Risk-on / Risk-off: How the Yen, Franc, Gold and the VIX Read Market Mood

Risk-on / Risk-off describes the overall mood of the market. When people feel brave (risk-on), money tends to flow into stocks and higher-yielding currencies. When they’re scared (risk-off), money tends to retreat to assets seen as safe. Once you understand this, it’s clear why several pairs that look unrelated can all fall on the same day. This article explains the mechanism, gauges such as the VIX, why relationships can flip, and works through the numbers to show a correlation turning from negative to positive.

What risk-on and risk-off mean

Picture all the world’s investors as one crowd. On good-news days, when the economy looks fine, they’re willing to hold things that swing hard in the hope of higher returns. When something frightening happens, such as a financial crisis, a war or a pandemic, they sell the risky things and hold cash or whatever they think won’t fall.

Nobody announces “today is risk-off”. It’s a label people use after the fact to describe which way many assets moved together. It’s a description, not a driver.

Why some currencies are safe havens

The interesting part is that “safe” doesn’t mean that country has the strongest economy. It comes from how money flows.

• JPY, the Japanese yen. Japan holds very large net assets abroad, and in a scare Japanese investors tend to bring money home. The other mechanism is that Japanese rates have been low for a long time, so people borrow yen to invest in higher-yielding currencies, known as the carry trade. When markets panic, everyone rushes to close those positions and has to buy yen back at the same time, so the yen strengthens sharply. Early August 2024 is a clear example: unwinding carry trades sent the yen up fast while Japanese stocks fell hard. The full mechanism is in The carry trade explained
• CHF, the Swiss franc. Switzerland has large net foreign assets and stable politics, so money flows in when Europe is in trouble. The inflows were strong enough that the Swiss National Bank intervened several times, for example capping the franc at 1.20 per euro in 2011 and then abandoning the cap without warning on 15 January 2015
• USD, the US dollar. It’s the world’s main currency for borrowing and settling debts. In a severe crisis everyone needs dollars to repay loans or post collateral (a dash for cash), so the dollar can rise even when the trouble started in the US, as in 2008 and March 2020. But when the market is worried about US policy itself, the dollar can fall together with US stocks, as in April 2025 amid worries about US import tariffs. To track the dollar, see The dollar index DXY
• Gold. It carries no risk of an issuer defaulting, so it’s often a refuge. But in the first days of a heavy sell-off gold can fall too, because people sell whatever is easy to sell to meet margin calls. In mid-March 2020 gold fell together with stocks before recovering. For rates and gold, read How interest rates move gold and currencies

On the risky side, AUD and NZD are tied to commodities and the Chinese economy, and have often paid higher rates than many currencies, which makes them typical carry-trade destinations. Emerging-market currencies behave similarly. Stocks and crypto depend directly on appetite for risk. Since 2020 the correlation between bitcoin and US tech stocks has risen clearly, so crypto has behaved more like another risky asset than a refuge.

What the VIX actually measures

The VIX (Cboe Volatility Index) is nicknamed the “fear gauge”, but what it really measures is the implied volatility of the S&P 500 over the next 30 days. It’s calculated from the prices of S&P 500 index options, both calls and puts across a range of strike prices, and shown as an annualised percentage.

Why does it rise when markets are scared? When stocks fall, people rush to buy puts to protect their portfolios, option prices go up, and the volatility backed out of those prices rises. That’s why the VIX usually rises on days when stocks fall.

You can turn a VIX reading into the size of the expected move (one standard deviation) like this:

1. The VIX is annualised, and 30 days is about 1/12 of a year
2. Volatility grows with the square root of time, so divide by √12 = 3.46
3. VIX 20 ÷ 3.46 = 5.77%: the options market expects the S&P 500 to stay within ±5.77% over 30 days about two-thirds of the time, if moves were normally distributed
4. For one day use √252 (trading days per year) = 15.87, giving 20 ÷ 15.87 = 1.26% per day

VIX	30-day range (±)	Per day (±)
12	3.46%	0.76%
20	5.77%	1.26%
30	8.66%	1.89%
40	11.55%	2.52%

Three things to know. First, the VIX gives the expected size of moves, not their direction. Second, it’s the price the market pays for insurance, so most of the time it sits above the volatility that actually follows. Third, real prices have fatter tails than a normal distribution, so days outside the range happen more often than the numbers suggest. Currencies and bonds have their own implied-volatility gauges too, such as the MOVE index for US Treasuries.

Relationships flip, and they have flipped

“The yen strengthens when markets are scared” or “bonds rise when stocks fall” sound like rules, but they’re the net result of several forces over a period. When the biggest force changes, the relationship changes.

• In 2022 the yen fell hard even with a war and global stocks falling, because the US Federal Reserve raised rates quickly while the Bank of Japan kept its rates low. The rate gap beat market mood
• US stocks and Treasuries tended to move opposite each other from roughly 2000 to 2020, so bonds could cushion a stock portfolio. In 2022 high inflation pushed rates up and both fell together, so the correlation turned positive
• The dollar rises in crises where everyone scrambles for cash, but can fall when the worry is about the US itself

So always ask what the main force is right now: rates, inflation or fear. The answer changes with the market regime. Read more in Market regime change.

Worked example: a correlation flips from negative to positive

Correlation (ρ) measures how much two return series move in the same direction, from −1 to +1. It has to be calculated from returns per bar, not from prices (the reason is in Correlation and portfolio risk).

