Comment 71: What is a "recession"? Are there any mechanical forewarning signals about incoming recessions? (Aug. 27, 2009)

The dreaded word "recession" now is in the mind of many, not just investors and Wall Street. Naturally the question whether there are any forewarning signs for the incoming recession looms large. Actually there have been quite a few forewarnings about this recession from various sources, including this website, but those forewarnings were ignored by many main stream economists, major news media, Wall Street and policy makers alike so that the general public was totally blindsided as the global economy has plunged into this deep recession. The above mentioned forewarnings are not mechanical in nature, but are based on the analysis and intuition of various students of macro economics. The mechanical forewarning signals in our mind is a set of predefined rules that will spell out warning signals automatically when certain conditions are met. The purpose of this comment is to introduce readers to such a mechanical forewarning signal system that has successfully forewarned the arrival of this recession. The mechanical forewarning signals, if available, will not only improve the accuracy of projections of the incoming recessions, but will also advance our understanding about recessions in general.

Before discussing about mechanical forewarning signals of incoming recessions, we need to ask ourselves what a recession really is. Many will say that a recession starts from the quarter when real GDP (means "inflation adjusted Gross Domestic Production") registers a negative growth quarter and then followed by another negative growth quarter immediately. The recession ends when real GDP experiences a positive growth quarter after the plunge into the recession. This definition of recessions based on two consecutive quarters of negative growth in real GDP is heavily promoted by news media and Wall Street so let us call this definition the popular definition of recessions. Unfortunately this popular definition of recessions has serious shortcomings. For example, during the economic downturn of late 2000 to 2002 following the burst of the dot-com bubble, 2.3 million jobs were lost, and DOW (Dow Jones Industrial Average) lost 30% of its peak value, whereas NASDAQ index, that is the home of high tech and dot-com companies, lost 78% of its peak value. As has been pointed out in Comment 70, the economic pain inflicted by the downturn was as severe as many other downturns that were classified as recessions by the popular approach. Nevertheless, according to the popular definition the late 2000 to 2002 economic downturn was not a recession since there were no two consecutive negative growth quarters of real GDP during the period of concern. Furthermore, there are some recessions that the growth rate of real GDP oscillated around zero for a while at the tail end of the recessions. By requiring only one positive growth quarter of real GDP to declare the end of those recessions, the general public was puzzled in those cases since the economic pain was still continuing but the recessions were declared prematurely to have ended. Apparently vast improvements on the popular definition of recessions are needed to make it more credible and less erratic so that it can be used to search for mechanical forewarning signals of incoming recessions.

