Searching for Signal: what Google searches say about risk appetite, volatility and markets, October 2026

We combined Google Trends from markets across the Americas, Europe, Asia, the Middle East and Africa with five years of weekly US data and 22 years of monthly data, tested against FinzData prices. Searches lag returns, forecast volatility, nowcast jobless claims, and caught France's unrest in real time.

• Lags: Searches follow the market. A bad week for stocks lifts next week's searches (Holm-adjusted p < 0.001); no search term forecasts the direction of returns.
• Vol: Searches add to volatility forecasts, even beyond the VIX and out of sample, but the gain is small (under one point of R²), and trading on it would have lowered the Sharpe ratio.
• Low: Attention is low. Index searches are 8% to 41% below their 12-month averages in all six core markets, and AI-investing searches have collapsed.
• France: France's unrest shows up clearly: protest searches peaked on October 17 and fuel-price searches flagged stress in February. Search confirmed the unrest; it did not predict it.
• Beyond the big six: Across 11 more markets in Asia, the Middle East and Africa, people search currencies and gold far more than stock indexes. Gold peaked between late January and late March; bitcoin peaked the week of February 1 in most of them, as in the big six.

• 53 wks: Six-country snapshot: weekly Google Trends, 20 local-language terms per country in four themes, last 4 weeks vs 12-month average
• 253 wks: Lead/lag study: US weekly searches Oct 2021-Oct 2026 vs FinzData SPY, VIX, realized volatility and first-release jobless claims
• ASVI: Abnormal search = log search minus log of its 8-week median (Da, Engelberg & Gao); regressions control for recent returns and VIX, Newey-West errors
• Checks: Granger tests both ways, Holm correction, 2,000-draw permutation tests, split samples, alignment shifts, 22 years of monthly data

Searches for each country's main stock index were below their 12-month averages in all six core markets in the four weeks to October 17, 2026: Korea -41%, the US -25% (searches for 'stock market'), Japan and Germany -22%, the UK -20% and France -8%. Retail attention to equities is low.

AI-investing searches boomed in the first half of 2026 and have faded everywhere: Korea -80%, the US -67%, Japan -42%, the UK -17%, France -16% and Germany -4% against their 12-month averages.

Interest rates are the one economic worry rising: searches are above average in Germany (+19%), Japan (+11%) and France (+7%), and below average in the UK, the US and Korea.

Inflation searches are below their averages in all six countries, most of all in Germany (-43%) and the US (-33%).

Bitcoin searches peaked in the same week, starting February 1, 2026, in all six countries, and are now 29% to 49% below their 12-month averages: one global retail cycle.

Korean market searches move with both the size and the direction of the market. The weekly correlation between KOSPI searches and EWY's absolute weekly move was 0.45, and with EWY's price level 0.61: Koreans search more when stocks are high. The busiest week, starting July 26, came with a 3.6% EWY fall; the October sell-off produced the latest spike.

The most-searched economic and trading terms differ by country: unemployment in the US, the exchange rate in Korea, interest rates in Japan and Germany, the Budget in the UK. Gold leads trading searches in the US, UK and Germany. In every market, index searches rise with the size of weekly market moves.

France's unrest is unmistakable in search. Daily searches for 'manifestation' (protest) climbed from mid-September, when school walkouts began in the Paris suburbs, and hit their 90-day peak on October 17, the day 550,000 to 800,000 people marched and violence broke out in several cities. Searches for 'émeutes' (riots) appeared only that week.

A critical caveat: search confirmed the unrest as it happened; it did not predict it. Protest searches rose with the school walkouts, not before them, and one episode cannot show that search anticipates social unrest.

The cost-of-living strain was visible months earlier: French searches for fuel prices ('prix carburant') surged between February and April 2026. Strike and protest searches then took off from mid-September. France's unemployment searches are 10% above average and recession searches 44% above, from a small base.

The fuel-price surge came seven months before the October protests. It fits the story of building cost-of-living strain, but with a single episode we cannot say whether fuel-price searches reliably precede unrest.

