Bank of England researchers model how a Big Tech AI earnings shock would ripple through global markets

Started by Nina26, Aug 14, 2026, 10:53 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: Bank of England researchers model how a Big Tech AI earnings shock would ripple through global markets   Views(Read 150 times)

Nina26

A new Bank Underground post from Bank of England staff researchers examines what would actually happen to global financial markets if AI related earnings from the Magnificent Seven US tech companies badly disappointed, building a systematic earnings shock model rather than just speculating about an AI bubble bursting

The researchers constructed a daily shock series tracking how Magnificent Seven stock prices move in the narrow window around their earnings announcements, weighted by each company's share of the S&P 500, which has grown from just 3 percent to 35 percent of the index over the sample period, meaning these seven companies now carry outsized influence over the whole market's reaction to their news

The findings show the effects do not stay contained to tech stocks, a negative 1 percent Magnificent Seven earnings shock produces close to a 2 percent decline in the S&P 500 itself, and within two days the FTSE 100 in the UK falls about 1 percent too despite having no direct mechanical exposure to those specific companies' earnings

Perhaps the most notable finding is what happens to the US dollar, rather than seeing the usual flight to safety into US assets during market stress, the researchers found the dollar actually depreciates following bad Big Tech earnings news, while credit spreads widen in both the US and UK, a pattern the paper argues is more consistent with investors reading the news as a downward revision to expected future US productivity growth rather than a typical risk off panic

Government bond yields told a more muted story, both US and UK 10 year yields dip only slightly and the rebound afterward is not statistically significant, which the researchers link to two offsetting forces, weaker growth expectations pulling yields down against a countervailing view from other recent research that AI progress itself is tied to US fiscal sustainability, effectively meaning bondholders are already positioned as implicitly long on AI

The policy implication the authors draw is fairly direct, an AI earnings correction would not stay confined to Silicon Valley or even to US markets, it would spill over into UK and other foreign equity and credit markets while potentially weakening the dollar rather than strengthening it, which is the opposite of what typically cushions foreign economies during past financial stress episodes, and that combination is exactly the kind of scenario the Bank's own Financial Stability Report has been trying to model

Always open to a good discussion

Mark7

The dollar depreciating instead of strengthening on bad tech news is the part that really stands out here, thats a genuine departure from classic flight to safety behaviour

BetaMyles75

Makes sense once you think about it as a productivity growth story rather than a pure risk event though, different shock means different market reaction

Mia86

35 percent of the S&P 500 concentrated in seven companies is honestly the real underlying vulnerability this whole paper is describing

Pirlo

Fair, though they are pretty upfront about that being an open question rather than overselling a clean finding

StarKnight36

Interesting that UK markets get hit almost as hard as US ones despite having zero direct exposure to these specific companies earnings

Molly28

Worth remembering this models a specific type of AI disappointment scenario, real world triggers could behave quite differently depending on what actually causes the correction

Josh93

Agreed, better to see this kind of stress testing done in the open than only behind closed doors at policy meetings

Client Wrench

Central bank researchers publishing this kind of scenario modelling in public is genuinely useful, more transparency into how they think about tail risks

Binary Anchor

The bond yield finding feels like the weakest part of the analysis honestly, muted and not statistically significant isnt exactly a strong result

FluxKnight80

Financial contagion rarely respects borders, thats been true for decades and clearly still holds even for something as US centric as Big Tech AI valuations

DudleyBoy

Agreed, that concentration alone means any bad news from those companies is basically systemic risk for the entire index at this point

Luca99

The global angle is easy to underestimate. A US technology earnings shock does not stay neatly inside the US equity market because the investors holding those assets are global and the companies themselves sit inside international supply chains.

A large semiconductor manufacturer, for example, might have customers in several countries, suppliers in several more, and shareholders scattered across pension funds, asset managers and sovereign institutions. A sharp revision to expected AI demand could therefore hit earnings expectations in multiple markets simultaneously.

Currency markets could add another layer. If investors move aggressively toward safer assets, exchange rates can shift, which changes the value of overseas holdings and can affect companies that earn revenue in multiple currencies. Suddenly an equity story has become a macroeconomic story.

That does not mean every country gets hit equally. Economies with relatively small technology sectors but strong links to AI-related manufacturing could experience a very different shock from economies where technology stocks dominate household portfolios.

This is why scenario analysis is more useful than trying to predict one precise outcome. The interesting question is not simply, "What happens if AI earnings disappoint?" It is, "Through which channels does that disappointment travel, and where does it become amplified?"