In contrast to the popular definition of recessions, there is the determination of peaks and troughs of business cycles by The Business Cycle Dating Committee of National Bureau of Economic Research (abbreviated as NBER). Right after the peak of a business cycle until the succeeding trough, the economy is in the phase of contraction, so this period can easily be identified as a recession, though NBER deliberately avoids the usage of the word "recession". The Business Cycle Dating Committee is composed of seven prominent economists, and the determinations of peaks and troughs are made by the majority vote among those seven economists. Each economist is allowed to use whatever measures and methods to reach his/her conclusions about the peaks and troughs. Let us use the current recession as the example to see how the determination of NBER compares with the popular definition. NBER has declared that the peak of this business cycle was reached in 2007Q4 (2007Q4 stands for the fourth quarter of 2007, this type of abbreviation will be used throughout this comment for convenience). This means that this recession has started in 2008Q1. The peak of this business cycle in 2007Q4 coincided with the peak of DOW and the top of nonfarm payroll number. Declaration of NBER about the peak has been made in December 1, 2008, fully one year after the actual peak. Without any doubt the peak of DOW and the top of nonfarm payroll number in 2007Q4 have weighed heavily in the determination of NBER. Inflation adjusted Personal Consumption Expenditure, called real Personal Consumption Expenditure and abbreviated as real PCE, provides another interesting aspect about the determination by NBER. In the graph at right monthly values of real PCE before the annual revision of 2009, which has been published by Bureau of Economic Analysis on August 5, 2009, are plotted as the red curve, and the monthly values with and after the annual revision of 2009 are plotted as the green curve. At the time of NBER determination of the peak at 2007Q4, that is, on December 1, 2008, only the red curve in the graph was available. The red curve shows that the peak of real PCE was reached in 2008Q2 but not at 2007Q4. NBER economists apparently trusted the peak of DOW and the top of nonfarm payroll number more than real PCE. The green curve of real PCE with the annual revision of 2009 now shows that the peak of real PCE was indeed reached in 2007Q4, a victory for the economists. To conclude the evaluation of NBER decision, we need to look at real GDP itself. In the second graph shown at left quarterly values of real GDP before and after the 2009 annual revision are shown as the purple and green curves respectively. Both curves peaked at 2008Q2. The annual revision of 2009 did brought down the height of the peak of 2008Q2 somewhat, but not enough to change the position of the peak away from 2008Q2 in spite of reduced real PCE. A close inspection of the strong real GDP in 2008Q2 reveals that the biggest contributor to the strength was the net export. A foreign buying panic of American commodities tied to the skyrocketing commodity price in 2008Q2 lifted real exports of goods, whereas the runaway crude oil price at that time reduced the domestic consumption and thus the quantity of imported crude oil. This was certainly just a temporary aberration and the economists at NBER were not swayed by the peak of real GDP in 2008Q2. On the other hand the popular definition of recessions was thrown off the track by the strong rise of real GDP in 2008Q2, and needed to wait for two consecutive negative growth quarters of real GDP in 2008Q3 and 2008Q4 to declare the start of this recession at 2008Q3, two quarters out of synchronization with the real economy. Since the real GDP data of 2008Q4 was only available at the end of January of 2009, the declaration of the start of this recession from the popular approach was almost two months behind the declaration of NBER.

The declaration of NBER with regard to the peak of this business cycle, though was accurate as discussed in the previous paragraph, is not mechanical in nature. People can only wait patiently for the sporadic declarations from the organization. Certainly we cannot derive any forewarning signals of incoming recessions based on NBER declarations. The popular approach is mechanical in nature and is based on real GDP, supposedly the most broad economic indicator available, but failed badly in determining the start of this recession. The trouble of the popular approach is its too close association with the quarterly growth rate of real GDP that can jump erratically at crucial times. Quarter-to-quarter % changes of real GDP multiplied by a factor of 4 in order to mold it as the rate of annual change, are plotted as the blue-bar curve in the lower portion of the second graph. It was the sticking out peak at 2008Q2 in the blue bar-curve that threw the popular approach into disarray as discussed already. Earlier shortcomings of the popular approach as have been pointed out before, they are, the failure to classify the late 2000-2002 economic downturn as a recession, and the arbitrary cutting short of some recessions are also caused by too close association with the jumpy quarter-to-quarter comparison of real GDP. Thus the way to improve the real GDP based popular approach becomes obvious, that is, the jumpy quarter-to-quarter % changes of real GDP must be smoothed out so that the underlying trends of real GDP can be extracted and the forewarning signals of incoming recessions can be searched.

There are two widely used smooth out methods for quarterly data like real GDP. One is "4-quarter moving average" of quarter-to-quarter % changes of real GDP, and the other is year-to-year % comparison. Those two methods have added advantage of minimizing the chance of inadequate seasonal adjustments. The 4-quarter moving average method is used in Comment 64 to study bear market bottoms. In article 12 it has been pointed out that two methods yielded only insignificant differences as long as real GDP is concerned, so easier to calculate year-to-year % change method is adopted. Let us take 2009Q2 real GDP data to illustrate how year-to-year % comparison works. Compared to 2008Q2 real GDP, real GDP of 2009Q2 has declined by 3.9%. Since year-to-year comparison is looking at the average condition of the past year, this -3.9% number does not reflect the economic condition of 2009Q2. The best way is to assign this -3.9% as the condition of the quarter just in the middle between 2008Q2 and 2009Q2, that is, 2008Q4. Thus a new index of -3.9% describing the average economic condition of 2008Q4 is constructed. In article 12 no name was assigned to this new index. Here for convenience of reference we will call it "Retarded by 2 Quarters Real GDP Index", and abbreviated as R2QrealGDP index. In the lower portion of the second graph, R2QrealGDP index is plotted as the red bar-curve to show how the blue quarter-to-quarter % change curve has been smoothed out. We should note that R2QrealGDP index stops at 2008Q4 in the graph because 2009Q2 real GDP data only allows us to calculate R2QrealGDP index up to 2008Q4. This loss of timeliness is the price we must pay in order to smooth out the jumpy blue bar-curve.