Every market term we tracked is searched less now than over the past year. Over the last four weeks, searches for "stock market" ran 25% below their 12-month average, "S&P 500" 38% below, "Nasdaq" 36% below and "Nvidia stock" 32% below. Bitcoin is still the most-searched term in the group (37% of its searches over the year) but has fallen furthest, down 49% from its average after peaking the week of February 1.

Unemployment is what Americans search most about the economy: 43% of searches in this group over the past year, peaking the first week of January. Inflation is second at 25%. All five terms are below their 12-month averages, and tariffs have dropped the most, down 67% since their peak in the week of February 15.

Gold price dominates US trading searches: 89% of the group over the year, with a peak the week of January 25. Searches for options trading, day trading and futures trading barely register beside it, and "0DTE" is too rare to measure. Even gold searches are now half their 12-month average.

"AI trading" was the quant-and-AI search story of 2026: 70% of searches in this group, rising to a peak the week of June 7. Since July it has collapsed and is now 67% below its 12-month average. Searches for algorithmic trading (-72%) and backtesting (-63%) have fallen even more.

Market searches rise when the market moves hard. Across 52 weeks, the weekly correlation between searches for "stock market" and SPY's absolute weekly move was 0.40. The busiest search week, starting March 1, saw SPY fall 2.0%. Searches were lower when prices were higher (correlation -0.48), which fits today's quiet, low-attention market.

The fastest-rising searches linked to "stock market" over the last 90 days are mostly practical: whether markets are open on Columbus Day and Labor Day. Searches about Warren Buffett on CNBC (+4,700%) and a story on AI images of market claims also broke out. Overall, the most common related searches are "stock market today" and "stock market news".

Korean searches for the KOSPI (코스피) are the most popular market search in the country, 41% of the group over the year, and they follow the market closely. They peaked the week of July 26 and then faded, but in the week of October 11 they jumped 2.6 times, from 17 to 45, as Korean stocks sold off: EWY fell 15% that week. Samsung Electronics (-45% vs average) and SK hynix (-65%) searches are still well below their peaks.

Koreans search about the exchange rate (환율) more than anything else in the economy: 48% of searches in this group, with a peak the week of December 21, 2025. Real estate (부동산) is second at 36% and has held up best, only 10% below its 12-month average and peaking as recently as the week of August 2. Searches for recession (경기침체) barely register.

ETF searches dominated Korean trading searches in 2026, 75% of this group, and peaked the week of May 24. They have since collapsed to 79% below their 12-month average. Searches for US stocks (-53%) and foreign stocks (-73%) have fallen too, and short selling (공매도) searches are down 48%.

AI investing (AI 투자) is Korea's dominant quant search, 68% of the group, and it peaked the week of May 10. It is now 80% below its 12-month average. Searches for quant (퀀트) spiked once, in the week of March 22, and are 41% below average; auto-trading (자동매매) is down 36%.

Over 90 days, the fastest-rising searches around 주식 (stocks) point abroad: Moderna (+1,000%), Coca-Cola (+120%) and Roblox (+100%) stock. Stock splits are rising too (+50%). The most common related searches remain Samsung, US stocks and SK hynix.

Japanese investors mostly search for stock prices in general (株価), 78% of searches in this group, which peaked the week of June 14. Searches for the Nikkei 225 are 22% below their 12-month average and bitcoin 29% below, a gentler cooling than in the US. NISA, Japan's tax-free investment account, is down only 10%.

Interest rates (金利) dominate Japanese economic searches: 74% of this group over the year, with a peak as recent as the week of August 30. They are the one big economic term searched more now than on average (+11%). Searches for the weak yen (円安, -36%) and the Bank of Japan (日銀, -39%) have fallen.

Futures (先物) are Japan's most-searched trading term, 59% of the group, but they are 33% below average since a peak the week of March 8. FX searches are going the other way: 16% above their average, with the 12-month high in the week of October 4. Searches for US stocks (-34%) and high-dividend stocks (-32%) are down.

