Bayes hosts econometrics conference against turbulent backdrop

Bayes Business School welcomed academics from across Europe for the 27th Dynamic Econometrics Conference last week.

Professor Giovanni Urga, co-chair of the programme committee and chair of the local organising committee, said: “This was a very timely scientific event where world leading researchers in the field discussed research and policy issues related to climate risk, use of factor models and financial analyses with big data, the appropriate use of ultra-high frequency data and monetary policy.”

In his presentation, Professor Urga showed the results of a new econometric specification of the Taylor Rule equation highlighting that the Federal Reserve’s decision to lower borrowing costs occurred notwithstanding a rise in consumer price inflation, which increased from 2.7 per cent (year-on-year) in July to 2.9 per cent in August, reflecting the inflationary effects of the ongoing trade conflict under the Trump administration.

“At the same time, the central bank’s preferred measure of underlying price dynamics, the personal consumption expenditures (PCE) price index, registered 2.6 per cent, above the stated 2 per cent target. From the perspective of standard Taylor-rule prescriptions, such conditions would normally call for a tightening of policy rather than an easing. This is also confirmed by my TVP-SVAR Augmented Taylor rule framework.

“The decision therefore highlights a credibility trade-off: while the rate cut sought to cushion the real economy against trade-related uncertainty, it also represented a departure from rule-based policy guidance, raising concerns regarding the consistency of the Fed’s commitment to its inflation objective. Recent research has emphasised that such deviations may reflect asymmetric policy responses to perceived downside risks or attempts to incorporate the balance of inflation risks into a risk-adjusted Taylor rule framework.”

Presenting a recent working paper, Dr Jack Fosten, Reader in Data Analytics at Bayes, said that sparse” models, which reach conclusions based on fewer variables than their “dense” counterparts are often more interpretable. He outlined how he and his co-authors developed a Hausman-type statistical test to check the number of predictors relevant for predicting a range of key statistics at the macroeconomic level.

The paper, Sparsity Tests for High-Dimensional Time Series Regressions, assessed the number of variables that economists and others need to include when predicting key statistics such as inflation and industrial production. Economists work with a mountain of data but the paper makes the case for using stripped back sparse models which are easier to interpret.

Applying their new statistical test to key macroeconomic and financial data for the United States, the authors found strong evidence of sparsity: most variables (like inflation and financial indicators) depend on fewer than five predictors. However, 10-15 predictors were more appropriate for industrial production and other economic indicators, which is fewer than suggested by some previous studies.

Dr Fosten said: “The test provides a new way to formally check for sparsity and the empirical evidence suggests U.S. economic and financial forecasting models are often much sparser than previous research indicated.”