White Papers

When Bad News is Good News: Market Forecasts Based on Investor Reaction to Unexpected News

Biff Robillard with Seth Bender

Working Paper

May 2002 (Revised 2010)


A stock market rally in response to surprisingly high unemployment news, during a contraction, is a buy signal for investors.

Stock markets, on average, will rally on higher-than-expected unemployment news during expansions, yet fall on the same news during contractions. This is because unemployment news contains two types of primitive information relevant for valuing stocks: information about 1) future interest rates and 2) future corporate earnings and dividends. A surprisingly high unemployment rate typically signals a decrease in future interest rates which is good news for stocks, as well as a decline in future corporate earnings and dividends, which is bad news for stocks.  The relative importance of the two depends on the state of the economy. Information about future interest rates dominates during expansions (low interest rates while business is good is a positive factor for stock prices), and information about future corporate earnings and dividends dominate during contractions (in a slowdown, earnings expectations drive stock prices, regardless of interest rates). However, under certain rare circumstances, markets may actually contradict the common sense interpretation of worse-than-expected unemployment news and rally. This event can be significant to investors.

We find that during contractions, market rallies in response to higher than expected unemployment news (which appears contradictory) signal economic recovery and imminently higher stock markets.  This often occurs within a few months of the end of a recession. Investors apparently begin to value low future interest rates over slightly lower growth expectations, i.e. they are transitioning from a recession response (when earnings matter most) to an expansion response (when interest rates matter most). This is an early indication that the economy is in a transition from contraction to expansion, and more importantly a sign equity markets are recognizing the imminent end of the contraction.  With this understanding, we make our most important discovery: A stock market rally in response to surprisingly high unemployment news, during a contraction, is a buy signal for investors.

G. L. “Biff” Robillard, III


Bannerstone Capital Management, LLC

Deephaven, Minnesota

952 249 8888


Seth Bender

Research Assistant

1. Introduction

This study attempts to make market forecasts based on investors’ reactions to surprising unemployment news during contractions. The particular news event we consider is the Bureau of Labor Statistics (BLS) monthly announcement of the unemployment rate. We establish that the economy is beginning to recover and equity markets are rebounding when stocks rally in response to higher than expected unemployment news during contractions. On average, such a rally in response to a surprisingly high rate occurs  a few months before the contraction ends.

Stock prices are based on three things: 1) the risk free rate of interest, 2) the expected growth rate of corporate earnings and dividends, and 3) the equity risk premium. During expansions, when surprisingly high unemployment data are announced, the expectations for possible interest rate cuts dominate over potential downward revisions in growth rates– and thus the stock market goes up. Investors do the opposite  in contractions: During a contraction, when surprisingly high unemployment data come out, investors again revise their growth expectations downward  but they now allow it to dominate over any good-news effect of future  interest rate cuts- and thus the stock market falls. It is important to note that these reactions happen on average, and as every investor will attest, it is impossible to successfully predict exact stock market movements, especially over short periods of time.

So how is this information useful for creating a market-timing tool? What does it mean when the United States’ economy is in a contraction and the equity markets make a move upwards in response to higher than expected unemployment news? Could this be an early signal from equities that the economy is coming out of a recession, and therefore a signal to start buying stocks? This is our fundamental question.

Investors invariably don’t realize a recession is ending until it is no longer useful to know. Stock markets historically move ahead of the actual economy. Markets will typically sell off months in advance of a contraction, bottom while the recession is worsening, and begin recovering months before the contraction  ends.  Therefore by the time conventional econometrics tell investors the recession is over, it is usually late for traders. A well-timed realization an economic contraction is coming to an end can clearly yield rewards for investors. If we can pinpoint locations  investors are transitioning to bull market behaviours from bear market behaviours, before prices have significantly advanced, than it is reasonable to claim  these points  mark entry points into recovering equity markets.

In searching for this evidence, forecasts of unemployment rates are compared to actual announced rates during contractions. Specifically, we are looking for trading days during recessions when the unemployment rate data are higher than expected yet the equity markets move upward. These circumstances will be referred to as “Boyd Events”.  Boyd Events are evidence investors are beginning to exhibit behaviour usually associated with economic expansions, when investors overweight the significance of falling interest rates over diminished growth expectations.

In order to optimize investment profits from a market rebound, traders must invest well before conventional economic evidence confirms the recession is over.  A Boyd Event is a clear and important hint that equity market investors are anticipating the end of a contraction and therefore it is also an immediate signal to implement a bullish equity strategy. How equity market investors intuitively know that the end of the recession is near is not the focus of this paper.

Why focus on unemployment rates? Perhaps no economic report exerts as much impact on speculative prices or is as closely followed by market participants. Furthermore, since unemployment rates have a long and accurately dated time series, findings are more credible than results from examining many other macroeconomic events. The unemployment and the market data are both complete and large. 

2. Explaining recessions and the market as a business cycle predictor

What is a recession? The conventional definition is, “a prolonged period of time when a nation’s economy is slowing down, or contracting”. Such a slow-down is characterized by a number of different trends, including consumers buying less, a decrease in factory production, growing unemployment, a slump in personal income and an unhealthy stock market In addition,  a recession is classified as two consecutive quarters of shrinking GDP (NBER)*.   There have been seven  recessions in the U.S. since 1961. During the approximately 408-month period from February 1961 to November 2001 (the last recession with an NBER recognized end) the US economy was in a contraction for about 65 months. The generally accepted dates of the contractions are: January through November 1970, November 1973 through March 1975, January through August 1980, July 1981 through November 1982, and July 1990 through March 1991, March 2001 through November 2001 and of course, the current contraction which officially started in December 2007. The average duration of a contraction was about eleven months while the average duration of an expansion was over seventy months. For the ten completed business cycles since 1945, the average contraction lasted about ten months and the average expansion about fifty seven months. Unemployment was higher (6.8% average) during contractions and lower (6.1% average) during expansions.  On average the unemployment rate increased during contractions and declined during expansions. We focus on these seven economic contractions and the Boyd Events that occurred during them.