In article 12 it is defined that when R2QrealGDP index falls below +1.0%, a recession starts, and when the index rises above +1.0% the recession ends. Within a recession, the lowest point of the index is defined as the nadir of the recession. The nadir of a recession is the point where the speed of economic contraction reaches its maximum. The nadir of a recession has an important implication for the stock market. Whether in the same quarter as the nadir or just one quarter after the nadir of a recession, stock market always rallied strongly during the past recessions since 1948. The timing of the stock market rally seems not related to the timing of the exact end of the recession as has been observed in Comment 64. Of course, whether those powerful rallies were the first legs of new bull markets or just a powerful bear market rally all depended on the strength of the ensuing recovery as has been pointed out in Comment 70. Based on this observation, we believe that the current powerful stock market rally started in March of 2009 implies the nadir of this recession been reached either in 2008Q4 or in 2009Q1 already.

Mechanical forewarning signals of incoming recessions are also uncovered in article 12 based on R2QrealGDP index. It turns out that the drop of the index below +2.0% has always been associated with a recession. The way for the index to drop below +2.0% can be sorted into the following three categories:

Category 1: R2QrealGDP index drops through both +2.0% line and +1.0% line into a recession simultaneously.

Category 2: R2QrealGDP index drops below +2.0% but above +1.0%. Then in the following quarter the index falls below +1.0% into a recession.

Category 3: R2QrealGDP index falls below +2.0%, but stay above +1.0% for quite a while, sometimes even bounce back above +2.0%. Within six to seven quarters the index falls below +1.0% into a recession.

In the case of Category 1, no forewarning signal is issued since when the index drops below +2.0%, the economy is already in the recession. In the case of Category 2, we should be aware that the actual time to obtain a value of the index is three quarters later than the indicated time of the index. For example, suppose the index drops below +2.0% in Q1, and then drops further below +1.0% into a recession in Q2. Since to calculate the index at Q1 requires real GDP data of Q3 that is not available until Q4, the actual time we get the signal that the index has fallen below +2.0% in Q1 is in Q4, but by that time the economy has been in the recession for more than two quarters already. However, a Category 2 signal does speed up the confirmation of the recession by one quarter. In the example considered above, the confirmation that the index has fallen below +1.0% into a recession can only be made in Q1 of the next year when Q4 real GDP data becomes available, but the confirmation of the recession is pushed up to Q4 of the same year by using the Category 2 signal.

Category 3 type signals are genuine forewarning signals of the incoming recessions. When those signals are issued, usually economic conditions looks like robust so few will heed to the forewarning signal. Let us use the entrance into this recession as an example. The red R2QrealGDP index curve in the second graph dropped below +1.0% in 2008Q1 (labeled as "B" in the graph) so this recession has started in that quarter, the same conclusion as the declaration of NBER. Since the R2QrealGDP index at 2008Q1 can be calculated only when 2008Q3 real GDP data became available at the end of October of 2008, the confirmation that this recession has started in 2008Q1 was made at that time. This confirmation from the R2QrealGDP index system is about one month ahead of the declaration from NBER and three months before the call of this recession from the popular method. The R2QrealGDP index of 2007Q3 (labeled as "A" in the second graph) had dropped below +2.0%. The R2Qreal GDP index of 2007Q3 was calculated by using real GDP data of 2008Q2 that became availabel at the end of July of 2008. By the end of July of 2008 nonfarm payroll number had been declining for six months and DOW had lost 20% of its peak value already. Thus the forewarning signal issued at the end of July of 2008 was a Category 2 signal, and we could conclude that this recession had started in 2008Q1; the confirmation of this recession is thus pushed up by five months ahead of the declaration of NBER and seven months ahead of the popular approach if the R2QrealGDP index system is used.