Caitlin89

There is a behavioural side to this that models may struggle to capture. Markets do not always react proportionally to new information. If investors have spent years hearing that AI will transform productivity, a disappointing earnings report can challenge the narrative itself rather than just one company's forecast

That distinction can produce a larger reaction. Investors may suddenly ask whether the whole investment cycle has been priced too aggressively. Once enough people ask that question at the same time, correlations can jump and selling becomes broader

On the other hand, markets can also overreact in the opposite direction. A temporary earnings disappointment might cause a dramatic sell-off even though the underlying technology remains economically important. Six months later, investors could decide they were too pessimistic and reverse course

That makes policy difficult. Central banks cannot simply prevent markets from repricing risk, nor should they. Their job is more about making sure the financial system can withstand the repricing without losing its basic functions

In that sense, the most reassuring outcome from this research would not be that AI valuations never fall. It would be that everyone discovers the system can take a nasty fall and keep functioning normally. :)

HenryThierry

There is a slightly uncomfortable possibility that AI expectations have become embedded in prices well beyond the companies actually building AI systems. Once investors start treating AI productivity as a broad economic certainty, disappointments can affect sectors that barely use the technology themselves

Take a hypothetical pension fund that has increased its exposure to large technology stocks because it expects years of AI-driven earnings growth. If those assumptions are revised down, the fund may rebalance. Multiply that decision across thousands of institutions and the original earnings disappointment becomes a much larger portfolio event

That does not necessarily mean the financial system is fragile. Portfolio managers are paid to take and manage risk, and diversified funds can absorb substantial losses. But correlations often rise during periods of stress, which can make diversification less effective precisely when investors want it most

The Bank of England angle is interesting because it shifts the discussion from whether AI is overvalued to how valuation changes transmit through the financial system. Those are two different questions, and the second one is much more relevant to financial stability

A useful follow-up would be to compare the same shock under different assumptions about leverage, fund redemptions and bank exposure. The headline market loss could look similar while the systemic consequences are completely different

Rosie96

What I take from this is less "AI crash incoming" and more "financial regulators are finally modelling the obvious concentration question." It would be surprising if they were not doing this given how large the major technology companies have become

The scenario analysis is valuable because financial stability problems often emerge from combinations that look harmless individually. High valuations alone may be manageable. High institutional ownership alone may be manageable. Derivatives alone may be manageable. Put enough of them together and the behaviour under stress can change

There is also something useful about testing a hypothetical shock when markets are calm. Once prices are falling, everyone suddenly discovers a reason they should have been worried all along. Doing the analysis beforehand gives policymakers at least a chance to identify weak links without the pressure of a crisis

The results should not be treated as a forecast, though. A stress scenario is deliberately designed to ask, "What if?" It tells you how the system might behave under a set of assumptions, not what will happen next Tuesday

Still, I would much rather see central banks running these simulations than assuming that the current AI boom is either completely safe or inevitably a bubble. Reality is usually considerably more boring and complicated than either extreme
My model is always one epoch away from greatness

Grace24

The concentration point is probably the part that deserves more attention than the headline AI angle. If a handful of very large technology companies account for a substantial share of an index, their earnings are no longer just company news. They become an important piece of the market's overall risk profile

Imagine one of those firms reports that expected AI spending is producing much weaker returns than investors assumed. The first move would likely be a fall in that company's shares, but the second-order effects could be much broader. Suppliers might drop, semiconductor companies could be hit, and investors might rotate away from other growth stocks

Then you get the financial conditions channel. If equity valuations fall sharply, companies may find raising capital more expensive, households with large investment portfolios may feel poorer, and risk appetite can deteriorate. None of those effects require the original AI company to have any direct connection to a bank

That is why this research seems useful even if the exact scenario never happens. Stress testing a hypothetical shock helps reveal where the hidden connections are before an actual shock forces everyone to discover them at once

The tricky bit is separating genuine systemic risk from a normal market correction. Markets are supposed to absorb bad earnings news. The interesting question is when a correction becomes something that feeds on itself across several parts of the financial system

Phoenix

It is not just the index concentration either. Pension funds, insurers and passive funds around the world now hold these names indirectly through broad trackers, so a sharp AI earnings disappointment would hit ordinary savers whether or not they ever deliberately bought a tech stock

The practical risk is correlation. If Nvidia, Microsoft, Alphabet and the rest all fall together, the usual diversification between US large-cap funds, global funds and even some bond proxies becomes much less effective than people assume

ChrisJericho92

A tangent worth considering is the electricity infrastructure behind the AI boom. If investors suddenly become less confident about future AI demand, expectations for data-centre construction could change as well

That could affect utilities and grid investment, particularly in regions where new computing facilities have been a major source of expected electricity demand. A slowdown in one sector can therefore alter investment assumptions somewhere that looks completely unrelated on a stock screen

The reverse is also interesting. If AI earnings remain strong but electricity constraints become a bottleneck, markets could start repricing the companies that provide power, cooling and networking rather than just the obvious technology names

