Todd E. Clark’s research while affiliated with Federal Reserve Bank of Cleveland and other places

... The paper also contributes to an active literature on forecast disagreement in macroeconomics and finance. This literature considers measures of disagreement to either measure economic conditions (as in Zarnowitz andLambros 1987, Lahiri andSheng 2010 and the studies surveyed by Clements et al. 2023 andClark andMertens 2024), or to test economic theories (as in Coibion and Gorodnichenko 2012, Dovern 2015and Andrade et al. 2016. The characterization derived in this paper suggests principled measures of forecast disagreement for various applied settings. ...

Reference:

A Kernel Score Perspective on Forecast Disagreement and the Linear Pool

... The first contribution of this paper is the application of ML methods to identify key financial predictors of growth vulnerabilities as measured by industrial production (IP) growth. Our paper is hence related to recent work by Prüser and Huber (2024) and Carriero et al. (2024), but our focus differs in at least three ways. First, we focus on the 0.05 quantile as this is the threshold commonly used in policy analyses, whereas Carriero et al. (2024) center their analysis on the 0.1 quantile. ...

Reference:

Machine-learning Growth at Risk

... The focus on density predictions, and especially tail forecast metrics, is due to the recent emphasis on predicting macroeconomic risk in the wake of Adrian et al. (2019). The most popular targeted objects, often with single-equation variants of quantile regression (QR), are the quantiles of GDP growth, which are commonly referred to as Growth-at-Risk (see, e.g., Adrian et al., 2022;Clark et al., 2024b;Plagborg-Møller et al., 2020). But there are other econometric approaches that have been shown to perform equally well or better than QR (see Carriero, Clark and Marcellino, 2022;Clark et al., 2023;Delle Monache, De Polis and Petrella, 2023, among others), and many papers have considered a wider set of economic indicators, such as inflation and unemployment (see, e.g., Adams et al., 2021;Galbraith and van Norden, 2019;Manzan, 2015;Pfarrhofer, 2022, among others). ...

Reference:

Predicting Tail-Risks for the Italian Economy

... Analysis further addresses scenario information beyond a single point forecast, specifically to use scenario tail percentiles that reflect measures of scenario risk. This links to the desirability of scenario hypotheses that represent more radical perturbations of the baseline than has been typical (e.g., Justiniano and Primiceri, 2008;Fernández-Villaverde et al., 2011;Adrian and Boyarchenko, 2012;He and Krishnamurthy, 2012;Brunnermeier and Sannikov, 2014;Fernández-Villaverde et al., 2023 with structural models, and Adrian et al., 2021;Caldara et al., 2021;Carriero et al., 2024 with reduced-form models). That scenarios considered by policy institutions often represent only modest perturbations of the baseline is also partly addressed by our use of the synthetic backstop scenario. ...

Reference:

Scenario Synthesis and Macroeconomic Risk

... One caveat is the identifiability issue of the margins, which could hinder sampling and interpretation. Second, developing techniques for obtaining convergent margin Markov chains can potentially lower the uncertainty in the alternative DICs specified in Section 4. Third, heteroskedasticity plays a critical role in VAR applications within econometrics (Clark and Mertens, 2023), suggesting that adding heteroskedasticity such as stochastic volatility to TVP-TVARs is worth investigating. ...

Reference:

Time-varying Parameter Tensor Vector Autoregression

... These variables were commonly used in studies on inflation. For example, global oil price was used in studies by Garzón and Hierro (2022) (2022); the global supply chain pressure index was used in studies by Diaz et al. (2024), and Gordon and Clark (2023); and unemployment rate was used in studies by Hall et al. (2023) and Yilmazkuday (2022). According to Chen et al. (2016), the SVAR model has advantages in analyzing dynamic relationships among relevant time sequence variables. ...

Reference:

Empirical Investigation of the Sources of Inflation in Sri Lanka: Assessing the Roles of Global and Domestic Drivers

... The most popular targeted objects, often with single-equation variants of quantile regression (QR), are the quantiles of GDP growth, which are commonly referred to as Growth-at-Risk (see, e.g., Adrian et al., 2022;Clark et al., 2024b;Plagborg-Møller et al., 2020). But there are other econometric approaches that have been shown to perform equally well or better than QR (see Carriero, Clark and Marcellino, 2022;Clark et al., 2023;Delle Monache, De Polis and Petrella, 2023, among others), and many papers have considered a wider set of economic indicators, such as inflation and unemployment (see, e.g., Adams et al., 2021;Galbraith and van Norden, 2019;Manzan, 2015;Pfarrhofer, 2022, among others). ...

Reference:

Predicting Tail-Risks for the Italian Economy

... of forecasters influence the obtained results (e.g. Glas and Hartmann, 2022;Pavlova, 2024 Ganics et al. (2022) and Krüger and Plett (2024) indicate that forecast distributions based on past point forecast errors perform similar to, or better than, subjective histogram-type forecasts as provided by the SPF and ECB-SPF. These findings motivate our use of the IMF as an external point forecast, together with a suitable set of historical forecast errors. ...

Reference:

Simple Macroeconomic Forecast Distributions for the G7 Economies

... However, most studies only consider low-dimensional quantile regression models estimated by frequentist methods. Meanwhile, studies that use high-dimensional Bayesian quantile regressions, such as Plagborg-Møller et al. (2020), Carriero et al. (2022b) and Prüser and Huber (2023), are mostly interested in macroeconomic variables such as gross domestic product (GDP), inflation, and unemployment. ...

Reference:

Variable selection in macroeconomic stress test: a Bayesian quantile regression approach

... In this context, shrinkage describes the penalization of numerous parameters toward zero or toward a simpler structure, so as to enforce parsimony. For VAR models, shrinkage is particularly needed when the number of time series variables (d) or the chosen lag order (p) is large relative to the available sample size, a common scenario in macroeconomic forecasting (Bańbura et al., 2010;Bai et al., 2022). ...

Reference:

Bayesian Shrinkage in High-Dimensional VAR Models: A Comparative Study