Outlook-at-Risk provides an objective way of quantifying the risks around future real GDP growth, the unemployment rate, and inflation by capturing how such risks evolve as financial conditions ease or tighten.

Adrian, Boyarchenko, and Giannone (2019) showed that risks to real activity evolve as economy-wide financial conditions change, with the probability of a decline in real activity rising as financial conditions tighten. This negative relationship between real activity and financial conditions is strongest when the chances of a pronounced decline in real activity are high—in other words, when the economy is “at risk.” Following a similar approach, Adams, Adrian, Boyarchenko, and Giannone (2021) showed that the risk that the unemployment rate rises in the future or that the rate of inflation rises or falls from its current level is also presaged by tight financial conditions.

In the spirit of this work, we follow a similar methodological approach and estimate the conditional distribution of future real gross domestic product (GDP) growth, unemployment rate, and consumer price index (CPI) inflation based on contemporaneous financial conditions for each month since January 1989.

The world is uncertain and so quantitative approaches to measuring the changing risks to the outlook are a valuable tool. Although there is extensive work and numerous existing resources available that endeavor to characterize the likely path of the economy, relatively little work has studied the entire distribution of future economic outcomes. Outlook-at-Risk provides a useful complement focusing on other aspects of the distribution, such as economic outcomes that are less likely (often referred to as the “tails” of a distribution).

There are two key ingredients to our approach. First, we use near-term forecasts (one quarter to four quarters ahead) for our economic variables of interest: real GDP growth, the unemployment rate, and CPI inflation. We obtain monthly forecasts from the Blue Chip Economic Indicators Survey (BCEI). These data are proprietary but are available for purchase. For the realized variables, we use the third release of real output growth available from the Federal Reserve Bank of Philadelphia Real-Time Data (available here) corresponding to real gross national product (GNP) growth prior to 1992 and real GDP growth thereafter. We also use the civilian unemployment rate (U-3 unemployment rate) and the consumer price index for all urban consumers, both available from the Bureau of Labor Statistics. Second, we use a measure of overall U.S. financial conditions, the Composite Indicator of Systemic Stress (CISS), which is produced by the European Central Bank and is available here. Importantly, when we estimate each distribution, we aim to utilize only information that would have been available at the time. This avoids “look ahead” bias, which could lead to misleading inferences.

In our main tabs above we present the current estimates of the conditional and unconditional distributions for each of the three economic variables. We also present the time series of the 10th, 25th, 50th, 75th, and 90th percent quantiles (or percentiles). A percentile provides information about how likely events are to occur above or below a certain value. For example, the 10th percentile of the distribution of future real GDP growth is the rate of growth such that there is an estimated 10 percent probability of future growth being less than or equal to it. To give a specific example, in the second chart in this blog post, the shaded area represents the probability that average real GDP growth over the next four quarters is below -0.1 percent, which, in November 2019, was assessed to be 10 percent. Thus, -0.1 percent is the 10th percent quantile (or percentile) of the conditional distribution of average real GDP growth over the following four quarters. A more negative (lower) estimate of the 10th percentile indicates greater downside risks to real GDP growth.

Because we estimate the entire distribution of future outcomes, we can study a number of important features about the outlook. For example, the most likely value is referred to as the mode of the distribution and it is the value on the x-axis corresponding to the highest point in the distribution. As another example, the difference between a high and low percentile, such as the difference between the 90th and 10th percentile, is a measure of uncertainty, with a longer gap indicating that a wider range of possible future values are likely. Measures such as this are appropriate for CPI inflation and indicate the risks of a wider range of future inflation occurring. For real GDP growth, where we might be concerned about reduced activity, percentiles in the left tail (such as the 10th percentile) provide information about the risks of future anemic growth or contractions. Similarly, for the unemployment rate, where we might be concerned about deterioration in the labor market, percentiles in the right tail (such as the 90th percentile) provide information about the risks of such a slowdown.

To obtain the estimated distributions, we rely on so-called quantile regressions, which allow us to assess how certain features of the distribution of future macroeconomic variables (in this case, particular quantiles) relate to the level of financial conditions. Suppose we are interested in the 25th percent quantile of future real GDP growth. Quantile regressions allow us to estimate how the 25th percent quantile of our desired distribution co-moves with financial conditions. Repeating this exercise for different quantile choices begins to inform us on the overall shape of the conditional distribution of real GDP growth. To estimate the full distribution, we then choose the best fit from a large family of candidate distributions to the predicted 10th, 25th, 75th, and 90th percent quantiles.

For real GDP growth and CPI inflation we consider risks around the average quarterly growth rate over the next four quarters; however, for the unemployment rate we consider risks around the quarterly average in four quarters’ time. Why the difference? The time series of real GDP growth and CPI inflation are subject to considerable quarter-to-quarter transitory variability that can be “smoothed out” by taking averages over multiple quarters. In contrast, the unemployment rate has little to no transitory variability, and so taking averages over multiple quarters is unnecessary.

Selected percentiles (P10, P25, P75, P90) of the forecast error distribution for each series are available for download.