Forward Looking Stock Market

Although the business cycle and the stock market are not perfectly correlated, they are widely considered linked. As one Charles Schwab analyst put it “The stock market, a barometer of business, economic, political and social conditions, also often serves as a leading indicator of economic cycles. The performance of the stock market can provide valuable clues of economic turnaround, or at the very least some insight to where the investors think the economy is heading” (Charles Dow would doubtless agree). The stock market is forward-looking, and current prices reflect investors’ estimates for the future earnings potential, or profitability, of corporations and profitability is directly linked to economic activity. The “wealth effect” is also often regarded as a reason for the stock market’s predictive ability. When the stock market is rising, investors are wealthier and may spend more. As a result the economy expands. When the stock market is falling, the opposite happens. (Comincioli)  This relationship can be shown in Graph 1 (quarter S&P vs. US GDP) in the appendix. At several spots on the chart it is evident that the S&P troughs before actual GDP.


This observation is relevant to our work: Since markets often begin to rebound months before the contraction actually beings to end, the hard data released stating that the recession has ended is  too late for investors. The early bird gets the worm, but too early and prices continue to fall. Traders perpetually face the dilemma of  investing  before the end of a contraction is economically “visible” to optimize profits,  yet not so early as to suffer heavy unnecessary losses. 

3. Related Literature

McQueen and Roley (1993) were among the first to observe correlations between fundamental macroeconomic news and stock prices. Much previous research found that macroeconomic news had little effect on stock prices. However, McQueen and Roley point out that each of these previous studies assumes that investors’ response to news is the same over different stages of the business cycle. For example, Summers (1989) assumed that a positive surprise in industrial production at the end of the Great Depression evokes the same response as a surprise in late 1969, after nearly a decade of expansion. But as McQueen and Roley explain, a positive surprise in industrial production during the Great Depression could indicate the end of a recession. This announcement would be “good news” to the stock market. Conversely, in late 1960, with business running near full capacity, a positive surprise in industrial production may result in fears of an overheating economy, inflation, and an effort by the fed to increase interest rates. Such an announcement could then be “bad news” for the stock market. McQueen and Roley showed in their paper that stock prices indeed do respond to macroeconomic news, but the responses vary over different stages of the business cycle.

Krueger (1996) studied the market rationality of bond prices to labor news.  Krueger focused on market reaction to the availability of more reliable information. His paper used an innovative approach to determine whether markets respond efficiently to new information. The test is based on the improved techniques in the 1980s by the Bureau of Labor Statistics (BLS) to estimate payroll and employment. The sample size increased, and data became more reliable. Due to these new measures, the Root Mean Square Error (RMSE) of estimated employment fell from 121,000 in 1979 to 71,000 in 1994. However, contrary to expectations, the effect of reported employment changes on interest rates was at least as strong in the early 1980s as it was in the mid 1990s. Despite the unexpected results, Krueger in his researched reiterated the findings of McQueen and Roley, but specifically pointed to the strong effects unemployment announcements had on market prices. As he said “In the markets, the monthly unemployment report has become the single most important indicator for economic strength, potential for inflation and fed strategy.” He further supported this relationship through data. He reports in Section III and IV of his dissertation that movements in the BLS survey data have a significant and sizable effect on the 30-year Treasury bond yield and on short term Treasury bill yields on the day the employment numbers are released. An unanticipated increase of 200,000 jobs is associated with about a 4 to 8 basis point increase in the long bond yield, and a larger increase in the short-term bond yield. In addition Krueger noted other elements in his research relevant to this paper. First, interest rates are not significantly affected by announcements of revisions to past months employment data; only the last month’s unemployment data seems to matter. And also, unexpected changes in unemployment are insignificantly related to long-term interest rates, but significantly related to short-term rates.

But clearly there has been no bigger contributor to our work than John Boyd, Ravi Jagannathan and Jian Hu. They find  on average an above consensus  unemployment rate is “good news” for stocks during expansions and “bad news” during economic contractions. This, they explain, is due to two primitive types of information relevant for valuing stocks- information about future interest rates and future corporate earnings and dividends. A surprisingly high unemployment rate typically signals a decline in interest rates, which is good news for stocks, but a a decline in future corporate earnings which is bad news for stocks. The relative importance of intertest rates versus future earnings changes over time depending on the state of the economy. For stocks  information about interest rates dominates during expansions. Conversely, information about future corporate earnings and dividends dominates during contractions. Campell and Mei (1993) attempt to explain this odd pattern of market responses through three factors: the risk-free rate of interest, the expected rate of growth of corporate earnings and dividends, and the equity risk premium.

Boyd and his colleagues explain  stock prices are not solely based on information about future interest rates that are contained in the unemployment news. If this were the case stock and bond prices would react in the same manner because bond prices, especially U.S. Treasury issues, are almost exclusively determined by interest rates. But stocks and bonds  in fact react quite differently part of the time: during contractions. During contractions stock prices react significantly and negatively to rising unemployment, but bond prices do not react in any significant way. Therefore, since bond prices do not respond,  unemployment news contains little or no information about future interest rates during contractions. However, unlike bond prices, stock prices do respond significantly to unemployment news during contractions, therefore unemployment news must contain information about growth expectations and the equity risk premium, just as Cambell and Mei had pointed out. In expansions stocks and bonds re-establish a positive correlation to unexpectedly weak unemployment data. During an expansion, both bond and stock prices rise when unemployment numbers are surprisingly high. Thus in expansions, weaker than expected labor market news causes future interest rate expectations to decline.

