Leading Economic Indicators
The Conference Board publishes various lists of Business Cycle Indicators. Those indicators are categorized as "leading," "coincident," or "lagging." This post will take a look at indicators 6-10 on the list of leading economic indicators. Indicators 1-5 will be covered in a second post at a later date.
Average weekly hours, manufacturing
Average weekly initial claims for unemployment insurance
Manufacturers’ new orders, consumer goods, and materials
Vendor performance, slower deliveries diffusion index
Manufacturers’ new orders, nondefense capital goods
Building permits, new private housing units
Stock prices, 500 common stocks
Money supply, M2
Interest rate spread, 10-year Treasury bonds less federal funds
Index of consumer expectations.
Looking at the above chart, I find it hard to believe that anyone thinks the stock market is a valid indicator of anything.
Starting in 1960 and using a decline of 10% as some sort of leading indicator would have generated five false positives, one miss, and one success. At the start of the 1980 recession, the S&P was up year over year about 5%; at the start of the 1982 double-dip recession, the S&P was up close to 30%; at the start of the 1991 recession, the S&P was up over 10%; and at the start of the 2001 recession, the S&P was nearly flat. The S&P did not decline 10% before, during, or after the 1960 recession. As a coincident indicator, the results would have picked up 1982 and 1991, but would still have missed 1960 and 1980. In 1987 and 2003, the stock market declined nearly 20%, but there was no recession.
Just for fun, let’s take a look at the Dow Jones home construction index:
Did the above chart "lead anything," or did the index peak a month or so after the June 13, 2005, Time magazine Home $weet Home cover?
Time and time again I hear, "The stock market acts six months in advance." Six months in advance of what? I fail to see how it is acting six months in advance of anything. If one is looking for leading economic indicators, the stock market is surely not one of them.
Also note that if one wants a stock market indicator, the economy is surely not it. Look at the plunging GDP in comparison with the stock market for recent proof. Look at the homebuilder chart above for recent proof. Look at the historic S&P 500 chart for proof. Seriously, the S&P is a hopeless leading economic indicator, and the economy is an equally hopeless stock market indicator.
Yet the myth (and the weighting) that the stock market is a leading indicator still persists. It’s no wonder that nearly everyone is confused, given that nearly everyone is looking for correlations that simply do not exist.
This chart of the University of Michigan consumer sentiment index seems to have some merit as a coincident indicator, but little as a leading indicator, at least in the time frame for which this data is available. The indicator also suffers from what seems to be a high percentage of false positives per correct call. As a coincident indicator, it has three false positives and four successes. One could draw the trigger line differently to avoid the false positives, but then the big recession in 1982 would be entirely missed.
Seven out of the last eight times the annual rate of change on permits was negative 20% or lower, the economy went into a recession (not counting the current situation). Currently, the chart shows that building permits in November dropped 31% from the year-earlier level.
In all seven recessions since 1959, building permits declined year over year. Using the 0% line as a threshold would have picked up the recession in 2001, but would have also resulted in false signals in 1965, 1985, 1987, and 1996. This can be summarized as seven out of seven with four false positives.
Using 20% as the threshold, the only false signal was 1987, but there would have also been a missed signal in 2001. This can be summarized as six out of seven with one false positive and one miss.
This is actually a reasonably good performance, especially if one uses a cross of the 0% line as a strong warning signal while waiting for continued confirmation. Note that dips below the 0% line tend to occur well before the onset of recessions. Leading indicators are supposed to lead, and this one does. A crossover of 0% is a strong warning, and a continuation below the zero line shows that a recession is likely.
In "Money Supply and Recessions," I introduced "M’" (M Prime) as a leading indicator based on sound Austrian principles and definitions of money. Those who have not seen how or why I came up with M’ can click on the above link to see just what M’ is all about. What follows now is a recap of M’ versus M2 as a leading indicator.
Leaving the current status as unknown, six of the last six recessions were marked by a major dip in M’. Note how the indicator clearly led the recession. Also note that six of eight sustained dips below an annual growth rate of 5% in M’ led to a recession. On that basis, we have two potential false signals (1985 and 1995).
Unlike M’, the direction of M2 does not seem to give clear economic signals. Note that M2 was rising into the double-dip recession of 1982 and the 1991 recession. Also note that the single largest dip in M2 was in 1993, while M’ was soaring. The years between 1992-1995 are all problematic. Finally, note that unlike M’, where a dip below 5% annual growth was a huge warning sign, the dotted line above shows no such significance. M’ seems to be far superior to M2 as a leading indicator.
M Prime CPI Adjusted
On a CPI-adjusted basis, we see that there has been a recession on six of seven sustained dips below the zero line of year-over-year growth in M’. The 1985 excursion below 0% was extremely brief, in stark contrast to all of the labeled recessions, and thus can be discounted. 1995 is still a miss, but nowhere nearly as pronounced compared with M’ unadjusted. 1995 also happens to correspond to the start of a huge ramp-up in sweeps. Perhaps that is significant, and perhaps not. Nonetheless, M’ CPI-adjusted gives a cleaner signal, arguably calling for seven recessions, of which six happened.
