Today, we are pleased to present by Filippo Natoli and Fabrizio Venditi Director of the General Directorate of Economics, Statistics and Research of the Bank of Italy. The views expressed in this note are those of the author and do not necessarily reflect the views of Bank of Italy.
A pre-estimated negative reading of U.S. GDP growth in the second quarter surprised private forecasters, leading to recession talks. Using standard forecasting models, we show that an impending recession in the US and UK was highly probable even a month ago when high CPI inflation and tight labor markets are factored in.
The momentum of the global economic cycle deteriorates sharply in the second quarter of 2022 Commodity prices have been underpinned by the war in Ukraine as global trade remains constrained by persistent supply bottlenecks, while high inflation continues to erode consumers’ purchasing power. Meanwhile, financial conditions have tightened sharply as central banks responded aggressively to stubborn inflationary pressures. Against this backdrop, the U.S. economy experienced negative GDP growth for the second time in a row, following the decline in economic activity in the first quarter. The outlook for other advanced economies appears equally worrisome, notably energy importers whose terms of trade have deteriorated rapidly in the wake of Ukraine’s invasion. Forecasters have factored in these headwinds and lowered their global growth forecasts for 2022 and 2023 (see, for example, International Monetary Fund) and The problem seems to be when instead of if recession is imminent: Searches for the term “recession” have been on the rise on Google since March of last year (Figure 1).
Figure 1: Google Search Strength for the Term “Recession”
(Number of hits ranked between 0 and 100 during the time period considered). Source: Google Trends
in recent work (Natoli and Venditi, 2022), We contribute to this debate by jointly assessing the relevance of financial and macroeconomic factors in predicting recessions in the US and UK since the late 1990s. In our analysis, we rely on the standard probabilistic forecasting framework pioneered by Estrella and Hardouvelis (1991),
The dependent variable is a dummy variable equal to one (zero) if the economy was (not) in recession time + hours, H is the forecast range, x is collection of regressors, F(.) is the standard normal cumulative distribution function and is the error of a normal distribution.
We benchmark the slope of the government bond yield curve (the difference between 10-year and 3-month yields) as a unique predictor, benchmarking against standard norms in the literature. However, the ability of such minimalist models to predict recessions has recently been questioned in the literature (Kanizova and Lee, 2014; Ercolani and Natoli, 2020; Keeley, 2022Wait). Therefore, we first added measures of benchmark indicators of market stress—financial conditions and stock market volatility—because the slope of the yield curve alone may not fully capture the deteriorating funding conditions that led to the crisis.The Financial Conditions Index (FCI) is an unweighted average of 10-year yields, monthly stock returns, and corporate bond yield spreads, based on the Arrigoni et al. (2022), while stock market volatility is captured by the VIX. In the third specification, we also added two variables that summarise the macroeconomic environment – CPI inflation and unemployment. Figure 2 shows the coefficient estimates for the United States over different forecast ranges (1 to 12). They confirmed that financial indicators and macroeconomic conditions provide additional predictive power. In particular, they show that (i) a flat yield curve (ii) tight financial conditions (iii) high financial market uncertainty (iv) high inflation and (v) low unemployment (roughly similar to the current economic environment) The period is likely to be followed by a recession. Estimates for the UK are very similar.
Figure 2: Average Marginal Effects, US Model
When key financial and macroeconomic indicators enrich specifications based on the slope of the yield curve, forecasting performance improves significantly, and the probability of an impending recession jumps to values very close to 1 in the current environment. Figure 3 shows the time series of predicted recession probabilities (6 months ahead) for a model that relies only on the slope of the yield curve (blue line), intermediate specifications including VIX and FCI (green line) and also introduces The full model of inflation and unemployment (red line). The last model is the best performing model according to the standard measure of fit. Estimates for all models are from January 1998 to May 2022 – ie, already available around mid-June. Historically, financial conditions have proved to be more important than inflation and unemployment in predicting recessions in the early 2000s. Furthermore, they are very nervous ahead of the pandemic shock that will eventually devastate the global economy in 2020. Both financial and physical factors played a role in the Great Recession of 2008.On the other hand, at this juncture, the strongest recessionary signals come from record high inflation and a tight labor market, which Dovish and Summers (2022). An interesting observation is that the sample period includes years of anchored inflation expectations and sound monetary policy: our results suggest that even in such an environment, an aggressive monetary response — aggressive enough to generate a recession — is required. – to curb inflation. Taken together, our results suggest that a soft landing — engineering deflation without triggering a recession — is highly unlikely at this juncture. The actual combination of a hot labor market, high inflation, and tight financial conditions often accompanies recessions.
Figure 3: Probability of recession for the next six months, time series
Panel A
panel B
Note: Specified recession ranges are based on OECD recession indicators
Source: Natoli, F. and Venditti F. (2022). The role of financial and macroeconomic conditions in forecasting recessions (29 July 2022). Available at SSRN: https://ssrn.com/abstract=4176581
This article is by Filippo Natoli and Fabrizio Venditi.