This pattern can be explained because rising unemployment is always followed by slower growth. But this relationship is much larger in contractions than expansions. Since equity investors are rational agents who study the real sector data, they would be expected to revise their growth expectations more significantly during contractions than during expansions. Therefore expectations about growth have a more pronounced change during contractions than expansions.

In sum, the change in growth expectations during expansions due to the bad effect of surprisingly high unemployment news is not sufficiently strong to cancel out the good news effect of a revision in expected future interest rates. Consequently, in expansions, stock prices respond positively to an unexpected increase in unemployment. During contractions, the relative importance is reversed. The bad news effect of rising unemployment on growth expectations is relatively strong. However, as bond prices illustrate, interest rates do not respond significantly to unemployment news during contractions. Therefore the good news effect on interest rates of rising unemployment rates is weak or nonexistent. The net effect is that during contractions the bad news effect of lower growth expectations dominates over the good news effect of lower interest rates, and stocks fall on higher than expected unemployment news.

As the fundamental framework for this paper, it is helpful to further explain Boyd’s work. To prove their hypothesis Boyd, Jagannathan and Hu developed a model to measure the “anticipated” and the “unanticipated” news component of the unemployment figures announced each month. A regression analysis was used to forecast the unemployment data and the reaction to the stock market. First, actual unemployment rates were used to forecast the forecasted unemployment rates. These forecasted unemployment rates were then compared to actual rates to determine the distribution of surprises, classified by whether the unemployment rate increased by more or less than the forecasted rate. Table 1 reports the results of Boyd’s model for average daily return on announcement days and non-announcement days during contractions and expansions. Table 2 computes average daily returns for the day of the announcement, when the data is sorted into “good news” and “bad news” unemployment surprises. News is categorized according to a common sense apparoach: “good”  when the announced unemployment rate is less  than forecasted using the model. “Bad” is when unemployment is higher than forecasted. As it is evident, a very definitive pattern emerges in the response of stock prices. In contractions, average stock returns are positive on good news and negative on bad news. But the opposite is true during expansions: average stock returns are negative on good news but positive on bad news. All these regressions are statistically significant to the 95% confidence level.

Boyd also points to the equity risk premium and growth expectations as variables that in addition to interest rates explain the stock markets reaction to unemployment rates. Boyd and colleagues explain  these factors impact stock price movement through the Gordon Growth Model.

Gordon constant growth model:


Where r is the interest rate, P is the price of a security portfolio, D is the period dividend, g is the expected (constant) rate of growth, and pie is the risk premium investors require to invest in stocks.

Through a series of regressions Boyd, Jagannathan and Hu show that in contractions there is no predicted stock price change due to news-induced interest rate changes. However, the estimated total effect of unemployment news on stock prices is negative, suggesting that either the risk premium (pie) must be rising; the expected future growth rate g must be falling, or both. Conversely, the sensitivity of stock returns to unemployment news due to its effect on the interest rate alone is about 3. However, the total effect of unemployment news on stock prices is about 1. That is, the predicted effect on stock prices through the interest rate factor is much larger than the actual combined effect of all three factors. Thus, during expansion, either the equity risk premium must be rising, growth expectations must be falling, or both. This is a logical implication of the observed responses of stock and bond prices.

How are the equity risk premium and growth expectations affected by unemployment news? Boyd and colleagues first explain  the equity risk premium is notoriously difficult to measure at high frequencies, and test to show correlations do not yield statistically significant returns. However, growth expectations are clearly affected. There is no direct measure for growth expectations, thus Boyd, Jagannathan and Hu used a couple of proxies. The first approach assumes  equity investors are rational agents who study data and make good forecasts. Using this basis, the researchers estimate the true relationship between the announced unemployment rate and dividend growth using the Index of Industrial Production (IIP) as a monthly proxy for corporate dividends. The idea was to see if this real sector measure of growth expectations is significantly different in contractions than it is in expansions. Boyd, Jagannathan and Hu studied the relationship between the Index of Industrial Production in the same month and one to four months following the reference month of unemployment rate announcement. There is a negative correlation for all regressions, but coefficients are much larger for contractions periods. This suggests that rational equity investors would be revising their growth expectations much more frequently and strongly in response to unemployment surprises in contractions than in expansions. Or in other words, growth expectations play a bigger role in determining stock prices during contractions than expansions. These findings are consistent with the researcher’s predictions.

The Rest of the Study

The remainder of our study proceeds as follows: Section 4 describes our attempts to locate forecasted unemployment and actual unemployment rates in order to pinpoint Boyd Events (market rallies, during a contraction, in response to weaker-than-expected unemployment data). Section 5 reports the findings of our research. Section 6 analyzes our results through regressions and averages in an attempt to make equity market forecasts. Finally, Section 7 summarizes and concludes.


4. Methodology

To reference Boyd Events we must obtain historic forecasted unemployment rates for the current, and six previous recessions. This was much more difficult than it may seem. In inquiries to several research departments at large financial institutions, just one firm was able to yield any sort of archived forecasted data. But this forecasted data was only available back to January 1992. Clearly this data would be insufficient because the recessions examined in our research date back to January 1970. We also looked into news services such as Bloomberg and Reuters, but were even less successful.  Yes, we considered using Boyd’s data. But Boyd, Jagannthan and Hu used past unemployment to regress what the forecasted unemployment rate would have been. This is not the direction we wanted our research to go. We wanted our research to be strictly empirical, encompassing only actual real-time market information — pushing aside models in favour of practical events. This is one area in which our research differs from others. We feel that actual data, defined as data from measuring and examining numbers from authentic market occurrences, is more persuasive and useful than data derived from models.  We were able to obtain the essence of forecasted unemployment rates in the Wall Street Journal archives (which, thanks to the Internet, has improved mightily since this working paper was first published in 2002!).

