puts a figure on the real risk number, which could be used to determine a capital add-on while the bank fixes the model. By contrast, the results of VAR backtesting do not indicate what the correct value of VAR or expected shortfall should be, says Acerbi.

“As opposed to the VAR backtest, the expected shortfall backtest gives you an estimate of the risk prediction discrepancy, and allows you to correct the model,” says Acerbi. “The VAR backtest only tells [whether] your model is wrong [and] has to be abandoned and replaced.”

Sailing beyond the limit

It’s not just risk managers that are nervy about VAR models being miscalibrated. Risk managers often also use VAR to set limits for the amount of risk individual traders and trading desks can take on day-to-day. The senior trader at the European bank highlights two risky scenarios. First, a bank may opt to lower a desk’s risk limits if the traders aren’t reaching their full headroom: in other words, use it or lose it. But when volatility rises, the desk is in danger of exceeding its limits. Second, a desk may have increased its trading to make full use of its VAR limit whilst markets were benign. But the traders also risk blowing through their limits unless they can unwind their positions.

What happens after a trader breaches their limit depends on the bank’s policies and the risk manager’s assessment of the situation. Risk managers could tell the trader to immediately unwind their positions to fall back under the limit. If they fear that rapid risk reduction would inflict unacceptable losses, they could decide to give a deadline for when the trader has to come back under the limit. Senior management is also made aware of breaches.

“I’ve seen different schemes in different companies, but it’s a matter of responsibility,” says Acerbi. “If you’re [a trader] beyond the limits, there is a big loss, and if nobody warned you, the blame is on the risk management. So, the risk management has to impose upon you to stay within the limits immediately, because otherwise they would take the responsibility of the potential loss.”

Editing by Philip Alexander and Alex Krohn