Economics 2012 41
Economics 2012 41
Economics 2012 41
2012-41
Citation Eduard Gracia (2012). Predicting the Unpredictable: Forecastable Bubbles and Business Cycles und
Rational Expectations. Economics: The Open-Access, Open-Assessment E-Journal, Vol. 6, 2012-41.
http://dx.doi.org/10.5018/economics-ejournal.ja.2012-41
Author(s) 2012. Licensed under a Creative Commons License - Attribution-NonCommercial 2.0 Germany
The next major bust, 18 years after the 1990 downturn, will be around 2008,
if there is no major interruption such as a global war
Fred Foldvary (1997)
In the wake of the 2008 recession, as of every major recession for the last 150
years, the question of why the downturn happened and whether the dismal science
should have predicted it has been posed again. In a way, this follows an old tradition: the Long Depression in the 1870s triggered the Marginalist Revolution, the
Great Depression in the 1930s gave birth to Keynesianism, and the Oil Crisis in
the 1970s marked the ascendancy of the Rational Expectations Hypothesis. In its
21st Century edition, a substantial part of the challenge seems to focus on the latter
hypothesis, and whether a return to adaptive expectations / bounded rationality
would improve the models explanatory power respective to empirical observations. Unfortunately the debate, at least at a popular level, is vitiated by a logical
fallacy: what we could call the fallacy of probability diffusion symmetry.
The popular controversy goes as follows. Mainly under the neoclassical banner, defenders of the Rational Expectations / Efficient Markets hypothesis claim
that, should markets not behave according to it, they would create arbitrage opportunities that would enrich anyone clever enough to spot them, and it would
therefore suffice to add a few rational players in the mix for their irrational competitors to be driven out of business. This, they argue, means that observed market
values must follow a random walk, and therefore stock market bubbles and crashes
be utterly unpredictable, for any predictable patterns should already be discounted
out. Those in the opposite camp take then this conclusion as the central contention
point, against which they pose numerous examples of departures from the random
walk hypothesis in the observed data, as well as names of economists who, against
the mainstream opinion, were able to forecast the 2008 crisis well before it hit.
From a strictly logical perspective, however, both sides might well be wrong.
Indeed, both camps are implicitly assuming that, if expectations of future market
prices were rational, they would reflect the mean (or expected) path calculated on
the basis of all the information available; therefore, if current prices reflect the fuwww.economics-ejournal.org
ture expected path, their observed trajectory should also approximate the mean.
Hence, if arbitrage ruled out any predictable patterns along the mean path, then
they should also be absent from the observed path and, as a consequence, proving their existence in the observed data series would also prove that expectations
were not rational in the first place. Yet, sensible as it sounds, this implication is
only true if the underlying probability diffusion process is symmetric.
An intuitive example may make this point clear. Imagine a game of triple-ornothing: you make a bet of, say, $10, toss a coin and triple the investment if result
is heads, or else lose it all if it is tails. Evidently, if the game is played only once,
the distribution of results is symmetric, with a 50% probability of making triple or
nothing, and with both mean (i.e. average) and median (i.e. the value that leaves
50% of the distribution on either side) being $15. Yet if we play the game, say, ten
times in a row reinvesting the profits every time, the distribution changes: now
there is a 0.098% probability of making $590,490 and a 99.9% of losing
everything, so the mean value is $577 (a substantial profit respect to the $10
investment), but the median is obviously zero. Under Rational Expectations, $577
is of course the investments expected value; yet, for an external observer viewing
the data series, there is a 99.9% probability that the observed value after ten tosses
be nil i.e. closest to the median path. In other words: the prediction with the
highest probability of success is not the mean, but the median.
This is not just the case for this straightforward example but for any
asymmetric distribution, including those most frequently used in standard financial
modeling. Appendix 1 illustrates this for perhaps the simplest random-walk asset
return model: the Brownian motion. It follows that, if probability diffusion
processes are asymmetric (as virtually all the standard asset pricing models are),
then neither does the observation of predictable market patterns imply irrationality, nor does the Efficient Markets Hypothesis rule them out, for there is nothing
preventing them from appearing on the median path.1 Hence, the classical papers
by Fama (1965) and Samuelson (1965) postulating that valuations in a rational,
efficient market must preclude any cyclical or otherwise arbitrage-enabling pattern
remain completely valid only, they apply solely to the mean path.