The general methodology of our research is as follows:  1. Determine release dates for unemployment announcements in past recessions.  2. Obtain forecasted rates for each month during past and current recessions. 3. Compare forecasted to actual rates. 4. Retrieve market data for occurrences in which the actual unemployment rate is above forecasted.  5. Search for Boyd Events- i.e.where the initially reported unemployment rate is greater than expected, during a recession, and the market goes up. 6. Observe market performance over various time periods after Boyd Events.

In order to utilize the Wall-Street journal archived data we had to retrieve release dates of the unemployment data throughout history. In most cases, the Bureau of Labor Statistics releases the unemployment data for a given month on the first Friday of the following month at 8:30 am Eastern Time. For example, the unemployment rate for March will usually be released on the first Friday in April. However, due to a variety of reasons, there are occasional exceptions due to the vagaries of holidays. Since we are examining market response on the day of the announcement, it is necessary to know exactly when the unemployment announcement occurred. On announcement days, the Bureau of Labor Statistics releases information in addition to the most recent unemployment rate. This includes the total number of employed and its distribution across regions and industries. It also releases employment totals, weekly and hourly earnings and weekly hours worked. This study focuses on BLS initial unemployment rate announcements only but we readily concede similar studies based on other data series is called for.

Obtaining the forecasted rates for the past recessions was  a tedious process requiring some flexibility. Every Monday the Wall Street Journal releases an article called “Tracking the Economy.” On the Monday prior to the Friday unemployment announcement, the consensus unemployment rate forecast can be retrieved from this section. Unfortunately,  “Tracking the Economy” only dates back to 1990 ( and was retired recently in the mid 2000’s). We were, however, still able to comfortably determine whether or not there was an unemployment surprise of any practical proportions using Wall Street Journal archives for  recessions prior to 1990.   To determine whether or not the announced unemployment rate was above or below expected, we used common sense and the historical information available. For example, on October 2, 1970 the unemployment rate announcement for September 1970 was released. Contained in the Wall Street Journal article on Monday, October 5th   was an article about the unemployment rate announcement. Words used to put the number into context were “Surprised”, and “Shell Shocked”.  We concluded the reported initial rate was above estimates, perhaps well-above consensus. For every release we found definitive language placing the announced unemployment rate into some type of practical context for our purposes. Prior to 1990, we make no representations about the minimum size of a “miss” to qualify, but simply assessed the article and graded the report “below”, “expected” or “higher than expected”.

As a practical matter, this was unambiguous. For example, continuing with the 1970 recession, the WSJ issue subsequent to the November 6th reporting date qualified the announced rate as “Well below forecasted”. Thus it was clear the announced unemployment rate was below forecasted (and therefore in no way a Boyd Event). Similarly, the Monday WSJ issue following the Friday, December 5th announcement (for the November 1970 unemployment rate) used phrases such as “Economists believe steady rate climb to continue” to describe the rate. Seeing no conclusive evidence that the unemployment rate was either above or below expected, we qualified the unemployment announcement as expected. On rare occasions, the Wall Street Journal, when putting the recently announced unemployment rate in context, would provide certain unemployment forecasts. On these occasions, we were able to check if our interpretation of the article matched the actual data. In every case it did. This method was used to determine whether the initial unemployment rate announcements were above, below, or as expected during the recessions beginning in 1970, 1973, and 1980. For all other recessions, an actual forecasted rate was obtained.       It is also important to note that we retrieved the forecast and unemployment rate data not only merely during the recession proper, but for the month before each recession began. This is because the  announcement for that month prior was actually made during the contraction, the first Friday of the next month. We examined every BLS unemployment rate announcement in each of the recessions after 1970. It is also unnecessary to examine unemployment numbers for the final month of a recession, since these unemployment rate announcements are actually made during the first month of an expansion.

After retrieving the forecasted unemployment rates during recessions, the next step is to compare this data to the actual unemployment rate. For the 1970, 1973, and 1980 recessions, where the surprise was estimated within the context of the Wall Street Journal reporting instead of an actual survey consensus expectation, obtaining the percentage of “beat” is not possible. Retrieving the actual initial unemployment data seems like a simple task. The Bureau of Labor (BLS) keeps an updated record of actual unemployment rates dating back to 1948.  The problem with the listed data on the website, of course, is it has been revised many times with the legacy data disappearing without a trace. The initial BLS data  for many series are often substantially revised, including the unemployment rate.  At the end of each calendar year, for example, the Bureau of Labour Statistics re-estimates the seasonality of unemployment, and other labor series, including another full year of data (McIntire).

This revised data is, of course, useless to our research because it was not the information investors experienced on the day of the initial announcement. Therefore, in order to appropriately conduct our study, it was essential we obtained the actual initial unemployment rate released on the first Friday of the month, instead of any later- revised number. Revisions can be quite substantial. For example, the November 1990 unemployment rate number (announced in early December) was initially reported at 5.9%, and later revised up to 6.2%. The consensus forecast for the November 1990 unemployment rate was 5.9%. This meant the announced data was a “met expectations”. Using the revised number as if it was the initial number, we would have interpreted the announcement as above expectations and possibly misinterpreted any market action that day,.  Revised numbers could  lead to inaccurate results.  We found the actual initially reported unemployment rate in the  Wall Street Journal archives and ignored the potentially revised BLS website data.

