Recessions aren’t fully predicted, to be sure. We’d be able to better plan for them or possibly avoid them if they were. However, there are a few warning indicators that economists may use to predict the onset of a recession. These signs are referred described as “leading indicators” by economists. There are other lagging indications that appear after a recession has started. The most notable lagging signal is a high unemployment rate.
An inverted yield curve is a prominent leading indicator. The link between the yields of a short-term government bond and a long-term government bond is known as an inverted yield curve. The long-term yield will be higher in normal circumstances. When the yield curve inverts and the long-term yield falls, it indicates a lack of confidence in the economy and the possibility of a recession. Since 1970, every recession in the United States has been preceded by an inverted yield curve.
Manufacturing job losses are another symptom of impending recession. Less demand for produced items can indicate lower consumer spending, so if companies lay off workers or stop employing new ones, it could signal job losses in other industries. Falling housing prices, a stock market correction, and a lack of new small enterprises are all leading indicators.
How do you know when a recession is approaching?
Real gross domestic product (GDP), or goods produced minus inflationary impacts, is the economic measure that most clearly identifies a recession. Income, employment, manufacturing, and wholesale retail sales are some of the other major indicators. Each of these areas suffers a drop during a recession.
Is there going to be a recession in 2021?
Unfortunately, a worldwide economic recession in 2021 appears to be a foregone conclusion. The coronavirus has already wreaked havoc on businesses and economies around the world, and experts predict that the devastation will only get worse. Fortunately, there are methods to prepare for a downturn in the economy: live within your means.
What are the three warning signals of an impending economic downturn?
What Are the Signs of a Coming Recession?
Falling asset prices, such as the cost of housing and stock market falls.
Which method accurately forecasts recessions?
Because it uses a wide variety of monthly indicators rather than a single quarterly indicator, the NBER method for dating recessions produces more realistic results than the GNP method. For predicting recessions, there are two separate econometric methods.
What is the most reliable predictor of a downturn?
In the past, indexes that integrate numerous macroeconomic variables have done a better job than other indicators at predicting recessions up to a year ahead of time. Economists monitor a variety of economic and financial data series to assess the current state of the economy and future possibilities. According to the National Bureau of Economic Research, leading indicators are indications of U.S. recessions (NBER). I look at how useful various economic and financial indicators have been in the past at “predicting” recessions and what these indications signal for the future. I show that indexes that integrate numerous macroeconomic variables have historically outperformed other indicators in terms of predicting recessions (and expansions) up to a year ahead of time. Furthermore, I confirm that financial market indicators, particularly the slope of the Treasury yield curve, have proven reliable predictors of recessions one to two years in advance. I also calculate recession prediction criteria for all of the leading signs I analyze using historical data. Then, to aggregate the data from these indicators, I create a new index that indicates the percentage of leading indicators that anticipate a recession at any particular period. This basic index exceeds existing indicators in predicting a recession six to nine months ahead of time.