AI investing (AI 投資) is 63% of Japanese quant searches and peaked the week of May 24. At 42% below its average it has held up far better than in the US or Korea. Searches for backtests (バックテスト) are slightly above average (+3%), while system trading has fallen 58%.

Nikkei searches rise in weeks when Japanese stocks move hard: the weekly correlation with EWJ's absolute move was 0.39 over 52 weeks. The busiest week, starting March 1, came with an 8.2% fall in EWJ. Searches had almost no relationship with the price level (correlation -0.14).

The fastest-rising stock searches in Japan over 90 days are specific: a Nomura fund focused on structural change in Japanese stocks (+3,150%), Tanaka Kikinzoku, a precious-metals group (+1,450%), and rare-earth stocks (+150%). Investors are also comparing all-world index funds (オルカン) with high-dividend stocks.

The FTSE 100 is Britain's most-searched market term, 42% of this group, peaking the week of March 22. In the last four weeks it was 20% below its 12-month average, the S&P 500 32% below and bitcoin 38% below. Searches for the stocks and shares ISA, the UK's tax-free investment account, are down 20%.

The Budget dominates British economic searches: 63% of this group over the year, with a peak in the week of November 23, 2025, around the Autumn Budget. Inflation and interest-rate searches are both 16% below average, and recession searches have more than halved (-52%).

As in the US, gold leads UK trading searches: 77% of the group, peaking the week of August 9, and now 51% below average. The trading app Trading 212 has held up best (-12%), while day-trading searches are down 32%.

AI trading is 67% of UK quant searches and peaked as late as the week of August 30; it is only 17% below its average, compared with 67% in the US. Searches for quant trading are 19% above average, the only rise in this group.

FTSE 100 searches rise in weeks when UK stocks move more: the weekly correlation with EWU's absolute move was 0.39. Searches were slightly lower when prices were higher (correlation -0.28).

The fastest-rising UK searches around shares are about tax and pay: an HMRC charge on stocks and shares ISAs (+4,100%) and a breakout story on OpenAI UK wages and shares. Mitie and Lloyds Banking Group shares are also rising.

Bitcoin (39%) and the DAX (26%) lead German market searches. In the last four weeks the DAX was 22% below its 12-month average, stocks (Aktien) 19% below and ETFs 35% below. Bitcoin is down 32% from an early-February peak, the same week it peaked in the US, Korea and the UK.

Interest rates (Zinsen) are Germany's main economic search, 55% of the group, and are 19% above their 12-month average after a peak the week of September 6. Unemployment searches are up 17%. Inflation searches have fallen 43% and the ECB 36%.

Gold price is 67% of German trading searches, peaking the week of January 25, and is now 43% below average. Searches for the broker Trade Republic are steady (+1%), and options searches are up 8%. Day trading barely registers in German.

Searches for 'Quant' jumped to their 12-month high in the single week of September 27, which lifts the four-week average to 46% above normal. French searches for 'quant' peaked the same week, which points to a story outside markets, so we treat the jump with caution. AI stocks (KI Aktien) are only 4% below average, while backtesting is down 79%.

German DAX searches follow the size of market moves closely: the weekly correlation with EWG's absolute move was 0.48, second only to France. The busiest week, starting March 1, came with a 7.9% fall in EWG.

The fastest-rising German stock searches include the Fed's rate decision (breakout), drone-defence stocks (+800%) and Zalando (+400%). Searches asking which stocks to buy now are up 110%.

'Bourse' (the stock market) is France's main market search, 55% of the group. It and the CAC 40 are only 7% and 8% below their 12-month averages, the smallest declines of any index search we tracked; both peaked the week of March 1. ETF (-42%) and bitcoin (-41%) searches have fallen much further.

Inflation is still France's top economic search (41% of the group) but is 25% below its average after a peak the week of March 29. Unemployment (chômage, 36%) is 10% above average and interest rates 7% above. Recession searches are 44% above average, but from a very small base: they are 1% of the group.