Armed with the forecasted and actual announced unemployment rates, the next step is determining market performance on announcement days when the actual rate was above the expected rate. We use a binary definition ( it is a surprise or it isn’t) for a surprise when using an announced rate versus an exact mathematical value for the expected rate. Whether the difference is as much as a .5% or as small as .1%, any surprise is a surprise. We treat price changes in the stock markets almost the same way with little interpretation of results: up is up and down is down, although “unchanged” in the daily data is subject to some interpretation, but certainly of no practical significance. We chose widely followed indices generally considered to  represent broad U.S. stock movements: the Dow Jones Industrial Average, NASDAQ Composite, and the S&P 500 Index.   The NASDAQ Composite began during the 1970 recession after our first Boyd Event. Only the S&P 500 and Dow Jones Industrials data were available for this earliest data surprise, but all three indices were used as soon as possible and thereafter.

After the Boyd Events have been identified, we next plot their location within the recessions and on price charts of the indexes.  We observe market performance in the days, weeks, months, and years following the Boyd Events  to evaluate the practical value of a Boyd Event as a market timing tool.

5. Statistical analysis

The recessions we examined are: December 1969 through November 1970, December 1973 through February 1975, February through August 1980, August 1981 through November 1982, August 1990 though March 1991,  March through November 2001, and the recession which began in December 2007 but has not been officially declared over. During the 1970 recession there were eleven unemployment announcements. Of these eleven announcements three were below expected, two were expected, and six were above expected. For the six announcements above expected,  two were Boyd Events (an above expected unemployment rate and the market reacts positively). The dates of these Boyd Events were February 6th 1970, and October 2nd 1970. Note the February 6th 1970 event was defined, by necessity, by the then-two major indexes: the S & P 500 and the Dow Industrials. The NASDAQ Composite did not exist yet. This is the only Boyd Event defined by two rather than three indexes.

The recession that began in 1973 was one the two longest recessions of the five (the most recent recession has not been declared ended by the NBER). In total, there were 16 unemployment announcements throughout the 1973-75 recession. Of sixteen announcements, four were below, seven were expected, and five were above forecasted.  Of the five above forecasted, there was only one Boyd Event- it occurred on January 3rd 1975.

During the 1980 recession there were seven unemployment announcements. Of seven announcements, two were below, three were expected, and two were above forecasted. Both of the announcements above forecasted were Boyd Events. They occurred on May 2nd and June 6th 1980.

The recession beginning in 1981 is the other very long recession. There were sixteen unemployment announcements as well. Of the 16 announcements, two were below, ten were expected, and four were above forecasted. Of the four announcements above forecasted, one of them was a Boyd Event. It occurred on November 5th 1983.

During the recession beginning in 1990, there were eight unemployment announcements. Of those announcements, five were expected and three were above forecasted. Only one of those above forecasted was a Boyd Event.  This occurred on November 3rd 1990.

The March 2001 through November 2001 recession had 8 unemployment rate announcements. Of these announcements four were below, none were expected, and three  were above forecasted.  Two were Boyd Events. They occurred on April 4th 2001**, and November 2nd 2001 (both Boyd Events will be explored in further detail later).

In total there have been nine Boyd Events in the recessions since 1961. TABLE 3 displays the announcements dates for the unemployment percentage during recessions. TABLE 4 is the same data in a spreadsheet.

6. Interpretation of results


The methods for reporting and interpreting our findings are as follows:

First we interpret data showing Boyd Events appear to occur during the transition from bear market to bull market, contraction to expansion, and explain research anomalies. Next we discuss using Boyd Events to forecast market performance, while proving they are, under particular circumstances, an immediate buy signal. Finally, we put the data into a format useful to prove statistically significant forecasts.

Predicting economic change

There are nine Boyd Events in the recessions since 1961, i.e. days when the major stock averages rallied when the BLS unemployment rate was higher than economists expected. It is important to note Boyd Events occur only if and when the unemployment rate is higher than economists expect: a “miss”. We do not examine the distribution of economists’ misses with respect to the unemployment rate in the course of a recession, although this is obviously relevant: If economists do not miss during a recession, no Boyd Event is possible. Curiously, however, there are no recessions since 1961 which have not contained at least one Boyd Event. These particular Boyd Events (especially with some practical considerations) tend to occur near the end of a recession, with at least two occurrences (February 1970 and April 2001) in the very early months of two of the officially defined recession. It is extremely unlikely a trader, in real-time, would have concluded such a response to bad news is a Boyd Event: the contraction is almost always unobvious at its onset, which is strictly defined much later. The timing of the Boyd Events is, of course, of critical importance to the trader. They must occur by virtue of both a miss by economists AND they must occur in a context the trade can recognize as a contraction. We treat the February 1970 event and the April 2001 event as too-early-to-tell.  Boyd Events occurring in the very early months of a recession (indeed in our study the two “exceptions” occur in the first three to ten weeks of the official contraction) may be invisible to the trader.  It is a common sense test. This topic receives amplification in our Research Anomalies and Conclusion sections which follow.

There are two Boyd Events in the 1970 recession: One in February 1970,  just six weeks into the recession and eight months before the end of the recession, and one in October 1970, eight months later and less than two months before the end of the recession. The  Boyd Event in the  1974 recession occurred fifteen months into the recession and less than two months before its end. The Boyd Event in the 1980 contraction took place four months after the onset and about three  months before  the recession was over. The Boyd Event in the sixteen month 1981-82 recession took place less than one month before its conclusion. The Boyd Event in the 1990 recession occurred five months into the recession and less than four months before the end. A Boyd Event occurred in April of 2001, about two months into the 2001 recession, and again in November of 2001, eight months into the recession and in the month it ended (rather similar to the 1970 recession).  There was at least one Boyd Event in the December 2007 recession, which occurred in November of 2008, about three months before the stock markets dramatically bottomed in March 2009. At this writing (July 2010) the NBER has not officially recognized an end to the December 2007 recession.