ρ = Σ(a − ā)(b − b̄) ÷ √[ Σ(a − ā)² × Σ(b − b̄)² ]

Try it on five days of returns: series A = 1, −1, 2, −2, 0 (%) and series B = 0.5, −1, 0.5, −1, 1 (%).

1. Both means are exactly 0, so the deviations equal the numbers themselves
2. Daily products: 0.5 + 1 + 1 + 2 + 0 = 4.5
3. Sums of squares: A = 1 + 1 + 4 + 4 + 0 = 10, B = 0.25 + 1 + 0.25 + 1 + 1 = 3.5
4. √(10 × 3.5) = √35 = 5.916
5. ρ = 4.5 ÷ 5.916 = 0.76

Flip the sign of every B return and every product turns negative, giving ρ = −0.76. Same size, opposite meaning.

A rolling correlation repeats this calculation on every bar using only the last N bars. We built two synthetic return series of 240 bars from one common factor (standing in for “market mood”) plus random noise specific to each series.

• Bars 1–120: series B takes the common factor in reverse (multiplier −0.8), like a safe haven that cushions a portfolio
• Bars 121–240: series B takes it in the same direction (+0.8), like a safe haven that has stopped working
• Bars 181–200: the common factor swings four times harder, standing in for a panic

What the numbers show:

1. Over all 120 bars of regime 1, ρ = −0.47, yet the 20-bar window swings from −0.75 to −0.02 even though the true relationship never changes. That’s the noise of a short sample
2. The true relationship flips at bar 121, but the 20-bar window first crosses zero at bar 132, 12 bars late. A 60-bar window crosses at bar 151, 31 bars late. A longer window is steadier but notices later
3. In the calm part of regime 2, ρ = +0.47. In the 20-bar panic, ρ = 0.93, and both series fall together, by 9.3% and 9.1%
4. Calculated over all 240 bars, ρ = +0.28, a single number that matches none of the periods

All of these numbers come from this synthetic data set, not from a real market. The mechanism is the same, though: the correlation you see on a website or in your platform is an average of the past, and it always lags reality.

Why correlations jump towards 1 in a panic

In the model above, each return = common factor + its own noise. If series B’s multiplier is b and the common factor’s volatility is σf, then:

ρ = b × σf² ÷ √[ (σf² + σA²) × (b²σf² + σB²) ]

Set b = 0.8 and give both noise terms a volatility of 1.

• Calm, σf = 1: ρ = 0.8 ÷ √(2 × 1.64) = 0.44
• Panic, σf = 4: ρ = 12.8 ÷ √(17 × 11.24) = 0.93

The link between the two series (b) is exactly the same. The common factor simply swings so hard that it drowns the individual noise, and the correlation rises by itself. The formula matches what the synthetic data gave (+0.47 and 0.93). Forbes and Rigobon (2002) made this point about stock markets across countries: higher correlations in volatile periods are partly a statistical result of the higher volatility, and don’t necessarily mean the underlying link changed.

Real markets add further pressure. As prices fall, leveraged accounts get margin calls and must sell whatever they can, funds cut risk at the same time, and liquidity dries up. All of that is the same common factor getting bigger.

Diversification that looks good on a normal day can vanish on a panic day. Plan your total risk as if correlations will be close to 1 on that day.

Use market mood as context, not a signal

You see market mood in prices that have already moved. By the time the VIX spikes, most prices have moved too. So it’s not an entry-timing tool, but it’s useful for three things:

1. Check the big picture before you trade. See whether US stock indices, the VIX, AUDJPY, gold and the dollar index are telling the same story. If they all agree, mood is today’s main force, and the chart of any single pair carries less weight. If they disagree, the main force may be country-specific news, so check the economic calendar
2. Size positions for volatility. When the VIX rises a lot, prices swing wider and your usual Stop loss gets hit more easily. If you widen the Stop loss using ATR, cut the position size so the money at risk per trade stays the same, as in Risk 1% per trade and Position size from your Stop loss
3. Count trades that are really one bet. Buy AUDUSD, buy NZDJPY, buy a stock index and buy bitcoin may look like four trades, but they’re all a bet that markets stay brave. Set a risk cap per group, not just per trade

If you want to measure it yourself, run rolling correlations of daily returns over two windows at once, such as 20 and 60 days. If the two are far apart or have opposite signs, the relationship is changing. Don’t rely on figures remembered from an old article or book.

Common mistakes

• Assuming safe havens are always safe. The yen fell hard in 2022 and gold fell in mid-March 2020. Every item in the diagram is a tendency that can change
• Relying on one long-period correlation. The overall figure can match no period at all, like the +0.28 in the example while the first regime was −0.47
• Opening a “hedge” and feeling safe. Trades that offset each other on normal days can lose together on a panic day, like the two series falling about 9% together. See Hedging and locking losing trades
• Treating the VIX as a reversal signal. A high VIX says the market expects wide swings. It doesn’t say when prices will turn, and it can stay high for a long time
• Blaming mood for every move. Some days a pair moves on one country’s rate news. If stocks and the VIX are calm, the main force may not be risk-off
• Forgetting crypto is a risk asset too. When markets are scared, crypto usually doesn’t cushion a Forex portfolio. For crypto sizing, read Crypto position sizing by volatility