_________________________
1
This, importantly, has nothing to do with the fallacy of ergodicity highlighted by authors such
as Paul Davidson (e.g. Davidson 2009): the binomial random process in the example above, for
instance, is totally ergodic and has a perfectly well understood probability distribution.
www.economics-ejournal.org
This, importantly, does not imply that we are in front of a money machine, for
it does not mean investors can resort to those predictable patterns to beat the
market but, at most, to fine-tune their probability of loss. For example, in the
game of triple-or-nothing, a rational investor may bet $10 and expect to end up
with $577, but by not betting may avoid suffering a loss 99.9% of the times and,
by selling short (assuming this were allowed in the game), might achieve a gain
99.9% of the times even with a negative expected value of the position more or
less like the trapeze artist who makes a bit of money every night at the cost of
risking life and limb. Similarly, investing in assets with, say, an abnormally high
P/E ratio may not indicate any form of irrational exuberance on the investors
side but a rational valuation of an asset with, say, a high expected gain as well as a
high probability of loss. 2 Under these conditions, to be sure, there is nothing
preventing as smart analyst from repeatedly issuing a successful contrarian
prediction: in the game of triple-or-nothing, for instance, our analysts could
consistently forecast a loss of $10, and be right 99.9% of the time.
There are unfortunately very few examples of papers in the literature where the
association of the observed time series to the median instead of the mean trajectory
plays a role in the core analysis. One of these few is Roll (1992), which proves that
portfolio managers who are measured against their deviation from a market index
are essentially being forced to track the market median path, which is suboptimal
respective to the mean. A much more direct precedent, however, is Gracia (2005),
which shows how an efficient market subject to a normal random-walk
perturbation may display a persistent, periodic cycle of asset valuation bubbles and
crashes along the median path, even though the mean remains cycle-free. 3
The main purpose of this paper is to develop a model of financial valuations in
an efficient, rational expectations market such that it would result in a persistent
cycle along its median path and hence, given a representative enough sample, in
the observed path. The model is directly inspired in Gracia (2005), although it has
been revised to make it both more general and more parsimonious. It portrays the
_________________________
2
Note that this differs from the rational bubbles literature following Blanchard and Watson (1982),
which requires asset values to diverge from their fundamentals something the model in this
paper does not require.
www.economics-ejournal.org
_________________________
4
www.economics-ejournal.org
As so many economists kept asserting that no one could have seen the latest downturn coming, 6 it became almost a popular pastime in heterodox circles to quote
those who had most egregiously missed it as well as those who had got it most
spectacularly right. There is certainly no shortage of the latter. A 2010 contest in
the Real-World Economics Review Blog, the Revere Award, shortlisted twelve
economists who made particularly accurate predictions of the crash (Dean Baker,
Wynne Godley, Michael Hudson, Steve Keen, Paul Krugman, Jakob Brchner
Madsen, Ann Pettifor, Kurt Richebcher, Nouriel Roubini, Robert Shiller, George
Soros and Joseph Stiglitz). Far from being complete, this list has been heavily
criticized due to some glaring omissions such as Marc Farber, Fred Foldvary, Fred
_________________________
5
For example, when a company invests, say, in a marketing campaign it is impossible to know to
what extent it contributes to improving its market recognition as an alternative supplier (which
would enhance efficiency by fostering competition), to what extent it transfers some demand, and
profits, from its competitors to itself (which might represent a mere transfer, leaving aggregate
rents the same) and to what extent it enhances its oligopolistic market power through product
differentiation (which would increase their oligopoly rents and therefore pull the market further
away from perfect competition). As all three effects would take place simultaneously, net
aggregate impact is uncertain.
For a good survey of such assertions see for example Bezemer (2009).
www.economics-ejournal.org
Harrison, Michael Hudson, Eric Janszen, Raghuram Rajan, Peter Schiff or Nassim
Nicholas Taleb. 7
Many if not most of these authors reached their conclusions by considering
various financial indicators in the context of non-neutral-money, boundedrationality or otherwise inefficient-markets models. Moreover, by relying on
cyclical market patterns some were able to make uncannily accurate predictions
even before the financial indicators would raise any grounds for concern. This is
the case of Foldvary (1997) and Harrison (1997) who, on the basis of the 18-year
cycle Hoyt (1933) observed in the Chicago real estate market, forecasted the 2008
global recession more than a decade in advance. Not that this was a one-off
success either: Harrison (1983), for example, resorted to the same real estate cycle
pattern to predict the 1992 downturn as a follow up from the one in 1974.
