The next crisis

Hemingway, in The Sun Also Rises:

"How did you go bankrupt?" ... "Gradually and then suddenly."

Start with what the commentariat worries about today. AI comes in with two entries, Mythos-facilitated hacking and an AI investment bubble. We then have long-term interest rates hitting 5 or 6% and excessively indebted sovereigns. Add to that the pressure on central bank independence, energy shocks, tariffs and private credit.

All this in times of increasing demand for government expenditure, like national defence. Throw in the lack of growth and, finally, ageing populations, climate and populism.

These come together in one vicious loop.

The debt identity

Crises are almost always about debt, and the natural place to begin is the government debt-to-GDP ratio. It evolves according to

$$ \Delta d_t \approx (r_t - g_t) d_{t-1} - pb_t $$

where \(d_t\) is the debt ratio, \(r_t\) the effective real interest rate on the debt, \(g_t\) real GDP growth and \(pb_t\) the primary balance, revenues minus non-interest spending, relative to GDP. Interest makes the debt ratio grow, while growth and primary surpluses shrink it.

The term \((r_t - g_t) d_{t-1}\) is the snowball, how much the debt ratio grows by itself. If the interest rate is below the growth rate, the debt ratio falls even with a balanced primary budget. But once it is above, the government needs a primary surplus just to keep the ratio stable, a surplus that rises with the debt. The danger is that the required surplus becomes economically or politically unattainable.

The interest rate in the identity is not the headline 5 or 6% we see in the bond market, since those are nominal rates on new long-term borrowing, for maturities of 10 years or more. The effective rate is what the whole stock of debt pays, net of inflation, and it only moves towards market rates as the debt matures and is refinanced. But governments do not repay their debt, they roll it over, so the market rates show where the effective rate is heading.

The chart shows government debt to GDP in 2025, with the average maturity of the debt next to each bar.¹

How it all goes pear-shaped

An indebted country may enter a debt death spiral, where the government faces increasingly costly obligations in an environment in which it is difficult to raise taxes. More and more needs to be borrowed to service the existing debt and meet demands for public services. In the identity, the debt ratio rises, investors demand higher risk premia, the effective rate goes up and the snowball grows, so the debt ratio rises further.

The way it likely ends is a fiscal heart attack, or in Ray Dalio's terminology, a debt-induced heart attack. The markets refuse to roll over the debt, and the government is left with two bad choices, either emergency measures or default.

We economists explain the sudden jump with the concept of multiple equilibria. The same fundamentals (debt, interest rates and growth) can support two outcomes. Confident investors happily roll the debt over at low rates and the debt is sustainable, but if they lose confidence, they demand high rates, and those high rates make the debt unsustainable. So the fear is self-fulfilling.

The fundamentals decide whether the bad equilibrium is possible, and then something small, a trigger, pushes us into it. That is why there is little point in trying to predict the trigger.

Running out of dry powder

The worse the identity, the smaller the event needed to trigger the fiscal heart attack.

Systemic risk depends on the government's ability to respond. An event it can deal with without too much cost is not systemic.

We could afford significant monetary and fiscal expansion in 2008 and 2020, but that might be impossible today. Governments are more indebted, and after the inflation that followed 2020, it is much harder to monetise interventions. We have run out of dry powder.

That has two consequences.

First, it takes less to set off the crisis. If a shock like 2008 or 2020 comes today, it is increasingly likely to push us into the bad equilibrium.

Second, the markets know this. They are more alive to the possibility that the authorities cannot come to the rescue, so they are quicker to sell at the first sign of trouble, and that selling can itself bring on the crisis. Endogenous risk, in other words, where the market participants' own reactions make a crisis more likely.

Back to the list

What matters for a crisis, then, is both what makes the identity worse and what can trigger the heart attack.

Mythos-facilitated hacking does nothing to the identity, but it can be the trigger. A cyber attack is most dangerous when it coincides with market stress. That is the double coincidence. If, as many commentators claim, an attack can do significant damage, it could be enough on its own.

And AI speeds up crises. Financial institutions use AI to analyse the state of the system far faster than a human can. In a crisis, survival means acting first, withdrawing liquidity and selling before everybody else, so what once took days or weeks may play out in minutes or hours.

