What actually happens when a central bank intervenes in the FX market?

Sometimes a central bank stops describing the exchange rate and starts trading it. The mechanics are less mysterious than the headlines suggest, and the arithmetic of scale explains most of what happens next.

Sterilised versus unsterilised

This distinction decides almost everything.

Unsterilised intervention lets the domestic money supply change. A central bank buying foreign currency creates domestic money to pay for it; the monetary base expands. That is not really FX policy — it is monetary policy conducted through the exchange rate, and it works for the same reason any monetary easing works.

Sterilised intervention offsets the operation with a domestic open-market transaction, so the monetary base is unchanged even though foreign reserves have moved. The central bank has swapped one asset for another and altered nothing about the stance of policy. Most G7 intervention is sterilised, and the evidence on whether sterilised intervention has durable effects is genuinely mixed. Two channels are proposed: a portfolio-balance channel (investors must be paid to hold a different mix of assets — generally found to be weak in deep markets) and a signalling channel (the operation reveals the authorities' intentions and their tolerance — generally found to be the stronger of the two).

Who actually presses the button

Not always the central bank. In Japan, the Ministry of Finance decides on intervention and the Bank of Japan executes as its agent. In the United States, decisions are shared between the Treasury, through its Exchange Stabilization Fund, and the Federal Reserve. Reading intervention commentary without knowing this leads to a lot of confusion about whose statement matters.

The arithmetic of scale

Suppose an authority sells $20 billion of reserves to buy yen at 150:

20,000,000,000 × 150 = ¥3.0 trillion of yen purchased

Impressive. Now the denominator. The BIS 2025 survey put the yen on one side of about 17% of a $9.6 trillion daily market — roughly $1.6 trillion of yen turnover per day. So a $20 billion operation is about 1.2% of one day's flow.

That ratio is the whole story of intervention. Concentrated into twenty minutes it is overwhelming and produces a violent move. Measured against a year of flow it is small. Which is why the standard finding is that intervention against a fundamental trend, without a change in policy behind it, tends to have a short half-life — while intervention with the rate differential, or accompanied by an actual policy change, lasts much longer.

Documented episodes

• Plaza Accord, 22 September 1985. The G5 agreed to act to depreciate the dollar. USD/JPY fell from roughly 240 to roughly 150 within about a year. The Louvre Accord of February 1987 was an attempt to stop the fall they had started.
• 17–18 March 2011. After the Tōhoku earthquake the yen spiked to a then-record high near 76.25 per dollar in early Tokyo trading on 17 March, and on 18 March the G7 conducted a coordinated intervention to weaken it — the first joint G7 FX action since 2000.
• September–October 2022. Japan bought yen for the first time since 1998, beginning on 22 September. Its Ministry of Finance later reported roughly ¥9.2 trillion of intervention across that window.
• Late April–early May 2024. Japan reported roughly ¥9.8 trillion of further yen-buying intervention.

The asymmetry that decides outcomes

A central bank selling its own currency can create that currency without limit. Its capacity is, in the literal sense, unbounded — the constraints are inflation and the size of the balance sheet it is willing to carry.

A central bank buying its own currency can only spend the foreign reserves it has. That stock is finite and observable, and the market can count it.

This is why defences of a weak currency fail and defences against strength usually do not — until, as the last lesson of this unit shows, the defender simply decides the balance sheet cost is no longer worth paying.

Try it now

1. USD/JPY over the longest window this page holds is below, as candles. Navigate to September and October 2022 and find the two sessions with the longest upper wicks. Measure each from its high down to its close: that distance is the intervention, and a daily bar is the most resolution a chart can give you for a reversal that happened inside one session.

2. Drop a Level at the high of the first of them, then follow the pair for the eight weeks after. Did the level hold?
3. An intervention spends reserves; a rate differential does not. Write one neutral sentence connecting what you observed to the fact that one central bank was raising rates through that autumn and the other was not. Describe only; infer nothing about the future.