The most important forewarning signal issued by the R2QrealGDP index system is a Category 3 signal at 2006Q3, labeled as "S" in the second graph. The index dropped below +2.0%, stayed below +2.0% for another quarter, and then rose above +2.0%. However, six quarters later the economy plunged into this recession starting from 2008Q1. The calculation of R2QrealGDP index of 2006Q3 required the real GDP data of 2007Q1 that became available at the end of April of 2007. At the end of April of 2007, DOW, nonfarm payroll number, real consumer spending were all rising, and the first financial firestorm of August of 2007 was still more than three months away. Nevertheless this mechanical forewarning signal was issued by the R2QrealGDP index system. This warning signal was even earlier than the "recession watch" issued on this website at July 28, 2007 which was based on the noticeable slow down of the trend of consumer spending painted by the annual revision of 2007, and also based on the wane of the runaway trade deficit. Of course, the whole R2QrealGDP index system was first introduced in article 12 that was posted on this website at January 20, 2009, so the forewarning signal of late April of 2007 was actually had not been discovered at that time yet. Only forewarning sign that beat this early warning signal from the R2QrealGDP index was the analysis purely based on the correlation of the flow and ebb of runaway trade deficit with the inflation and the burst of economic bubbles. Here we just quote the last sentence of Comment 25 posted on this website at October 18, 2005;

", so if an economic recession should come, it will be in 2008, not within the next 12 months, unless an unforeseen non-economic event, for example, a bird-flu pandemic, suddenly emerges."

Since 1948 only three Category 3 mechanical forewarning signals have been issued by this R2QrealGDP index system. The most recent one is related to this recession as has been discussed thoroughly in the previous paragraphs. Other two signals were the ones associated with the 1953-1954 recession and the 1957-1958 recession respectively. Anyway if a Category 3 mechanical forewarning signal is issued by the R2QrealGDP index system, we better heed to its warning. It will become the life savor for many investment portfolios. In concluding this comment, a table listing the track records of three types of definitions of recessions as discussed here is enclosed.

Comparison of recessions in three types of definitions
\ R2QrealGDP index NBER Popular definition
No. start end nadir duration signal available start end duration start end duration
1 1948Q4 1949Q3 1949Q2 3Q 1949Q2 1949Q1 1950Q1 4Q 1949Q1 1949Q3 2Q
2 1953Q2 1954Q2 1953Q3 4Q 1952Q4 1953Q3 1954Q3 4Q 1953Q3 1954Q2 3Q
3 1956Q1 1956Q2 1956Q1 1Q none none none none none none none
4 1957Q2 1958Q2 1957Q3 4Q 1956Q4 1957Q4 1958Q3 3Q 1957Q4 1958Q2 2Q
5 1960Q2 1960Q4 1960Q3 2Q none 1960Q3 1961Q2 3Q none none none
6 1969Q3 1970Q3 1970Q2 4Q none 1970Q1 1971Q1 4Q 1969Q4 1970Q2 2Q
7 1973Q4 1975Q2 1974Q3 6Q 1974Q2 1974Q1 1975Q2 5Q 1974Q3 1975Q2 3Q
8 1979Q4 1980Q3 1980Q1 3Q 1980Q1 1980Q2 1980Q4 2Q 1980Q2 1980Q4 2Q
9 1981Q3 1982Q3 1982Q1 4Q 1982Q1 1981Q4 1983Q1 5Q 1981Q4 1982Q2 2Q
10 1990Q2 1991Q2 1990Q3 4Q 1990Q4 1990Q4 1991Q2 2Q 1990Q3 1991Q2 3Q
11 2000Q4 2001Q3 2001Q2 3Q none 2001Q2 2002Q1 3Q none none none
12 2008Q1 ? 2008Q4? >=6Q? 2007Q3 2008Q1 ? ? 2008Q3 ? >=4Q?