Nowcasting

AFR - Nowcasting

Google’s on Top of Today, Australian Financial Review, 20 September 2011

Some days, it seems that everyone has a crystal ball. Bank economists boldly predict exchange rate movements. Political pundits use polls to predict the next election. And fund managers vie to be the best stock-picker.

Alas, many of these forecasts aren’t much good. Exchange rates are equally likely to rise as to fall. Polls years out from an election have little predictive power. And the typical managed fund underperforms the All Ordinaries index.

Faced with the dismal performance of forecasting the future, one firm is taking a more modest tack. The wonks at Google are hoping that their new project will tell us what’s happening today.

If it sounds unambitious, consider that economic statistics are typically released with a substantial lag. The Australian Bureau of Statistics produces unemployment figures about six weeks after the end of the month, inflation numbers about eight weeks after the end of the quarter, and growth numbers 12 weeks after the end of the quarter. As the saying goes, this makes counter-cyclical policy like driving a car down a winding road while watching out the rear vision mirror.

One of the main forces behind what Google calls its ‘nowcasting’ project is its chief economist, Hal Varian. Formerly at the University of California Berkeley, Varian is a doyen of the field known as ‘the economics of information’.

In a recent presentation to the Australian Conference of Economists, Varian showed how the firm went about creating ‘Google flu trends’, which provides real-time measures of influenza prevalence based on searches for flu-related terms (such as symptoms and medication).

Since the Australian Influenza Surveillance Reports generally appear with a two week lag, Google flu trends can provide early warning of a sudden spike in flu cases. Search data won’t beat government statistics for accuracy, but its value is to provide a decent proxy that’s available in real time.

In the realm of economics, UK house prices have been shown to track searches for ‘estate agents’, while Australian consumer confidence fits closely the number of searches for new vehicles (and, surprisingly, crime).

In the case of unemployment, searches for ‘welfare’ and ‘unemployment’ spiked in the US in mid-2008, just as the national jobless rate passed 5 percent. Before the welfare data and labour force surveys had been compiled, search data could have indicated to economic policymakers that storm clouds were gathering. And given the well-known lags in fiscal policy, search data is worth using anytime we’re worried about a future downturn.

A cute feature of using search data to look at joblessness is that it also points to distinct patterns of search terms among the unemployed – many of whom are young men. Varian finds that the first set of terms to spike are labour market related (eg. ‘jobs classifieds’, ‘unemployment benefits’). The second phase sees an increase in searches for new technologies (eg. ‘ipod apps’, ‘free ringtone’). The third stage of unemployment searches are for low-cost entertainment (eg. ‘guitar scales beginner’, ‘home workout routines’). The fourth stage of unemployment searches are for adult content (eg. ‘adult video’, ‘porn tube’).

Rivalling Elisabeth Kübler-Ross’s ‘five stages of grief’, Google’s ‘four stages of unemployment’ is a touching story about how the US recession has affected everyday life. The stages of unemployment searches are as much a part of life as the fact that increased search volume for ‘vodka’ is followed by a spike in searches for ‘hangover cure’.

Some companies (such as credit card firms and travel agents) already use real-time data to monitor their businesses. But others might benefit from drawing on search data. For example, in a demand-driven system, universities should take notice if search volumes suddenly shift from ‘accounting ATAR’ to ‘engineering ATAR’.

Nowcasting is just one of the features that makes Google an interesting company to watch. In human resources, the firm has a policy of giving all employees a day each week to work on their own projects. In evaluation, it makes extensive use of randomised policy trials; as Varian points out, ‘any time you use Google you are in many treatment and control groups’. It runs many laboratory projects, including self-drive cars, which navigate using Google Street View.

It may not get everything right, but in its 13-year history, Google has shown itself to be one of the world’s most progressive companies. Its future is hard to forecast, but right now, the folks at Google are producing more than their share of the world’s innovative ideas.