How to evaluate leading indicators
In this study, I evaluate numerous leading indicators to see which ones have been more successful in the past at predicting recessions based on their historical classification abilities of data aligned with future recessions and expansions. I specifically assess a list of leading indicators compiled by the Conference Board from a range of sources. Data on employment, manufacturing activity, housing, consumer expectations, and stock market returns are among these indicators. However, I substitute the more commonly used difference between the ten-year yield and three-month yield (the long-term spread)3 and another version of the yield curve designed to capture monetary policy expectations for the Conference Board’s measure of differences in Treasury securities’ interest rates across maturities (or the slope of the yield curve) (the near-term forward spread). 4 I also substitute the Chicago Fed’s National Financial Conditions Index (NFCI) and its nonfinancial leverage subindex for the Conference Board’s measure of credit conditions. 5 I look at the Conference Board Leading Economic Index for the United States (the average of its list of indicators); the Brave-Butters-Kelley (BBK) Leading Index, which two collaborators and I recently created from a panel of 500 monthly time series and quarterly U.S. real gross domestic product growth;6 the University of Michigan’s Index of Consumer Expectations; and the value of debit balances in broker-dealers’ s accounts (GSCI). 7 The 17 indicators I evaluate have been normalized throughout this analysis, so that negative values reflect a decline in economic activity. 8
Finally, I’d like to be able to compare a specific observation for any of these indicators to historical levels and determine whether or not a recession is imminent. This means I’m looking for a threshold below which the indicator has always been while indicating a recession (or always above when signaling an expansion). These forecasts will inevitably be flawed, and there will be instances during a recession when an indicator exceeds the selected threshold (and times during an expansion when it is less than the threshold). The accuracy of an indicator refers to the total number of times it correctly defines a certain period based on a set of criteria. Unfortunately, for the purposes of recession prediction, accuracy is a problematic metric because successfully diagnosing a recession is viewed the same as correctly classifying an expansion. In the most extreme example, predicting that a recession will never happen is 88% accurate because recessions have only happened in 12% of all months since 1971. Obviously, having a predictor that delivers a meaningful signal about impending recessions, even if it is less than 88 percent accurate, would be desirable.
A statistic known as the area under the receiver operating characteristic (ROC) curve, or AUC value, is a superior criterion for evaluating these indicators.
9 An indicator’s categorization ability based on a pair of data points is measured by its AUC value. Assume we were given two indicators and informed that one is connected with a recession and the other is associated with an expansion. The AUC value represents the likelihood that the lower observation is linked to a recession. AUC values range from zero to one, as with any probability; a value of one indicates that an indicator perfectly classifies a random pair of observations. 10 Even if an indicator is unrelated to future recessions, it has a 50-50 chance of properly predicting one, resulting in an AUC of 0.5.
The imbalance in the number of recessionary vs expansionary periods observed has no effect on the AUC value because it is related to random pairs of observations. Furthermore, the AUC value may be computed without first deciding on a threshold (unlike accuracy), making it a more reliable measurement of how much information an indicator transmits about future economic conditions.
AUC values of leading indicators
I alter the indicators’ observations to correspond with whether or not a recession happened a specific number of months in the future up to two years ahead of time to evaluate each indicator’s AUC value. The results are shown in figure 1 as colored lines, with composite indexes and Treasury yield curve measures (those with generally larger AUC values) in panel A and the remaining measures in panel B. Based on these findings, I’ve come to the following conclusion:
- The Conference Board Leading Economic Index for the United States is the best at predicting recessions and expansions up to nine months ahead of time. I reject the idea that other indicators are similarly good at predicting a recession one to six months ahead of time, based on a statistical test11. The Conference Board’s leading index remains the strongest predictor for seven to nine months ahead, but I can’t rule out the possibility that three other indicators are just as good (the BBK Leading Index and the two yield curve measures). In the short term, the Conference Board’s leading index is extremely accurate, with an AUC value of 0.97 one to three months ahead.
- The long-term Treasury yield spread (i.e., ten-year minus three-month Treasury yields) is the best predictor of a recession or expansion far in advance. At a 16 to 20-month horizon, I can rule out the possibility that alternative indications are just as excellent. The long-term yield curve slope remains the best predictor for 14 to 15 and 21 to 24 months ahead, but I can’t rule out the possibility that at least one of three other indicators (the NFCI’s nonfinancial leverage subindex, the S&P GSCI, and the University of Michigan’s Index of Consumer Expectations) is just as good. The AUC values are lower than those for short horizons because to the added uncertainty that comes with longer horizon predictions: At 14 months ahead, the long-term yield spread reaches an AUC of 0.89, then steadily drops to 0.75 at 24 months ahead.
- Several leading indicators generate similar AUC values ten to thirteen months ahead. At these horizons, the AUC values of the Conference Board Leading Economic Index for the United States, the BBK Leading Index, the two yield curve slopes, and the NFCI’s nonfinancial leverage subindex are all between 0.84 and 0.89. Statistical tests are ambiguous as to which one performs best at these horizons, emphasizing that both should be taken into account for forecasting medium-term recessions.
