Notice: Trying to get property 'ID' of nonobject in /var/www/www.macrobond.com/wpcontent/plugins/reusabletextblocks/textblocks.php on line 268
In the barren fields of economic research, there are only a handful of academics devoted to cultivating fundamentally new ideas. One of those is Professor Roger Farmer (of Warwick, UCLA etc.), who has long questioned central tenets of the newKeynesian heterodoxy. Among other things, he has written on ‘Endogenous Business Cycles’ (with steady state indeterminacy) and wants to replace NewKeynesian sticky prices with what he calls a ‘belief function’. Speaking as a layman I would label Farmer’s work as essays for developing a macro framework in Behavioral Economics, despite Farmer’s assertion that he’s an oldschool (or perhaps ‘rightschool’) Keynesian. Now, I must admit to not seeing eye to eye with all I have read of Farmer, but today I thought we would study an interesting alternative hypothesis to business cycles, in particular the Global Financial Crisis that he has put forward.
Note: In the graph above I, like Farmer, use the logs of Real S&P500 (deflated by the ‘money wage’, which is compensation of employees divided by FTE number of employees) and UNR (defined as (100*UNR)/(100UNR )). These transformations produce new variables that are unbounded above and below, which is important since there is evidence that the two transformed variables are nonstationary but cointegrated. In order for a series to be nonstationary it must be able to increase or decrease without limit, independently of its current value.
Chart 1: There is some correlation…
In short, what we will have a look at is how well asset prices, in particular stock markets, explain the real economy; in particular, the unemployment rate. Below is a bivariate VAR(2) estimated for two periods, pre and postVolcker (i.e. a variant of “the great moderation”).
Dependent variable 
1953 Q11979 Q3  1979Q42017 Q4  

p  u  p  u  
R2: 0.95  R2: 0.98  
p(1)  1.42 (0.0000) 
0.35 (0.0001) 
1.29 (0.0000) 
0.22 (0.0000) 
p(2)  0.44 (0.0000) 
0.26 (0.0127) 
0.30 (0.0000) 
0.21 (0.0000) 
u(1)  0.14 (0.0419) 
1.43 (0.0000) 
0.07 (0.5471) 
1.56 (0.0000) 
u(2)  0.12 (0.0487) 
0.55 (0.0000) 
0.08 (0.4330) 
0.60 (0.0000) 
c  0.13 (0.1844) 
0.20 (0.0709) 
0.06 (0.1499) 
0.02 (0.4939) 
Note: VAR(2) in levels. Probabilities within parenthesis.
Table 1: Quite steady coefficients, especially for the unemployment equation.
One thing that Farmer points out is that the coefficients seem quite stable, suggesting that the relation between stock markets and the real economy are also quite stable. This is an attractive feature in comparison to many other economic relationships that tend to break down during/after the great moderation. Admittedly, the coefficient stability is somewhat less remarkable in my calculations (then Farmer’s ditto) but, nonetheless, still decent. Farmer seems to suggest that this is a proof of no structural breaks in our series, though in my opinion this is an altogether different question.
Under any circumstances, when I proceed with putting the data into the Macrobond VECMframework, cointegration rank tests are automatically computed. These indicate both that there is a unit root and that the series are cointegrated (1), over the full sample as well as in the two subsamples (pre/postVolcker) suggested by Farmer.
1953 Q11979 Q3  1979 Q42017 Q4  1953 Q12017 Q4  

0 rank (no coint.) 
1 rank (coint.) 
0 rank (no coint.) 
1 rank (coint.) 
0 rank (no coint.) 
1 rank (coint.) 

