Todd Clark (Cleveland Fed)

Nowcasting -- predicting current quarter GDP growth -- is commonly viewed as an important and unique forecasting problem. Much of the nowcasting literature has focused on data available at a monthly and quarterly frequency. In 2020, the shutdown of significant portions of the economy to restrain the outbreak of the coronavirus has raised practical interest in high-frequency indicators of economic activity in the US and other economies. Apart from nowcasting considerations, a rapidly growing body of research has examined tail risks in macroeconomic outcomes, typically at a horizon of one quarter or one year ahead. Most of this work has focused on the risks of significant declines in GDP, and has relied on quantile regression methods to estimate tail risks.

Building on these strains of work, this paper assesses the ability of models to produce accurate tail risk nowcasts of GDP growth with a potentially wide array of information. We consider not only different models but also different methods for data reduction. Our starting point is a mixed frequency regression setup, in which, for nowcasting GDP growth within a quarter, each time series of monthly indicators is transformed into three quarterly time series, each containing observations for, respectively, the first, second, or third month of the quarter. At the moment in time that the forecast is formed, the model includes only the quarterly series without missing observations, which addresses the ragged edge of the data. Bayesian methods are used to estimate the model, which facilitates providing shrinkage on estimates of a model that can be quite large.  

Our paper makes three broad contributions. First, we extend the research literature's typical forecast calendar setup to use 15 different weeks as forecast origins for a quarter's nowcast, rather than a few months. This setup permits an assessment of the evolution of forecasts and tail risks with the week-by-week flow of information in the quarter. Second, we consider higher frequency data, making use of a number of indicators at a weekly frequency and not just monthly indicators. Finally, we examine nowcasts of tail risks to economic activity, focused on the 5 percent quantile. 

Our results show that, within some limits, more information improves the accuracy of tail risk forecasts. Forecast accuracy typically improves as time moves forward from week to week within a quarter, making additional data available. In a given week, large models often forecast as well as or better than small models. There is a clear benefit to adding a base set of financial indicators to the base set of macro indicators. Adding other weekly indicators of economic activity doesn't have much effect on forecast accuracy, either to help or harm. Among the models or estimation approaches we consider, our regression with stochastic volatility and our Bayesian quantile regression perform consistently, offering solid gains in tail risk forecast accuracy. Data reduction via factors and forecast averaging can also help forecast accuracy.

Eleonora Granziera (Norges Bank)

The recent shutdown of significant portions of the worldwide economy, in order to restrain the outbreak of the coronavirus, has triggered a global recession. The uncertain consequences of the rapid spread of the virus and the induced infection control measures have made it extremely challenging for forecasters and policymakers to quantify and assess the current and future outlook of the economy. This has raised a renewed interest in the search for reliable high-frequency indicators that can track the real economy in a timely matter.

In this paper, we document that debit card transaction data serve as an early and reliable indicator for household consumption in Norway. We use a novel data set covering all debit card transactions for Norwegian households to nowcast quarterly household consumption in Norway. These card payments data are free of sampling errors and are available without delay, and currently account for more than 35% of the total value of all household consumption expenditures. Therefore, they prove a valuable early indicator of household spending.

To account for mixed-frequency data, we estimate various mixed-data sampling (MIDAS) regressions using predictors sampled at monthly and weekly frequency. We evaluate both point and density forecasting performance over the sample 2011Q4-2020Q1. Our results show that MIDAS regressions with credit card transaction data improve both point and density forecast accuracy over competitive standard benchmark models that use alternative high-frequency predictors. Finally, we illustrate the benefits of using the card payments data by obtaining a nowcast of the first quarter of 2020, a quarter characterized by heightened uncertainty due to the COVID-19 pandemic. For this quarter the MIDAS model with BankAxept data successfully predicts the magnitude of the drop in consumption, while the other models forecast a much milder decline.

Matteo Mogliani (Banque de France)

Monetary policy decisions require a real-time assessment of different economic scenarios, including the most extreme ones. In this paper, we propose an extension of the Growth-at-Risk approach that assesses financial risks on GDP growth, by accounting for the high-frequency nature of financial conditions. In this respect, we propose a Mixed-Data Sampling (MIDAS) approach of the quantile regression of quarterly GDP growth rate on daily measures of financial conditions, using a recently developed Financial Conditions Index for the euro area. Overall, we show that financial risks on euro area economic growth can be monitored in real-time on a high-frequency basis using our approach. This high-frequency monitoring turns out to be useful for policy-makers, especially during crisis times. 

Kajal Lahiri (University at Albany, SUNY)

Given the balance budget rule, good revenue forecasts are critical for state policy makers to decide on their expenditure plans for the year. We develop a MIDAS model for revenue forecasting characterized by arbitrary patterns of missing data and machine learning. Since recessions and policy changes generate big errors in revenue forecasts, we pay special attention to various leading indicators for recessions and newly enacted tax policies. We forecast yearly tax revenues using monthly tax receipts and also two dynamic factors extracted from two large sets of monthly and quarterly indicators specific to the New York State and the U.S. economy. These three models are combined with optimally to generate monthly multi-period forecasts with bootstrapped uncertainty fan charts. To coincide with the budget process, our forecasts start at the 18-month horizon, and are continuously updated till the end of the fiscal year. The sample covers fiscal years 1986-2020, and data till 2007 is used in estimation to evaluate out-of-sample forecasts over 2008-2020. Our model successfully predicted the massive revenue shortfalls during the 2007-09 recession and also during FY 2019 following the Federal Tax Cuts and Jobs Act of 2017.

The pandemic caused by the novel coronavirus outbreak and the economic shut down has dramatically upended state tax revenue forecasts for FY 2021.With updated March data, the boosting forecasts for FY 2021 is revised nominally to -4.39% from an initial estimate of 2.86% with data up January. As new data will be released over the year, our model is uniquely poised to forecast future tax receipts in real time to aid appropriate expenditure decisions during the rest of the fiscal year.

Dalibor Stevanovic (Université du Québec à Montréal)

We consider simple methods to improve the growth nowcasts and forecasts obtained by mixed frequency MIDAS and UMIDAS models with a variety of indicators during the Covid-19 crisis and recovery period, such as combining forecasts across various specifications for the same model and/or across different models, extending the model specification by adding MA terms, enhancing the estimation method by taking a similarity approach, and adjusting the forecasts to put them back on track by a specific form of intercept correction. Among all these methods, adjusting the original nowcasts and forecasts by an amount similar to the nowcast and forecast errors made during the financial crisis and following recovery seems to produce the best results for the US, notwithstanding the different source and characteristics of the financial crisis. In particular, the adjusted growth nowcasts for 2020Q1 get closer to the actual value, and the adjusted forecasts based on alternative indicators become much more similar, all unfortunately indicating a much slower recovery than without adjustment and very persistent negative effects on trend growth. Similar findings emerge also for the other G7 countries.

Danilo Cascaldi-Garcia Danilo Cascaldi-Garcia (Federal Reserve Board)

We construct a model to provide GDP forecasts for the euro area and the three largest member countries---Germany, France and Italy. We show that our model improves significantly the accuracy of forecasts for all of these economies. Our model processes the large flow of data released continuously and asynchronously for these economies while accounting for their short-term heterogeneous dynamics. We evaluate the ability of our model to distill the large flow of data by conducting a real-time forecasting evaluation and by focusing on key specific episodes: the Global Financial Crisis, the European sovereign debt crisis, and the onset of the Great Lockdown.