It is widely believed that economic uncertainty, driven by uncertainty about government actions, has been unusually high for several years. Heightened uncertainty has been blamed for dampening business investment, household spending, output, and hiring.
Scott Baker of Stanford University opened this conference on the ramifications of policy uncertainty with a presentation that cut straight to the core of the matter: how much policy incertainty is there and how does it influence the economy?
Baker and coauthors Nicholas Bloom and Steven Davis have developed an index to measure the overall level of policy-related economic uncertainty and examine its dynamic relationship to output, investment and employment. Using their index to estimate uncertainty historically, they find spikes associated with major political and economic events. They find uncertainty rising sharply since 2008, and that rising uncertainty foreshadows declines in GDP and employment.
Baker, a Becker Friedman Visiting Fellow, explained that half of the index is based on the number of references to policy uncertainty in the news media. The other half of the index incorporates the number of federal tax code provisions set to expire, forecaster disagreement on the government purchases of goods; and forecaster disagreement on inflation (each weighted at one-sixth of the total.
The authors have leveraged the availability of electronic news sources and powerful data analysis technology to measure uncertainty in the news. They conduct a monthly search of 10 major market newspapers for articles containing key such as 'uncertainty' or 'economy' along with terms such as 'policy', 'regulation', 'tax' etc.
The tax code expiration factor is an annual component based on data from the Congressional Budget Office. The CBO applies a 50 percent discounting to estimates of the total dollar amount of the expiring provisions to come up with a discounted dollar-weighted index value of the number of expiring provisions. Temporary tax code provisions are a source of uncertainty for businesses and households, as Congress has a habit of extending them at the last minute and lead to murkier outlooks for federal spending and borrowing.
Baker showed an empirical observation that the dispersion among forecasters of government and local spending, as well as the consumer price index, spikes around significant policy events. However, one discussant claimed that only about 20 percent of variation in forecasters' estimates of government and local spending and consumer price index shifts for the next year can be attributed to economic policy uncertainty, suggesting that the inclusion of forecaster dispersion as a component of the index is of limited value.
One discussant questioned the suitability of the news component, suggesting that sophisticated investors may be better informed than a media pundit. This was countered with the observation that many unsophisticated agents driven by mass news act in the economy, and the uncertainty over their actions may still affect the decisions of a more sophisticated investor. Baker also commented that by analyzing the news the day after big stock market moves, they had observed over the last few years that the number of such moves driven by policy has dramatically increased.
Baker further addressed the issues of the suitability and accuracy of the index for measuring policy uncertainty A graphical comparison of the index with the Chicago Board Options Exchange volatility indext (VIX) shows that while they have moved in tandem to notable degree, in the most recent year, the VIX has dropped while policy uncertainty has remained high.
To confirm accuracy of their index, Baker said the research team performed a news article audit, reading 3,500 articles that had been picked up in their automated analysis based on search terms to judge whether stories were actually about economic policy uncertainty. They found a 65 percent accuracy rate in their search. This led to considerable amusement and inquiry about how the authors split up the reading workload. Baker noted that they were beginning to look at more sophisticated machine learning algorithms for classifying news stories.
Text analysis allowed them to drill down into news stories to pinpoint issues associated with uncertainty. Health care, labor regulation, national security and sovereign debt and currency issues were the most important sources of uncertainty.
The authors addressed the question of whether policy uncertainty drives overall economic uncertainty is addressed by using their method of counting keywords and combinations of words to construct a news-based measure of overall economic uncertainty. By tracking the ratio of news articles that meet their criteria for policy-related uncertainty to those that reflect the broader uncertainty measure, they concluded that policy-related concerns are an increasingly important aspect of overall economic uncertainty, and account for most of the movements in overall economic uncertainty in recent years.
Further empirical results provide evidence at least of important co-movements between the BBD policy-related uncertainty and index and real industrial production and employment, with some suggestive evidence on causation.
The BBD uncertainty index has received considerable attention in the research community and the press. It was cited by many other presenters throughout the conference. Their work has gained enough media attention that the authors find they must now filter their uncertainty search results to remove articles that describe their index.