Trace test  0.0002  0.8767  0.0000  0.4736  0.0000  0.7320 
Maximum Eigenvalue 
0.0000  0.8767  0.0000  0.4736  0.0000  0.7320 
Note: The fact that the Johansen test points to one cointegrating vector means, by default, that there indeed exists a unit root (which is why a unit root test is more nice than necessary).
Table 2: Cointegration is present
We cannot yet perform causality tests in Macrobond (keep’em support tickets coming!), but even main stream economists admit that empirical studies mostly support the idea that asset prices cause real economy responses. The fact that Farmer (and others) finds that stock markets cause unemployment does not, as Granger himself pointed out^{1}, imply a means of control. For this, we need an economic model that chisels out the mechanism of the causal chain. And this is where Farmer gets interesting.
According to the traditional – fundamental – view of the business cycle, the GFC was caused by an event that signaled depressed earnings and, hence, increased unemployment for an extended period of time (at least until the next fundamental signal)^{2}. Furthermore, according to this view, policy intervention is (or rather, would be) futile as it cannot stave off the fundamental causes of forward looking, unique equilibrium, market behavior.
^{1}Granger (1980): “Testing for causality: a personal viewpoint”. Journal of Economic Dynamics and Control, p. 329–352.
^{2}Absent any clearcut event, the traditional model completely fails to explain the depth, breadth and longevity of the GFC.
Invoking Keynes “Animal Spirits”, Farmer instead paints the causal chain thus: For some reason, the perceived risk of doing business increased, and lower earnings were anticipated. As a result, market participants sold off shares in the belief that future market prices would be lower. Stock markets and the price of paper assets dropped, causing households to curtail spending, which in turn lead to worker layoffs; and the reduced level of economic activity caused a selffulfilling drop in company earnings.
In addition, and in contrast to the fundamental view, Farmer’s ‘animal spirits’ is also able to explain why largescale asset purchases are an effective way of restoring confidence and resurrecting the economy. From a monetary policy perspective, this is an attractive and truly important feature of Farmer’s work.
– This begs the question: Is the stock market really the economy?
Granted, it would be easier to just apply Farmer’s model and run with it. However, due to lingering doubts of the stability of the VECM, or rather the stability of the cointegrating relationship^{3} I have nonetheless chosen to also estimate a simple bivariate VAR in levels^{4}.
^{3}Doubts that are only augmented when studying the outofsample performance when using the preVolcker or full sample coefficients.
^{4}This is also suggested by the MBapplication as the cointegrating relation with other specifications and (shorter) samples often is full rank (implying the variables are close to stationary).
[mathjax]
$\left[\begin{array}{c}\Delta {u}_{t}\\ \Delta {p}_{t}\end{array}\right]=\left[\begin{array}{cc}0.57& \u20130.50\\ 0.06& 0.41\end{array}\right]\left[\begin{array}{c}\Delta {u}_{t\u20131}\\ \Delta {p}_{t\u20131}\end{array}\right]+\left[\begin{array}{c}\u20130.13\\ 0.00\end{array}\right]\left[\begin{array}{ccc}1& 1.71& 1.72\end{array}\right]\left[\begin{array}{c}{u}_{t\u20131}\\ {p}_{t\u20131}\\ c\end{array}\right]$(VECM estimated PreVolcker)
As you will see, the Bivariate VAR outperforms the VECM during the GFC quite handsomely as the latter model seems to suggest considerably more persistence in unemployment than what postGFC experiences de facto indicate.
That said the coefficient of the cointegrating vector, ‘α’, has the expected negative sign, albeit only insignificantly so in the stock market equation. As ‘α’ is 0.13 (and significant) in the unemployment equation it nonetheless suggests that unemployment adjusts toward the long run solution, the cointegrating equation, with some 13% per year. Encouragingly, independent of specifications, the parameter values for ‘α’, remain close to 0.1 (in a range of 0.13 to 0.09).
Now, turning to using stock markets as a forecasting tool for the real economy I have studied the outofsample^{5} performance of Farmer’s model for various horizons but here, for lack of space, I will focus on one multiperiod forecast surrounding the GFC, in particular the period after the FED initiated the large scale asset purchases, LSAP by the end of 2008. Also, as my aim here is to find a functional forecast model, I have chosen to let the application find the best possible model^{6}.
^{5}Outofsample model evaluations is another feature I urge you to send us support tickets on!
^{6}Using the Bayesian Information Criteria (BIC/Schwartz) and the default maximum of three lags and regressors.
Chart 2 a & b: Unemployment and the stock market, out of sample performance
Albeit far from perfect, the Bivariate VAR handsomely outperforms Farmer’s VECMmodel, and this holds true even for different lag structures or estimation windows. Despite seemingly stable coefficients in Farmer’s model (the cointegrating vector), the generally poor outofsample fit suggests that the VECM is indeed misspecified as discussed above. This, in turn, puts into question the idea that lowfrequency movements of the stock market matter for the real economy.
That said, the occurrence of breaks in cointegrating equations are known to make ordinary VARs generate better forecasts even if the underlying data generating process is a VECM. Also, the lion’s share of model specifications I (incl. sample windows) suggest that the data is indeed cointegrated, implying there is some Granger causality, even if the direction is unclear.
In conclusion, and with the important caveat that Farmer’s claim of causality is correct, the statement that the stock market is the economy is at least plausible. Furthermore, the causal chain Farmer has laid out suggests that it can also be controlled, with the largescale asset purchases from FED and other central banks being glaring examples. Hence, the apparent success of these policies should come as no surprise to those adhering to Farmer’s Keynesian interpretations (‘Farmerism’?).
To the rest of us– if nothing else – it increases the anxiety levels connected to the FED’s quantitative tightening process. Let’s hope our anxiety won’t instigate too much stock market volatility though. The economy depends upon it.
Disclaimer: We don’t usually have views and opinions about economic and financial states of affairs, (not ones that we express publicly as a company, anyway). We do believe, however, that people can and do appreciate a variety of perspectives. What you’ve just read is the perspective of the author. While we think our writers are very smart, Macrobond Financial does not expressly endorse the views presented here. And, as the old adage goes, you shouldn’t believe everything you read (not without finding the data, performing a few analyses and presenting it in a nice chart). We want to make it clear that we are not offering this information as investment advice. That being said, if you have Macrobond, you can easily check everything that’s mentioned here, and decide for yourself. If you don’t have Macrobond, now you have a great reason to get it.