- The Conference Board’s leading index and the BBK Leading Index in panel A perform a better job of predicting recessions than the macroeconomic indicators in panel B, as seen in figure 1. Panel B’s macroeconomic indicators perform so poorly over longer time horizons that I can’t rule out the possibility that many of them are similar to random noise more than a year ahead. The AUC values of these leading indices approach 0.5 over extended time periods as well, but take longer than the macroeconomic indicators. These findings suggest that the indexes are operating as expected: they provide a clearer indication of future economic activity by reducing noise in their component indicators.
Recession prediction thresholds
While the AUC value provides information about a leading indicator’s ability to classify data in the past, it does not provide information about the threshold that should be utilized to anticipate a recession. The earlier problem, which I correctly identified, shows that a different strategy is required. To figure out which option is best, keep in mind that the threshold for each indicator has two effects: 1) the true positive rate, or how many months it correctly labels as a recession, and 2) the false positive rate, or how many months it incorrectly classifies as an expansion.
My goals for these two indicators are in conflict with each other regardless of the threshold I pick. I want to predict as many recessions as possible (a high true positive rate), but I also want to have as few cases as possible where the indicators are wrong “yell “wolf” (avoiding a high false positive rate). If I wanted to ensure that an indicator predicted every probable recession, I’d pick a high threshold to build a sensitive predictor with a high true positive rate at the cost of many false recession forecasts. In contrast, if I wanted to be sure that a recession was coming when an indicator predicted one, I would set a low threshold so that only the lowest values of the indicator predicted one; while this approach would miss some recessions, it would give me more confidence that one was coming when one was predicted.
Consider the scenario of an indicator that provides no information about a looming recession to resolve this problem. Whatever criterion is chosen, it merely alters the percentage of time when a recession is expected. Assume this random guess correctly forecasts a recession 20% of the time. This prediction would correctly predict 20% of recessions when the results are known whether a recession occurred or not. This assumption, on the other hand, would falsely forecast a recession 20% of the time when an expansion happened. This indicates that the genuine positive rate and the false positive rate will always be the same for such an indicator. The more informative a threshold indication is, the more it will deviate from this connection. Choosing a threshold that maximizes the difference between true positive and false positive rates gives you the most information about previous recessions for a given indicator. 12
Let me give you an illustration of what this threshold criterion means in terms of a single indicator: This is a good example “For the long-term Treasury yield spread (i.e., ten-year minus three-month Treasury yields) at 12 months ahead, the “maximum information” criterion is slightly higher than the frequently stated value of zero. Only 57 percent of recession months and 5% of expansion months are accurately classified using the zero threshold (also known as a yield curve inversion13). The highest information threshold fluctuates slightly among the horizons studied, but remains constant at 0.94 for the next eight to fifteen months. The long-term spread one year ahead correctly diagnoses 88 percent of recession months, but erroneously classifies 19 percent of boom months, according to this criteria.
The circumstance determines which of these levels to use. The lower false positive rate of the zero threshold is appealing to individuals who want to be more convinced that a recession is approaching when one is forecast. Instead, the maximum information strategy focuses on the true positive/false positive rate trade-off. The maximum information technique raises both rates by raising the threshold, but the true positive rate rises faster than the false positive rate, making it easier to discern prior recessions from expansions. 14
A summary index
To determine an optimal recession prediction threshold, the maximum information threshold criterion can be applied to each of the indicators at each horizon from zero to 24 months ahead. I calculate the fraction of the 17 indicators that are below their ideal threshold and anticipate a recession to present all of the indications under consideration as succinctly as feasible. This is, in effect, a new approach of generating a leading index to forecast future recessions. This “ROC threshold index” is notable in that it is merely the fraction of the indicators analyzed that have passed their recession prediction thresholds, rather than an estimated chance of a recession. I computed the AUC values for each of the 25 ROC threshold indexes at the corresponding horizon to evaluate them, and the results are represented as the black line in panel A of figure 1.