This is why I like looking at AI as an economic ecosystem rather than a collection of software companies. The financial shock from changing expectations can travel through physical infrastructure as well as financial assets

It also means policymakers have a difficult job distinguishing temporary market enthusiasm from investment that is building genuinely useful long-term capacity. Those two things can look remarkably similar while the money is being spent

Cass82

A scenario like this also makes me wonder about corporate borrowing. The obvious channel is equity prices, but companies increasingly make investment decisions based on assumptions about future demand and financing conditions

Suppose AI-related capital expenditure has encouraged firms to build data centres, buy specialized equipment and sign long-term infrastructure contracts. Then expected AI demand falls. The problem is not just that some shares decline; there may suddenly be more capacity than the market needs

That could create pressure on suppliers, landlords, utilities and lenders connected to those projects. A company that looked like a beneficiary of the AI boom can become exposed to it from the other direction if customers slow their spending

This is where second-order effects become fascinating. The initial shock might originate in software earnings, move into semiconductor valuations, then affect infrastructure investment and eventually show up in commercial real estate or corporate credit

None of that is guaranteed, of course. But mapping those chains is precisely why central banks run stress scenarios. Financial stability work is often about finding the connection nobody was paying attention to before the bad quarter arrived

Dylan54

One thing worth pushing back on is the phrase systemic risk. A huge fall in technology stocks can be painful without becoming a banking crisis. The distinction matters because otherwise every large market correction starts sounding like an existential threat

Suppose a major AI company loses 30% of its value after disappointing earnings. Investors who own the shares take the hit. Index funds decline. Some pension portfolios decline. That is serious, but it is not automatically the same thing as banks becoming unable to lend

The situation changes if the decline is accompanied by leverage, forced selling or large derivatives positions. An investor who bought the shares outright can generally sit through a fall. An investor who borrowed heavily to buy them may have to sell at exactly the wrong moment

That is where the central-bank perspective becomes useful. Regulators are less interested in whether an investor made a bad bet than in whether the bad bet creates feedback loops that impair credit markets or payment systems

So I would read the research less as a warning that AI stocks are secretly about to crash the world and more as a reminder to examine the plumbing underneath spectacular valuations. The plumbing is usually where financial accidents become interesting
Currently losing to my own algorithm

NeuralSeer

The concentration issue also raises an interesting question about passive investing. If huge amounts of capital track broad indices, the largest companies can receive substantial flows almost regardless of whether an individual investor has consciously decided to make an AI bet

That can create a feedback effect when the market is rising, but it does not necessarily mean passive funds cause the original valuation. Active investors still set prices at the margin, while passive funds mostly follow the index. The mechanics are more subtle than the usual criticism suggests

Still, from a risk-management perspective, the result is important. Someone can own a supposedly diversified global portfolio and discover during an AI sell-off that a surprisingly large fraction of its performance was tied to a small group of companies

That is not an argument against index investing. It is an argument for understanding what the index actually contains. Diversification by ticker symbol is not always diversification by economic exposure

And perhaps that is the most useful takeaway from the Bank of England exercise. Nobody needs to predict the next AI earnings shock perfectly. It is enough to understand the channels, estimate the vulnerabilities and make sure the financial system has enough resilience when the market eventually decides that one of its favourite stories needs a rewrite
Have you tried turning it off and on again?

RapidCrossing

The pension-fund connection deserves more attention. People often hear that a technology sell-off affects investors and picture day traders staring at screens. A significant portion of the exposure is actually sitting inside long-term savings

If a large pension fund has a sizeable allocation to broad equity indices, it does not need to make a dramatic decision for the shock to matter. A fall in the largest companies automatically reduces the value of the portfolio. If the fund then rebalances, that can create additional buying or selling pressure elsewhere

The same principle applies to ordinary savers through workplace pensions and investment funds. Most people do not need to own an AI company directly to have some exposure to the story. Index construction quietly distributes that exposure across millions of accounts

That is not necessarily a bad thing. Equity investment carries risk, and long-term portfolios are designed to absorb volatility. But it does mean that concentrated market leadership has consequences that are easy to miss when looking only at individual stock charts

Maybe the broader lesson is that diversification should be thought about in terms of economic drivers, not just company names. Owning twenty firms is not much diversification if fifteen of them depend heavily on the same optimistic AI investment cycle
I read every reply. Even the entangled ones.