Research Anomalies

Now we address the research anomalies that could be grounds for criticism.  In at least two recessions there were “premature” Boyd events, occurring so early in the contraction as to call into question any usefulness for trading. In other word, a trader heeding these “premature” signals would be in earlier than optimal.  These early Boyd events occurred in the 1970 recession (just a month after the recession began),  and in the March 2001 recession (two months after the recession began),  Both the 1970 recession and the 2001 recession contain two Boyd Events each, with the second Boyd Events offering more timely entries for the trader, within a few months of the end of the recession.  Boyd events have on average happened just a few months before the end of a recession.

We are comfortable, as stock market practitioners, eliminating these two very early Boyd Events from the official count. Any Boyd Event occurring very early in a contraction is apt to be miscategorised. The utility of a Boyd Event to a trader is dependent on the trader’s ability to discern a “rise on a surprise” in a contraction versus a “rise on a surprise” in an expansion. By any practical measure, a trader experiencing a market rally on worse than expected unemployment data in the very initial stages of an economic contraction would not likely suspect, let alone know, the economy was already in contraction. One could argue, especially in the context of the NBER definition of a recession* ( see the footnote), these early Boyd Events are simply typical expansion responses to weaker than expected data: the boundary of a contraction is subjective at the very frontier.

If you eliminate these two early Boyd Events, on average (excluding the current 2007 contraction data)  there has been a Boyd Event about 2.5 months before the end of each recession since 1961.  

Also, after looking at the date set in TABLE 4, you may wonder why we qualified November 2nd 2001 as a Boyd Event. The criterion for a Boyd Event is a market rally in response to a surprisingly high unemployment rate. On November 2, 2001 the NASDAQ actually fell but by one point(the other two indexes rose), less than .01%. The S&P and Dow Jones both made significant rallies. Even though the NASDAQ fell, it was still clear enough to a trader that the markets were, on balance, rallying in responding favourably to the bad news. The NASDAQ Composite is usually considered the least broadly-based on the three indexes we consider. But this is a potential criticism of this particular Boyd Event.

The Immediate Buy Signal

It was our hypothesis the first hint of investor tendencies to buy unexpectedly bad unemployment rate news (when the unemployment rate is higher than expected) during a contraction ( under conditions which it is recognizable as such), should be a good time to buy stocks. As the data shows, we were right: during a contraction, the first time investors reverse the tendency to sell bad news- which is exactly what they have been doing during the onset of the contraction- the psychology for stocks is changing, and it is time to buy. As proved, the Boyd Event is correlated with the end of economic deterioration. But more importantly, as a practical matter, it often coincides with the end of price deterioration for stocks. It is an early signal from equity markets that the expansion is approaching, which explains why equities soon begin to trend higher. Stock prices discount the future earnings, not so much the current earnings of stocks. It is well recognized that stocks usually bottom before the economy does. Boyd Events appear often to precede the end of recession, but coincide—at least as a practical matter– with market lows. Graphs 2, 3, and 4 show Boyd events since 1970 for the Dow and S&P and since 1971 for the NASDAQ. These graphs, as well as all other forecasts in the Appendix, ignore the two early signals already discussed. The graphs illustrate the Boyd events’ practical relevance for market timing. In almost every case, they occur at, or near market lows.

Note: Boyd Events and their practical value bring to mind Charles Dow, the creator of the world’s first stock market index. It is widely reported Dow was far more interested in the correlation of stocks prices with general economic conditions than with future stock prices. Dow postulated stock prices at any moment incorporate all that information known, but excludes “Acts of God”.  There are arguably two recent cases of “Acts of God” relevant to the efficacy of Boyd Events as a timing tool.

The September 11, 2001 NYC attacks deserve special consideration. In summary, the Boyd Event of November 2001 coincides nicely with the end of the recession and a temporary low in stocks. Major markets rallied significantly (over 10%) for eleven months. We consider November 2001 to have been an entry point of practical value. The gathering storm clouds of the Iraq War, not a subsequent recession, soon contributed to the next broad market decline through late 2002.

The Financial Panic of 2008-09 adds yet another wrinkle. The NBER has not announced an official end of the recession which began in late 2007, so we cannot know yet how the November 2008 Boyd Event will line up with end of this recession. It may be worth noting throughout 2009, after the Boyd Event in late 2008, there are many times stocks rallied on weaker than expected unemployment rate announcements. Economists consistently underestimated the weakness in the job market and investors consistently rallied stocks on those days. The stock markets themselves declined significantly for about four months after the November 2008 Boyd Event. This is the earliest Boyd Event in our data (in both time and price) versus the ultimate market low for the cycle. The market low was reached in early March and index levels attained November 2008 levels by early summer, some 3 calendar quarters after the Boyd Event.. In both the 9/11 and the Financial Panic of ’08, Boyd Events, while useful, contended with significant market and economic events in addition to mere recessions and recoveries. Like all market timing tools, context matters and the experience and skill of the practitioner play key roles for Boyd Events.

Economic cycles have deep and enduring influences on stock prices, but many important market rallies and declines occur throughout history outside the context of an economy transitioning from contraction to expansion or vice versa. Boyd et al have demonstrated the dynamic nature of investor interpretation depending on context. Dow’s “Acts of God” and other significant exogenous factors must not be ruled out when contemplating the significance of a particular Boyd Event.

Timing Statistics

In order to further demonstrate the affects of the Boyd Event, we examined market returns 3 months, 6 months, and 1 year after the timing signal and compared them to average market returns since 1970 (the first recession in our data set). NASDAQ data only dates back to 1971. 