The ROC threshold indexes are better predictors of approaching recessions than any of the other variables studied throughout time horizons of up to 11 months.
15 Using the same statistical test as before, I can rule out the possibility that any of the indicators considered here are as good as these indexes over a six- to nine-month timeframe. The predictive power of the ROC threshold indexes falls below that of the yield curve measurements over longer time horizons, but they remain moderately helpful. Intuitively, the ROC threshold indexes’ performance deteriorates as the predictive power of the leading indicators used to create them deteriorates. Because just a few indicators are substantially predictive more than a year in advance, the ability of the ROC threshold indexes to distinguish between recessions and expansions deteriorates when the prediction is made longer in advance.
Nine months ahead of schedule, the ROC threshold indices greatly surpass all other indicators. Figure 2 shows the ROC threshold index time series at this horizon, with the series pushed nine months ahead to exhibit the most recent data observation from August 2019 in May 2020. Because the goal isn’t necessary to extract as much information as possible, determining the right threshold to assess this index against is difficult (as it was with the individual indicators to construct the index). Using the maximum information strategy, a 50% threshold is obtained. This indicator properly forecast a recession in 83 percent of recession months based on the 50% threshold, but mistakenly projected a recession in 15% of expansion months. A generally used, more conservative criterion16, on the other hand, yields an 80 percent barrier. The genuine positive rate of the 80 percent threshold is 26%, whereas the false positive rate is only 3%. The decision between these criteria, as before, is determined by the purpose of the forecast. The 50 percent barrier is better if one is ready to accept a higher chance of misclassifying an expansion; the 80 percent level is better if it is more vital to be highly convinced that a forecasted recession is genuinely coming. Both criteria are presented in figure 2 since they are possibly beneficial.
ROC threshold index at nine months ahead
While the ROC threshold index for the next nine months rose beyond 50% based on data collected in December 2018 (plotted in September 2019 in figure 2), it has remained close to but below 50% for all data collected since then. Since around the end of the previous recession, this indicator has been significantly below the 80% mark. Given how volatile this indicator is, these slightly higher recent readings are worth noting, but it remains below the 50% level that is nearly invariably connected with a historical recession.
To be clear, the entire study is based on the assumption that when data is observed, it is known with confidence. This is obviously not the case, as data is released slowly and frequently changed months later. To better understand our abilities to foresee recessions before they happen, we need to do a real-time analysis of this technique.
Conclusion
The findings of this article reveal that the long-term Treasury yield spread has historically been the most accurate available “predictor” of recessions for timeframes of one year and longer. However, leading indexes have done a better job of predicting recessions in the short term than individual leading indicators or financial data. Because they are basically leading indexes that combine the information in the inputs to create a more precise evaluation of coming economic activity, the ROC threshold indexes constructed here have also performed well as recession predictions in the near term.
What will the state of the economy be in 2022?
“GDP growth is expected to drop to a rather robust 2.2 percent percent (annualized) in Q1 2022, according to the Conference Board,” he noted. “Nonetheless, we expect the US economy to grow at a healthy 3.5 percent in 2022, substantially above the pre-pandemic trend rate.”
How long do economic downturns last?
A recession is a long-term economic downturn that affects a large number of people. A depression is a longer-term, more severe slump. Since 1854, there have been 33 recessions. 1 Recessions have lasted an average of 11 months since 1945.
Is a recession expected in 2023?
Rising oil prices and other consequences of Russia’s invasion of Ukraine, according to Goldman Sachs, will cut US GDP this year, and the probability of a recession in 2023 has increased to 20% to 30%.
Do things get less expensive during a recession?
Lower aggregate demand during a recession means that businesses reduce production and sell fewer units. Wages account for the majority of most businesses’ costs, accounting for over 70% of total expenses.