In France, 'trading' (34% of the group) and crypto (33%) lead trading searches, followed by the PEA, France's tax-advantaged share account (25%). Trading searches hit their 12-month high the week of September 27 and have since fallen to 36% below average; crypto is 45% below and the PEA 20% below. Gold is a much smaller search in France than in the US, UK or Germany.

'Quant' is also a common French word (it means 'as for'), so its level says little about investing; its spike in the week of September 27 matched Germany's. Searches for AI and the stock market (IA bourse) are 16% below average, robot trading 79% below and backtests 75% below. Algorithmic trading barely registers in French.

French CAC 40 searches follow the size of market moves more closely than any other index we studied: the weekly correlation with EWQ's absolute move was 0.51. The busiest week, starting March 1, came with a 7.8% fall in EWQ.

The fastest-rising French searches around 'bourse' are single stocks: the biotech Biophytis (+450%), Moderna (+250%), Aton, Haffner Energy, Legrand and Schneider Electric. Searches for exchange-traded funds are up 40%. The most common related searches are brokers: Bourse Direct and Boursorama.

We correlated abnormal searches with SPY's weekly return at offsets from eight weeks before to eight weeks after. The strongest correlations are with the previous week's return (-0.35 for 'stock market', -0.25 for 'stock market crash'): searches rise after falls. Bars where searches come first mostly sit inside the 95% noise band.

Controlling for the last two weeks' returns and the VIX change, no search term predicts SPY's return over the next week or month. Several predict higher realized volatility next week (t +2.3 to +3.4), and spikes in market-fear searches are followed by a lower VIX 13 weeks later. Searches for 'unemployment' are the exception: they were followed by a higher VIX 13 weeks later (t +3.0).

Granger tests ask whether one series helps predict another. Returns help predict searches for 'stock market' (p 0.0001) and 'recession' (p 0.0007), and the VIX helps predict most search terms. Searches do not help predict returns at the 5% level for any term.

Our fear composite averages standardized abnormal searches for 'stock market crash', 'recession' and 'unemployment'. It spikes with the VIX: same-week correlation +0.21 with the VIX change and -0.19 with SPY's return. The biggest spikes came in the 2022 bear market, August 2024 and April 2025.

In the 26 weeks with the highest fear searches, SPY rose 0.57% on average the next week versus 0.20% normally, but a permutation test puts that gap within chance (p = 0.10). Volatility is the reliable effect: next-week realized volatility averaged 19.3% versus 15.6%.

US bitcoin searches move with the size of bitcoin's weekly move (correlation +0.38) but not its direction, and big moves raise the next week's searches (t +2.9). Neither searches nor price help predict the other in Granger tests.

Weekly searches for 'unemployment' track initial jobless claims as first published: the weekly changes correlate at +0.17, and adding searches to a model of claims raises its explanatory power from 7% to 10% (t +2.4). Searches are available days before the Thursday release.

Monthly since 2004, searches for 'stock market crash' peaked in February 2018, March 2020, April 2025 and August 2024, all inside sell-offs. Months in the top 10% were followed by a 13.6% average 12-month SPY return versus 8.5% for all months, or 10.3% versus 7.5% excluding 2020.

In Korea, Japan, Germany, France, the UK and the US, abnormal index searches rise with the size of the ETF's move in the same week (+0.22 to +0.36) and the week before. Germany (p 0.007) and France (p 0.022) are the two markets where searches also help predict next week's move size.

The hardest test for any new signal is whether it adds to what markets already price. Controlling for this week's realized volatility and the VIX, abnormal searches for 'stock market crash', 'recession', 'stock market' and 'VIX' still predict next week's realized volatility (t +2.4 to +2.9). But the explanatory power rises only from 88.4% to between 88.7% and 89.1%: statistically real, economically small.

To rule out hindsight, we forecast next-week realized volatility each week from June 2024 using only data available at the time. Adding the fear composite cut forecast errors by 2.4% relative to a model with past volatility and the VIX, and the Clark-West test (t 2.52) says the gain is significant.