Average market returns were calculated as:

I x { (1 + i)^ N }  =  C

“I” represents the initial price of the equity (The price on January 1, 1970 for S&P and Dow, and February 1, 1971 for the NASDAQ. “N” is the number of periods. For example, there have been 129.3333 3month periods since January 1st 1970. “C” represents the current price of the equity.  Note: ^ represents “raised to the power” The “i” represents the average monthly return for this period.  To get the average 3 month, 6 month, 12 month returns, etc., use your “i” and calculate:

{(1 + i) ^ 3} – 1; {(1 + i) ^ 6} – 1, etc.

The average returns for equity markets 3months, 6months, and a year after Boyd Event is significantly higher in each index compared to the average return on the market invested in any given 3month, 6month, and a year period. Not only are the average returns following Boyd Events significantly higher than the average, but the returns after each Boyd Event were almost always higher than average. Taking all data from the three indices into account, there were only 3 instances out of 66 where the average market return was greater than the return 3months, 6months, and a year after a Boyd Event. Table 5 summarizes percentage return following Boyd Events. Table 6 summarizes average market return. GRAPHS 5, 6 and 7 show average market returns vs. returns 3months, 6months, and 1 year following Boyd Events in the Dow Jones, S&P, and NASDAQ respectively.

Regression Analysis ( forecast)

Our regression analysis results were precisely as expected. Each regression “best fit” line showed an immediate acceleration of the markets upwards, and at about a year, the markets began to revert back to normal growth patterns. Hence the logarithmic shape. The regression graphs are shown in Graphs # 8, 9, and 10. To perform the regression analysis, data for all three indices were taken every month for two years following each Boyd Event. The average return for each month following a Boyd Event was used in the actual regression. Again the average returns were much higher than average. Table 7 displays the regression returns. All regressions have an R(squared) greater than 70%,  meaning 70% of the return is explained by time. In regressions using economic data, R(squared) greater than 70% is considered highly correlated, and statistically significant (Acastat). Additionally, all regressions yielded low standard deviations.

For the Dow Jones, market return following a Boyd Event can be found through the equation:

y = 0.0549Ln(x) + 0.0548

Where “y” is the percent return and “x” is the number of months after the Boyd Event. Using this equation the returns 3months, 6months and a year after are 11.43%, 15, 24%, and 19.04% respectively. The R(squared) is .8781. Suggesting that

88% of the return is explained by the by the number of months after a Boyd event. The standard deviation of returns was 1.7%

For the S&P, market return following a Boyd Event can be found through the equation:

y = 0.0453Ln(x) + 0.0475

The variables are the same as above. Using this equation the returns 3months, 6months and a year after are 9.73%, 12.87%, and 16.01% respectively. The R(squared) is .8494. The standard deviation of returns was 1.62%

For the NASDAQ, market return following a Boyd event can be found through the equation:

y = 0.0674Ln(x) + 0.0986

The returns 3months, 6months, and a year after are 17.26%, 21.94%, and 26.61% respectively. The R(squared) is .75. The standard deviation of returns was 3.3%

Regression analysis (proving statistical significance)

In order to prove statistical significance we initiate the use of dummy variables. We obtained return data every month since 1970 for the S&P and Dow and since 1971 for the NASDAQ. In our first regression, the six months following each Boyd event was highlighted with a “1” in the data set, and all other months were given a “0”. There are six dummy variables for the S&P and Dow, and five for the NASDAQ (only five Boyd Events occurred within NASDAQ’s history). If the coefficients (d1, d2, d3 etc.) are statistically significant, meaning that they are not 0, then we know that the Boyd Event, and not normal equity growth is playing a part in the price acceleration following market rallies to higher than expected news during contractions.  The results are shown in Table 8. Although not all statistically significant, many of the coefficients are larger than the standard error. Several are remarkably statistically significant. The p value for coefficient d2 is very small for each index. This proves to a very high probability that the Boyd Event caused the market rebound, and not arbitrary equity growth.

Since not all coefficients were statistically significant, we were interested in furthering the analysis in order to find conclusive statistical evidence. In theory, we can expect the same returns from each Boyd Event. In our study we have not indicated any reason for which an investor would expect higher returns from one Boyd Event over another. Our next regression then highlights every 6 month span after a Boyd Event with a “1”, and all other data with a “0”. In this case, there is only one variable. The results are shown in Table 9.  The P values are almost zero, illustrating that these Boyd Events are undoubtedly a source of market acceleration. To sum, taking each individual Boyd event does not consistently yield statistically significant results. However, the chance of similar market returns to those 6 months following Boyd events from six random dates (5 for the NASDAQ) is zero. Therefore clearly something other than normal growth is influencing the markets at these inflection points. This is conclusive statistical evidence, that the Boyd event was indeed the cause of subsequent increases in equity prices. 

7. Summary and Conclusions

We have proven through historical and regression analysis that when markets rally in response to worse than expected unemployment news during contractions, it signals investors’ realization that the economy is recovering which leads the markets to begin rebounding. This is because investors begin to favor expectations of interest rate cuts over growth expectations in their valuation of stocks and therefore they buy bad news instead of selling it.  Equity markets recover a few months before conventional econometrics tells us the recession is over, and a Boyd Event is a signal that the markets are immediately beginning to rebound. Market returns three months, six months, and a year after Boyd Events are substantially higher than average. Thus, this market-timing tool has significance for investors. In order to pinpoint Boyd Events, historic unemployment data was retrieved from the Wall Street Journal and market data was examined for the S&P 500, NASDAQ Composite, and Dow Jones Industrial Average on days where the announced unemployment rate, during contractions, was higher than the forecasted. A Boyd Event occurred when, during contractions, the announced rate was higher than expected and the markets rallied across all three major indexes.  This paper added to research of John Boyd, Ravi Jagannathan, and Jian Hu, who proved that on average, the markets responds positively to bad news during expansions, and negatively to the same news during contractions. This is because during expansion, investor’s expectations of future interest rates dominate over the downward revisions in growth rates. However, during contractions, investors revise their growth expectations strongly, which dominates over the expectations of interest rate cuts.