Even among the supporters of bounded rationality, the existence of predictable
long-range cycles spanning many years or even decades is frequently regarded
with skepticism. It has not always been so, though: until the 1930s it was
commonly accepted in academic circles that economic waves existed and had
well-defined frequencies. More than 150 years ago, Juglar (1862) observed a trade
cycle with a wavelength of 711 years associated to fixed asset investment. Then,
in the early 20th Century, similar findings came in quick succession: Kitchin
(1923) identified a shorter, inventory-driven cycle lasting around 35 years,
Kondratiev (1926) a long wave lasting 4560 years, and Kuznets (1930) an
intermediate swing lasting 1525 years, which he associated to building activity
(thus linking it to the 18-year property cycle Hoyt 1933 would identify shortly
afterwards). Yet, as Burns and Mitchell (1946) argued, under more strict empirical
tests the evidence was far from conclusive, so the view that business cycles are, as
Zarnowitz (1992) put it, recurrent but non-periodic gradually took hold.
Identifying cycles on the basis of aggregate GDP data is indeed fraught with
technical difficulties, as the data series are either not long enough or not
homogeneous enough. Yet stock market datasets, which are more granular, have
been empirically proven to reflect mean-reversion patterns akin to business cycles.
Thus, for example, Fama and French (1988) as well as Poterba and Summers
(1988) identified a 35 year cycle in stock market returns i.e. the same frequency
_________________________
7
References for each one of these authors predictions can be found, for example, in Bezemer
(2009) and Gaffney (2011) as well as in the Real-World Economics Review Blog:
http://rwer.wordpress.com/2010/03/31/shortlist-for-the-revere-award-for-economics-3/
www.economics-ejournal.org
_________________________
8
Of course, once one has abandoned the constraints of rational behavior it is always possible to
find a set of assumptions to fit any given set of observations, but this goes against Occams Razor
just as a regression with zero degrees of liberty achieves perfect fit but explains nothing at all.
www.economics-ejournal.org
Can an efficient stock market display a periodic cycle of bubbles and crashes?
How could such financial phenomena impact aggregate productivity (and,
through it, aggregate output) in a rational expectations economy?
Section 3 outlines how (a) could happen;9 a rationale for (b) is offered in Section 5.
www.economics-ejournal.org
would be worth more or less if sold in the market instead of being managed within
the companys framework. A straightforward way to measure this cost of
opportunity is the ratio of a companys market price divided by its assets
replacement value i.e. what is known as Tobins q: if the ratio falls below unity,
the firm is actually destroying value and therefore liquidation becomes an
attractive option. 10 Since liquidation (or, in a less dramatic version, reorganization
and dismissal of the worst-offending producers) puts a hard stop to the producers
rent extraction, it represents the investors ultimate weapon; yet, to the extent it is
neither instant nor cost-free, its deterrence power is also limited.
Under these conditions, the producers rent extraction takes place as follows.
In a given production process structure, the producer has a certain degree of
control that translates into a given percentage of the output being siphoned out as
rents. At every given point in time, a certain number of new opportunities to
modify this productive structure will randomly pop up. Other things being equal,
the producers decisions are of course biased in favor of the options that generate
the highest level of rents, but their power to choose is limited by the control
mechanisms imposed by investors. Rational producers will therefore pluck the
goose just enough to maximize their future expected rents, which also means
leaving just enough to keep investors happy. Markets will then price these
companies accordingly, so that the expected growth of their value equals their
discount rate less (plus) the weight of the net cash flow they distribute to (raise
from) investors. The higher this price stands above the threshold value (i.e. the
higher Tobins q ratio), to be sure, the lower the probability that investors
liquidate, and therefore the more freedom of action producers will have to
maximize their rents within the boundaries of the investors control mechanisms.
In an efficient market, the mean path is of course the one where both investors
and producers expectations are fulfilled, so asset values grow precisely at their
market discount rate less the dividends they cash out. Yet, as Appendix 1 proves
for a geometric Wiener process, when the probability diffusion process is
asymmetric the best approximation to the observed time series is the median, not
the mean path And along the median path (as along any path different from the
mean) the players expectations consistently fail to be met, and thus constantly
need to be realigned.
_________________________
10 Needless to say, Tobins q is the standard metric first put forward by James Tobin (1969).
www.economics-ejournal.org
where , 2 , > 0 are positive parameters
d t
= (q t 1)dt
t
(1)
_________________________
11 To be sure, when a companys assets are liquidated they are bought by (and thus incorporated
into) another company, so liquidation does not necessarily lead to a reduction in the volume of
assets put to productive purposes; yet, to the extent the firms liquidated are also those that
extracted the highest rents, the average rate of rent extraction must come down as a result. The
same outcome is of course to be expected from low key liquidations i.e. reorganizations where
the worst-offending producers are fired and the company processes are rearranged to further limit
shirking opportunities.
www.economics-ejournal.org
10
_________________________
12 Note that, to develop this diagram, the specific values of the parameters , 2 and have not
been selected to be realistic but merely to make the shape more evident to the viewer.
