Just Ask Google: Measuring Investor Attention With Search Data

Behavioural finance had a favourite word: attention. Investors have limited attention. They cannot analyse ten thousand stocks, so they buy the ones that grab them. Attention-grabbing stocks get bid up and later come back down. Neglected stocks are slow to reflect news.

It is a compelling theory, and it produced a large body of research. It also had an awkward hole at its centre: nobody could measure attention.

So researchers used proxies. Trading volume, on the theory that people trade what they notice. Extreme returns, on the theory that big moves grab eyeballs. News article counts, on the theory that coverage generates interest. Every one of these is indirect. Volume goes up for reasons other than attention. News is written about stocks for reasons other than investor interest. And crucially, all of these proxies measure attention after the fact, as a consequence of something that already happened.

Zhi Da, Joseph Engelberg and Pengjie Gao noticed that in the modern world, there is a place where attention leaves a direct, timestamped footprint.

The problem: proxies that measure the shadow, not the object

The trouble with proxy variables is that they always contain something else.

Suppose you use trading volume to measure attention. But volume is also driven by institutional rebalancing, by index changes, by algorithmic market making, by liquidity needs. When volume spikes, you genuinely do not know whether retail investors are paying attention or whether a pension fund is quietly reweighting.

Or suppose you use news coverage. But a story appearing in the Wall Street Journal tells you that a journalist paid attention. It does not tell you that anybody read it. There is a big difference between information being available and information being consumed, and every news-based proxy conflates the two.

What you want is a measurement of the thing itself: did an investor actively go looking for information about this stock?

The key idea via analogy: the footprint in the search bar

Here is the insight, and it is one of those ideas that seems obvious the instant it is said and was invisible before.

If you are interested in a stock, you search for it.

Not passively. Not incidentally. You open a browser, you type the ticker, you hit enter. That is a deliberate, revealed act of attention. Nobody accidentally googles a ticker symbol. It is the digital equivalent of watching someone walk across a room to pick up a specific book.

And Google publishes aggregate search volume data.

So the authors proposed Search Volume Index, the frequency with which people search Google for a stock's ticker, as a direct measure of investor attention. Not a proxy for it. A measurement of it.

The design choice to search on the ticker rather than the company name is important and clever. If you search for "Apple," you might be looking for fruit, or a recipe, or a phone. If you search for "AAPL," you are unambiguously interested in the stock. The ticker acts as a filter that isolates financial interest from general interest, which is exactly what you want.

They then established that the measure was actually doing what they claimed.

It is correlated with the old proxies, but different from them. It moves with volume and news coverage, which confirms it is picking up something real. It is not identical to them, which confirms it is adding new information rather than just repackaging what we had.

It is more timely. The old proxies register attention slowly and with a lag. Search volume registers it the moment it happens. That matters enormously if you want to know whether attention precedes price moves, which is the whole question.

It measures retail attention specifically. Institutional investors do not google tickers. They have Bloomberg terminals, sell-side research, direct company access. The person typing a ticker into a search bar is overwhelmingly an individual investor. So the measure gives a relatively clean read on the attention of exactly the group that behavioural theories are about.

With a real measurement in hand, they went on to show that it predicts things. A spike in search interest is followed by price pressure, and the pattern is consistent with the attention-driven story: retail buying pushes prices up in the short run, with a subsequent reversal. They applied it to IPOs, where the attention story is most vivid, and found that heavy search interest ahead of an offering was associated with a bigger first-day pop and weaker returns afterwards, exactly the signature of temporary attention-driven buying pressure.

Why it mattered

• It was one of the founding papers of alternative data in finance. The idea that a data exhaust stream from a technology company, generated for entirely non-financial reasons, could be repurposed as a financial signal is now the basis of an entire industry. Satellite images, credit card panels, app downloads, web scraping, all of it follows the logic this paper demonstrated: the digital traces of ordinary behaviour are data.
• It gave behavioural finance a ruler. A theory that cannot measure its central variable is in a precarious position. After this paper, attention was observable, and hypotheses about it became testable rather than assertable.
• It isolated the retail investor. Cleanly separating retail attention from institutional activity had been very difficult. Search data does it almost for free, which unlocked a lot of subsequent research on retail behaviour.
• It reversed the causal arrow. Because search is timely, you can ask whether attention leads prices rather than merely following them. That is the question that makes the concept tradeable, and this was the first measure that could address it.
• It is beautifully simple. No proprietary data, no exotic modelling. Anyone could download the search index and reproduce it, and thousands of researchers and practitioners did.

The honest limitations

• The data source is a black box you do not control. Google's search index is sampled, normalised, and its methodology has changed over the years without notice. It is not an audited financial data series, and a signal built entirely on a private company's freely-provided data stream is a business risk as much as a research one.
• The mapping from tickers to searches is noisy. Plenty of tickers are also ordinary words or acronyms. Careful filtering helps, but it does not eliminate the contamination, and the problem is worse for small companies.
• It works best exactly where trading is hardest. The attention effect is strongest in small, illiquid, retail-dominated stocks, which are also the stocks where transaction costs and short-sale constraints eat the profits. The signal is real; the tradeable version of it is a narrower thing.
• It has been crowded and it has decayed. Once the paper was published and Google Trends became a standard tool on every quant's desk, the easy version of this trade got competed away. The concept is durable. The specific edge documented in the sample is not.
• The search landscape has changed. People increasingly get financial information from social platforms, apps and chat interfaces rather than a search engine. A measure of attention built on Google in 2011 is measuring a smaller and less representative slice of attention today.

The one-line takeaway

Da, Engelberg and Gao realised that a retail investor's Google search for a ticker is a direct, timestamped act of attention, turned that into the first genuine measurement of a variable behavioural finance had been theorising about for decades, and in doing so wrote one of the founding documents of the alternative data industry.