Future Research

Our research led to many ideas for future investigations. The most obvious to us is investigating the “other” Boyd Event behavioural frontier: when investors begin selling worse than expected employment data during an expansion. Does this behaviour occur near useful turning points in equity markets at tops?

The facts we have presented raise a fundamental question about our research. How do investors and markets intuitively realize that the end of an economic contraction is near? As our studies proved, markets are forward looking. The historical ability for investors to do so is striking and intriguing.

The size of the surprise could also lead to potentially groundbreaking discoveries. That is, do big surprises mean more reliable signals? For example, if the unemployment surprise is larger, is that a better, equal, or worse predictor of the state of the economy than a lesser surprise where the markets rally? Moreover, could larger surprises lead to more prominent rallies during expansions or larger sell-offs during contractions?

Given the current near-zero real interest rate, it is difficult to anticipate any more cuts, no matter the state of the economy. Stocks rally in response to higher than expected unemployment news during expansions because of investors’ expectations of future interest rates. If the rate is already so low, can investors really expect further interest rate cuts? Or in other words, does the current interest rate level affect investors’ expectations of possible interest rate cuts? And if so, does this lead to greater or weaker market rallies in response to unemployment news? This research could provide clues to explaining the market response on May 3rd, 2002. The unemployment rate was reported at 6%, the highest in eight years and a big surprise to investors. The markets fell substantially, maybe because investors cannot expect cuts given the current rate.

Our study shows that a Boyd Event often marks an improvement in the equity markets’ rates of return. How long does the acceleration last? In 2001 it was only for a quarter or two at most, for example. How long until the markets revert to average rates of return? Our regression analysis provided data suggesting that it takes about a year. Additional research is called for.

Finally, the regressions yielded better results using third degree equations because of several distinguishable humps consistent through all the indices. There could be several explanations for this, all warranting future research.  In sum we claim for this study we have identified a market reaction phenomenon that predicts imminent economic recovery, and often signals an almost immediate equity market rebound.  This is a significant new market indicator for equity investors when combined with the many tools already in the trader’s toolbox.


*Statement of the NBER Business Cycle Dating Committee on the Determination of the Dates of Turning Points in the U.S. Economy

The NBER’s Business Cycle Dating Committee maintains a chronology of the U.S. business cycle. The chronology comprises alternating dates of peaks and troughs in economic activity. A recession is a period between a peak and a trough, and an expansion is a period between a trough and a peak. During a recession, a significant decline in economic activity spreads across the economy and can last from a few months to more than a year. Similarly, during an expansion, economic activity rises substantially, spreads across the economy, and usually lasts for several years.

In both recessions and expansions, brief reversals in economic activity may occur—a recession may include a short period of expansion followed by further decline; an expansion may include a short period of contraction followed by further growth. The Committee applies its judgment based on the above definitions of recessions and expansions and has no fixed rule to determine whether a contraction is only a short interruption of an expansion, or an expansion is only a short interruption of a contraction. The most recent example of such a judgment that was less than obvious was in 1980-1982, when the Committee determined that the contraction that began in 1981 was not a continuation of the one that began in 1980, but rather a separate full recession.

The Committee does not have a fixed definition of economic activity. It examines and compares the behavior of various measures of broad activity: real GDP measured on the product and income sides, economy-wide employment, and real income. The Committee also may consider indicators that do not cover the entire economy, such as real sales and the Federal Reserve’s index of industrial production (IP). The Committee’s use of these indicators in conjunction with the broad measures recognizes the issue of double-counting of sectors included in both those indicators and the broad measures. Still, a well-defined peak or trough in real sales or IP might help to determine the overall peak or trough dates, particularly if the economy-wide indicators are in conflict or do not have well-defined peaks or troughs.



                Acastat (2002). Pearson’s Product Moment Correlation Coefficient


Boyd, John., Ravi Jagannathan and Jian Hu (2001), “The Stock Market’s Reaction to Unemployment News: Why Bad News is Usually Good For Stocks”. NBER Working Paper 8092

Comincioli, Brad (1996), “The Stock Market as a Leading Indicator: An Application of Granger Causality”, The University Avenue Undergraduate Journal of Economics, Sample issue (1996)

Dow Theory (2001). The Stock Market as a Business Cycle Predictor.


Jagannathan, Ravi and Zhenyu Wang (1993), “The CAPM is Alive and Well”, Staff Report #165, Federal Reserve Bank of Minneapolis.

Krueger, A.B. (1996), “Do Markets respond More to More Reliable Labor Market Data? A Test of Market Rationality”, NBER Working Paper 5769.

McIntire, Robert (2001). Revision of Seasonally Adjusted Labor Force Series. Employment and Earnings. http://stats.bls.gov/cps/cpsrs.htm

McQueen, G., Roley, V.V. (1993), “Stock Prices, News, and Business Condition”, Review of Financial Studies 6, 683-707.

NBER, (2001), Commentary on 2001 recessions. NBER


Staff reports. (2001).  Economists call it a recession. CNNmoney. http://money.cnn.com/2001/11/26/economy/recession/

Veronesi, Pietro (1999), “Stock Market Overreaction to Bad News in Good Times:  A Rational Expectations Equilibrium Model”, Review of Financial Studies, Vol. 12, No. 5, 975-1007.







–          Peter Tollefson, for his help researching the Wall Street Journal archives

–          Scott Larson, of Merrill Lynch, for his interpretations and commentary

–          The media and internet equity analysts of Dresdner Kleinwort Wasserstein in London, for their assistance with Data Stream and other research applications

–          Dr. Hans Crockett of Dresdner Kleinwort Wasserstein in London, for his help designing and interpreting the regression analysis.