13 Although, incidentally, if any of them is infinite then the cycle also disappears by collapsing to
the trivial solution qt = t = 0 .
www.economics-ejournal.org
11
Tobins q
Rent ratio ()
time
Testing Strategy
Separating the value of producers rents from the marginal productivity of their
services is inherently tricky, if nothing else because it is precisely their privileged
knowledge that allows them to extract them from investors, so they must
necessarily be difficult for an external observer to detect. Tobins q is, on the other
hand, not too hard to calculate through various, fairly standard calculation
methods. Several authors (e.g. Wright 2004) argue convincingly that the value
obtained through these methods may be downward-biased due to systematic
overvaluation of asset values at replacement cost (more or less in the same way
that a used car is not worth the same as a new one, so its liquidation value is below
replacement cost). Yet, although this bias might quite possibly compromise a test
of Tobins q convergence to unity, it poses no particular challenge if all we want to
test is whether it does indeed display a cyclical behavior.
The key challenge is finding a data time series that is granular enough yet
covers a period long enough to actually test this. Indeed, however one may want to
www.economics-ejournal.org
12
estimate them, realistic values for the parameters in expression (1) are always
small (equity premiums and average annual variances are typically single-digit,
and the process of liquidating or reorganizing the management of an
underperforming company can also be quite time-consuming), which means cycles
may well stretch over many years. If we want to test whether such cycles do
actually exist, therefore, we need a reliable, homogeneous sample that covers the
span of the underlying cycle several times over so that it can provide a reasonable
degree of confidence to accept or reject the hypothesis.
In addition, there is a question on what test to perform. To prevent any bias
towards acceptance of the hypothesis, we will rule out any test that somehow
forces a positive answer: if nothing else, we should remind ourselves that, per
Fouriers theorem, any continuous function can be decomposed into an
aggregation of sinusoidal curves, so fitting a cyclical function to the sample, or
even decomposing it as a spectrum, does not necessarily prove the point either.
This is also why filters and calibrations are dangerous, as it is not always clear
whether the cycle actually exists in the underlying data or is just a result of the
interaction between the sample and the filter.
To avoid these pitfalls, this paper will resort to a simple autocorrelation test on
the unfiltered data, and demand that its correlograms display the same pattern as
the theoretical model, as illustrated in Figure 2.
Specifically, the autocorrelation function (ACF) should display:
www.economics-ejournal.org
13
Note that although, faced with this correlogram, it might seem natural to resort
to the Box-Jenkins methodology and fit a linear ARIMA function to the sample,
we will not do so here, as it would not help to prove or disprove the papers
hypothesis that the underlying process follows a non-linear expression such as (1).
Test Data and Results
The most complete dataset available for Tobins q is perhaps Wright (2004),
which spans from 1900 to 2002; Wright put forward several computations 16 of
which we will take here the one he calls equity q. Unfortunately an annual series
such as this is not granular enough to test long-wave autocorrelations, so we resort
_________________________
15 Although this may or may not be observable as it quickly falls below the significance threshold.
16 The data can be downloaded from http://www.econ.bbk.ac.uk/faculty/wright/.
www.economics-ejournal.org
14
www.economics-ejournal.org
15
only, due of course to the paucity of data, in this case the correlation wave exceeds
the threshold for 95% confidence a lot less comfortably. This may also explain
why, unlike the CAPE partial autocorrelation function (which essentially matched
the theoretical diagram in Figure 2), the q ratio PACF only displays the first
autocorrelation lag as statistically significant: arguably, this is due to the data
scarcity in this series, which pulls most autocorrelations under the significance
threshold.
Furthermore, this long-wave cycle has quite a high degree of explanatory
power: in fact, even a simple sinusoidal curve with a 31-year wavelength (i.e. the
cycle time suggested by the analysis in Figure 4) has a 72% correlation coefficient
respective to the equity q series or, what is the same, can be said to explain more
Figure 3: CAPE autocorrelation diagrams (monthly data, Jan. 1881 to Oct. 2011)
www.economics-ejournal.org
16
than half ( R = 52% ) of its observed variation. This, however, poses another
question: although we know the correlogram yields symmetric waves even against
an asymmetric wave, if a simple sinusoidal curve already has such a high
explanatory power, how do we know the underlying waves are not sinusoidal
anyway? After all, with a very small change in the basic assumptions of Appendix
2 one could have turned the models median path into a sinusoidal (i.e. symmetric)
wave. 19 The evidence, however, suggests that the actual path behaves just as in a
predator-prey wave, where the fall is steeper (and therefore also shorter in time)
than the climb-up. Indeed, according to Wrights time series there was an increase
of the equity q in 57% of the years in the sample and a fall in only 43% of them. In
2
_________________________
19 Specifically, it would suffice to replace the expression d = ( q 1) dt in Assumption 7
t
t
t
with the expression d t = (ln qt )dt to turn the median path into a pure sinusoidal wave.