The above chart clearly shows M’ CPI-adjusted to be a strong leading economic indicator.
M2 CPI-adjusted is certainly an improvement over M’ CPI-unadjusted. Note, however, that the 1982 and 2001 signals are not as strong as the corresponding M’ CPI-adjusted signals. The M2 adjusted signal for 2001 was particularly weak. More problematic for M2 adjusted versus M’ adjusted is the mass of Jell-O between 1988-1996. M’ adjusted was clearly giving an all-clear zero cross-signal by 1992, while M2 adjusted gyrated for years and did not really give an all-clear until 1998. Furthermore, M2 adjusted actually dipped back below the zero line in 1997, while M’ adjusted was soaring upward. Both M2 and M’ missed around the 1996 time frame, but even then, M2 did worse both in terms of an actual low and the Jell-O that preceded it.
10-Year Treasury Minus FF Rate Spread
A dip below zero preceded six of six recessions since 1965. Unfortunately, it generated five false positives, as well. Of course, one can set the line at negative 1, in which there were only three false positive. Or one can set the line at negative 2, in which case there was one false positive and one miss. Still, it would be much nicer if we did not have such curve-fitting. Can we do better than this indicator?
The Yield Curve
The above chart was generated by subtracting the symbol $IRX from $TNX where $IRX is a 13-week discount and the latter a 10-year yield. Ideally, both would be yields, but the difference is not that great on the 13 week. We use free data when available, and that data not only works well, it also happens to be free.
This chart is almost perfect. A three month-to-10-year inversion is six for six with one false positive in 1966. The current situation is considered unknown.
Note how the above chart does not confirm the false signal on the 0% line cross in 1995 on the previously shown real (CPI-adjusted) M’ chart. Unfortunately, the false signal in 1967 on this chart predates the beginning of our M’ series of charts, but I suspect there was nonconfirmation in the other direction, with M’ not confirming this chart.
Comments From Paul Kasriel
The "real" unadjusted monetary base (bank reserves plus currency) seems to provide fewer false recession signals than does real M2 growth. That does not necessarily mean that the real base does a better overall job of forecasting real GDP, just that it does a better job of forecasting official recessions. Mish Note: "Real" in this case means inflation-adjusted via the PCE price deflator, and "unadjusted monetary base" means a nonseasonally adjusted monetary base
I have used the PCE price deflator to get "real," rather than the CPI, for purely arbitrary reasons here, not theoretical — I don’t have time to explain now, but it is not a big issue. Mish Note: There is a potentially confusing mix of terminology here, but none of the charts in this post were seasonally adjusted (except perhaps for consumer sentiment, and on that, I am unsure). Our inflation adjustments used the CPI, and any references to "real" in what I wrote (as opposed to what Kasriel wrote) means CPI-adjusted. As Kasriel suggests, there is little difference between the two. We tried both and settled on using the CPI, because that is what Shostak did, as explained in "Money Supply and Recessions"
Starting with the recession of 1970, a negative spread on the 10-year Treasury minus the fed funds rate in conjunction with a contracting year-over-year change in monetary base/CPI has predicted recessions with no false signals. In Q3 and Q4 of 2005, the real monetary base contracted, but the interest rate spread still was positive. Now, the interest rate spread has turned negative, but the real monetary base is no longer contracting — just barely. Mish Note: The 10-year minus the three-month spread by itself has no false positives and no misses since 1970.
Real M’ is a very good leading indicator. It also performs better in theory and practice in comparison with Real M2. Real M2 performs better than M2, and M’ performs better than M2. M’ is thus a better indicator than M2, no matter how it is compared (real or not)
The 10-year minus the three-month is an exceptional leading indicator. It works better in practice than using the 10-year minus the FF rate
No false signals have been given by the 10-year minus the three-month spread since 1970. False signals were given by spreads using the FF rate alone
No false signals were given by a combination of real monetary base and the 10-year minus the FF rate spread (as per Kasriel, but not shown)
Housing permits provide a valid leading signal. When the 0% line is decisively penetrated, a recession usually follows
The S&P 500 is simply not a valid leading economic indicator. It is at best a marginal coincident indicator and perhaps should be ignored altogether. There are just too many false and/or missed signals
Consumer sentiment may have some value as a coincident indicator, but it does not function well as a leading indicator.
This post is an attempt to find a methodology that makes theoretical sense and works well in practice, too. Five of the 10 widely used leading indicators were reviewed, one of which should be discarded outright, one redefined as a coincident indicator, and one (housing permits) seen as valid as it stands. Two leading indicator components had substitutes that seem to work far better in both theory and practice.
The charts show that M’ and the 10-year minus the three-month spread are both superior to similar indicators on the list. As time permits, I will take a look at the remaining five widely used leading indicators. Thanks once again to Bart at NowAndFutures for providing many charts based on specifications that I requested, and also to Paul Kasriel at Northern Trust for his time and comments.
Mike Shedlock ~ "Mish"
January 15. 2007