That makes a crisis much more damaging. The authorities have less time to step in, and their liquidity interventions may come too late. And when everybody tries to get out first, the selling feeds on itself.

The AI investment boom cuts both ways. We are spending a lot of money on data centres, which boosts the economy today. That will likely raise productivity, and with it growth. And whatever happens to the companies doing the investing, we will be left with the investments.

But much of it is borrowed, and that puts the bond markets under strain and pushes up interest rates, including the ones governments pay. And when rates are rising while the price the AI companies can charge for their tokens keeps falling, their financing position becomes increasingly precarious, and the risk premia they face go up.

In the worst case, the AI bubble bursts. Growth falls, and when the government bails out the banks and props up the economy, private losses become public debt. Cue the fiscal heart attack.

Long-term interest rates of 5 or 6% feed into the effective rate as the debt rolls over, and so into the snowball.

Then there are the excessively indebted sovereigns, which is why we have run out of dry powder.

The pressure on central bank independence pushes long-term interest rates up. If the markets suspect the central bank will keep rates low to help the government, inflation expectations and term premia rise, and the effective rate follows. It also weakens the credibility the bank needs when the crisis comes.

Energy shocks lower growth, and when governments subsidise energy bills, as many European governments did in 2022, they weaken the primary balance too.

Tariffs hurt growth, and retaliation makes that worse.

Private credit is a trigger. It is opaque, and if it runs into trouble, the losses show up where nobody was looking.

Defence spending is a direct hit to the primary balance, and it is going up.

The lack of growth works directly on the identity. Europe used to keep pace with the US until 2008, but since then, its economy has fallen behind by perhaps 15% to 25%, and without growth, the debt ratio does not shrink.

Ageing hits all three terms. The labour force shrinks, so growth falls, and retirees run down their savings, which raises interest rates. Meanwhile, pensions, health and care eat into the primary balance.

Climate does too. Subsidies and adaptation cost money, the borrowing to pay for the transition and policy uncertainty push rates up, and growth suffers now, even if acting helps later.

Populism hits all three terms and the trigger as well. Unfunded promises and tax cuts weaken the primary balance, and protectionism and less immigration lower growth. The markets also demand higher risk premia when policy becomes unpredictable. On top of that, it undermines the authorities, and their ability to fight crises depends crucially on their credibility.

A growing snowball forces tax rises and spending cuts, which feed populism, and populism makes the arithmetic worse still. That is the vicious loop.

Minsky, the US and the rest

The US matters more than its size suggests. In my work with Marcela Valenzuela and Ilknur Zer, covering 73 countries since 1900, we find that global risk cycles affect growth more than local ones, through capital flows, investment and the quality of borrowers. And the US alone explains about 30% of the variation in global risk.

The mechanism is simple. When risk is perceived to be low, risk appetite rises, we invest more, and growth picks up. But one cannot make good investments forever. Investment quality falls, and a few years later, growth reverses and a crisis becomes more likely. Just Minsky — stability is destabilising.

Credit matters too. Only two variables predict crises, low perceived risk and credit growth. Low risk comes first, since it drives credit growth. If investors think the world is safe, it is perfectly OK to take on more risk, which means more borrowing.

That means low perceived risk is good for growth, unless we are in an excessive credit cycle. Then the consequence for GDP is negative, and the reversal is worst when low risk persists.

The AI boom in the US fits the pattern. It fuels growth at home and, through global risk appetite, investment in the rest of the world. If it turns sour, the reversal hits the US, but likely even harder elsewhere, especially in emerging markets.

What to watch

The trigger cannot be predicted, but we can track the fundamentals. Four things in particular:

1. How fast higher market rates reach the effective rate on the debt, which depends on how much of it is short term.
2. Whether credit, term and liquidity premia start to widen together, as the market rate then jumps at once.
3. Whether the central bank becomes a large and persistent buyer of government debt, or starts setting rates to keep debt servicing costs down, a sign of fiscal dominance.
4. Whether the US sees a long stretch of low perceived risk and fast credit growth, a warning sign for both America and the rest of the world.

So

There is only one remedy — economic growth. But we are not likely to get it, as I will discuss in a later blog.