A 1% fall in SPY raises next week's abnormal searches for 'stock market' and 'recession' (t +2.0 and +2.6), and a 1% rise lowers them. Searches for 'stock market crash' and 'VIX' respond more to rallies, which calm them, than to falls, though not all of these effects are significant at 5%.

Google Trends reports a sample of searches and rescales each request, so the same query can return slightly different numbers. We pulled four terms twice, independently. For busy terms the two pulls agree almost perfectly: levels correlate at 0.98 to 0.997 and weekly changes at 0.90 to 0.99. Rare terms are much noisier, and we do not draw conclusions from them.

Of 26 fear-search spikes in five years, 11 came during or just after a stress week, 4 came before new stress and 12 had no stress nearby. Seen from the other side, only 8 of 49 stress weeks (SPY's worst 10% of weeks or a VIX of 25 or more) had a spike the week before. Search tells you stress has arrived, not that it is coming.

Halving SPY exposure for a week after each fear spike, with 5 basis points of cost per change, would have returned 8.8% a year against 10.4% for buy-and-hold, with slightly lower volatility, a lower Sharpe ratio (0.57 vs 0.64) and no smaller maximum drawdown. Markets tended to rebound after spikes, so cutting risk then gave up gains. The threshold was set with hindsight, so even this result is flattering.

We asked an AI critic to attack every headline claim. The strongest objections: news coverage may drive both searches and markets; searches may be moving from Google to AI chatbots and apps, overstating the drop in attention; and France is a single episode. We show each objection with the evidence that would settle it.

Four limits matter most: Google captures a shrinking slice of attention as investors use chatbots and apps; five years contains few independent crises; Granger tests show precedence, not cause; and the France result rests on one episode.

• Lag: Treat search data as a thermometer, not a forecast: it measures how the public reacts to what markets just did.
• Vol: It belongs in risk models as a small, real addition to volatility forecasts, not in return or timing models.
• Low: Low attention today means the next sharp move will draw retail back fast, as Korea showed in October.
• Caveat: Some of the fall in searches may be people asking AI chatbots instead of Google; levels need care, changes are more reliable.

• Oct 22: Bank of Korea decision, after a 22% two-week fall in Korean stocks.
• Oct 28: Fed decision; US 30-year yields reached 5.69% this month.
• Oct 29: ECB decision, with German and French rate searches above average.
• Oct 30: Bank of Japan decision, with Japanese rate searches 11% above average.

• 170+: local-language search terms across 17 markets in Asia, Europe, the Middle East, Africa and the Americas
• 5 yrs: weekly US searches tested against FinzData prices, the VIX, realized volatility and first-release jobless claims
• 22 yrs: monthly US searches for 'stock market crash', 'recession' and 'unemployment' since 2004
• API: FinzData prices, point-in-time macro and licensed series via Python, REST and MCP for AI agents

• Da, Z., Engelberg, J. and Gao, P. (2011), 'In Search of Attention', Journal of Finance 66(5) (rady.ucsd.edu)
• Da, Z., Engelberg, J. and Gao, P. (2015), 'The Sum of All FEARS', Review of Financial Studies 28(1) (ideas.repec.org)
• Preis, T., Moat, H. S. and Stanley, H. E. (2013), 'Quantifying Trading Behavior in Financial Markets Using Google Trends', Scientific Reports 3 (nih.gov (PMC3635219))
• Challet, D. and Bel Hadj Ayed, A. (2013), 'Predicting financial markets with Google Trends and not so random keywords' (arxiv.org/abs/1307.4643)
• Choi, H. and Varian, H. (2012), 'Predicting the Present with Google Trends', Economic Record 88 (Google Research)
• French unrest: AP via ABC News and CBS News (Oct 18, 2026); Fortune (Oct 19, 2026); central-bank calendars (abcnews.com, cbsnews.com, fortune.com)