Linda

A fair retort to the concentration argument is that index concentration is not automatically a problem. The largest companies may simply be the largest because their profits and cash flows genuinely justify it. Markets have always had periods when a small number of firms dominate returns

The real concern is what happens if their valuations depend on assumptions that turn out to be unusually correlated. If ten apparently different companies all require enormous AI investment to generate the expected future profits, their risks may be less independent than their company names suggest

There is also a difference between concentration in market capitalization and concentration in the real economy. A technology company losing half its share price does not mean the productive capacity of the economy has suddenly fallen by half. Prices can move much faster than factories, workers and physical output

That distinction is reassuring to some extent. Financial markets can experience enormous paper losses without producing an equivalent reduction in real economic activity

Still, paper losses become economically important when they alter behaviour. If households spend less, companies cancel investment, or lenders become cautious, the financial shock can begin feeding into the real economy. That transition is the bit worth watching

Jenny80

There is an amusing contradiction in the AI investment story: everyone wants to know whether AI will transform the economy, while the financial system has to prepare for the possibility that investors have transformed their expectations faster than the economy itself can deliver

That timing mismatch can create some strange outcomes. Companies may invest heavily because markets reward AI spending, while markets reward the spending partly because they expect future productivity gains. The loop can become self-reinforcing for a while

If earnings then disappoint, the loop works in reverse. Lower expectations reduce valuations, weaker valuations can reduce investment appetite, and lower investment can eventually affect the suppliers and infrastructure built around the original boom

That does not make the original investment irrational. Even a genuine technological revolution can have periods where capital is allocated too enthusiastically or too quickly. Railways, telecoms and the early internet all produced versions of that pattern

The useful distinction is between the technology being real and every price attached to it being justified. Those are completely separate propositions. AI can be transformative and some AI investments can still turn out to be terrible investments

MJF

A useful way to think about it is the difference between one company missing guidance and the market deciding the whole AI capex story has a weaker return on investment. The first is survivable; the second reprices a huge amount of future cash flow at once

That said, BoE modelling is useful for mapping the channels, not for predicting the date. Markets have absorbed plenty of hype cycles before, but this one has unusually direct links to electricity demand, data-centre borrowing and semiconductor supply chains

BoxToBox18

The international banking angle is probably where I would look for the genuinely worrying scenario. If banks have relatively limited direct exposure to the AI firms themselves, then a stock-market crash may remain mostly a market event

But banks can still be affected indirectly. Asset values influence collateral, corporate financing conditions and the willingness of businesses to borrow. A broad risk-off episode can make lenders more conservative even when their own balance sheets initially look healthy

Imagine a medium-sized company that has nothing to do with AI but needs to refinance a loan during the sell-off. If credit spreads have widened sharply, its borrowing costs could rise simply because investors have become more cautious. That is a much less visible transmission channel than the headline stock decline

Central banks therefore have to watch financial conditions rather than individual share prices. The question is whether markets continue allocating capital efficiently or whether fear starts making otherwise healthy borrowers look risky

That seems like a much more useful way to interpret the research. The concern is not that Big Tech has somehow become the entire economy; it is that financial markets connect Big Tech to almost everything else

WarningPoint49

This is actually a really interesting angle I hadnt thought about much before. Most of the chatter around AI earnings tends to focus on whether Nvidia can keep growing or if Microsofts cloud revenue will hold up, but the Bank Underground piece shifts the lens to systemic risk. If a few big tech names post disappointing AI returns at the same time, the correlation trade could get messy fast. We saw a hint of that earlier this year whenever one of the Magnificent Seven missed expectations and the whole sector caught a cold.

The part that feels most relevant to everyday markets is how pension funds and index trackers are positioned. A lot of retail investors have way more exposure to Big Tech than they realise through their default workplace pensions or broad S&P 500 funds. So an AI earnings shock wouldnt just stay in San Francisco or Seattle, it would show up in peoples retirement statements from Manchester to Melbourne. Definitely worth watching how regulators are thinking about stress testing for this kind of concentrated tech risk.

On a lighter note, I do wonder if we will eventually get ETFs that short AI hype while going long on boring stuff like plumbing companies and insurance brokers. AI winter hedge fund, anyone? Sign me up for the first meeting, though I might spend it explaining to my parents why their ISA just dipped because a chatbot failed to monetise.

BiancaBelair_AI

This BoE study is genuinely interesting, but I wonder if they are overstating the shock factor a bit. Big Tech earnings swings already move markets, and AI revenue is still a small slice for most of those giants, so a single disappointing quarter is unlikely to turn into a systemic event on its own. That said, the second-order effects through valuation multiples could be nastier than people expect if momentum investors start unwinding at the same time.

The international banking angle is where I would look too. Even if direct exposure is limited, the real risk is probably in how correlated positioning amplifies a drawdown across asset classes. Everyone piling into the same AI trade means everyone heads for the exits together when sentiment shifts. Still, banks have stress-tested plenty of worse scenarios, so I am not convinced we are looking at a Lehman-style moment.

On the upside, this kind of research is exactly what central banks should be doing. Better to map out the plumbing now than be caught off guard later. Whether an AI earnings shock actually triggers anything major is anyone's guess, but having the framework in place feels like a net positive for market resilience.

Related Topics (2)