www.economics-ejournal.org
17
fact, even if we restrict this calculation to the period 1900 to 1993 (as the tracking
sinusoidal wave appears to reach in 1993 the same point the cycle as in 1900), the
average of climb and fall years still holds: 57% upwards and 43% downwards.
In sum, the evidence supports the hypothesis that the bubbles and crashes
observed for the 100+ years on record could largely be explained through an
efficient-markets, predator-prey cycle model. The underlying model itself may not
necessarily be exactly the one put forward in this paper (to begin with, a model
such as Gracia 2005 would behave in a similar way), but the hypothesis that a
long-wave cycle with steeper slopes downwards that upwards (i.e. with a predatorprey shape) explains most of the observed equity q behavior cannot be rejected.
Two side comments are probably appropriate at this point:
Although this long cycle is consistent with the model in this paper (for equity
premium and return variances are usually single-digit over long periods of
time), it is worth mentioning once again that its identification is not modeldependent. The finding of any cycle under these conditions is hence all the
more robust, as it was not imposed through any filtering, regression or
calibration that could make something visible when it actually does not exist.
This wavelength of just over 30 years is, on the other hand, quite different
from that of most classical studies: it definitely does not match the waves
identified by Kitchin (3-5 years), Juglar (7-11 years), Kuznets (15-25 years) or
Kondratiev (45-60 years). Whether the frequency we have identified here
might result from the combination of others or represent something entirely
different, however, is not a question we can answer on the basis of the
evidence above, and will therefore remain out of scope for this paper.
The next question is how these oscillatory financial phenomena could, under
rational expectations, have an impact not only on GDP growth but also on TFP
(which, per Kydland and Prescott 1991, typically explains 70% of the business
cycle). The standard Cobb-Douglas production function does not lend itself very
well to this (unless modified through ancillary assumptions, of course), as its TFP
growth rate is exogenous ex hypothesi. The answer is, conversely, quite
straightforward on the basis of the aggregate production function put forward in
www.economics-ejournal.org
18
www.economics-ejournal.org
19
activities: as the former enhance overall productive capacity, but the latter
diminish it, the aggregate impact of capital on output is ambiguous, and so its
statistical significance must be (as it actually is) low.
This links up directly with the rent-driven cycle model we developed in
Section 3 and Appendix 2. In good times, producers (meaning anyone with some
control over the means of production) gradually find ways to increase their rent
extraction, which of course imposes additional rigidity on the productive process.
During periods of growth, the weight of these rents over the rest of output (i.e. the
producers rent ratio t ) does not grow very quickly, so its impact on productivity
is also small. Yet, per Figure 1, when the bubble finally bursts and Tobins q starts
its downward spiral, the rent ratio shoots up because their variable portion (i.e. the
marginal cost) drops faster than the fixed costs (that is, the economic rents).
Hence, as the rent ratio goes up, productivity must fall.
In short, the prediction is that financial recessions will impact output growth
through the fall in aggregate productivity they cause. We can now resort to the
production function in Gracia (2011) to show how. Analytically, the function is as
follows:
Yt = At H
1
1+ t
t
(2)
Where Yt represents GDP, H t work hours, t the average rent ratio 22 and At the
productivity factor, which follows the expression:
2 t
= t + t
t dt + s t dWt
2
At
dAt
(3)
www.economics-ejournal.org
20
t2 t
should have a negative sign and therefore, the higher the rent ratio t ,
2
the lower the rate of productivity growth would be. Hence, if we assume
technology growth as well as the difference t2 t to be constant, the median
rent we depicted in Figure 1 translates into the aggregate productivity path shown
in Figure 5.
_________________________
time
t j ,t j ,t
j =1
j,t
j,t is defined as j ,t 2j ,t
j ,t
j ,t
variance of each product / production unit j that is not driven by a technology shock and j,t as
the relative share of labor input corresponding to each one of those products / production units.
www.economics-ejournal.org
21
In other words, productivity must behave over time just like the rent ratio
curve in Figure 1 (only upside down) i.e. displaying a cycle whose dips are steeper
than the previous or subsequent climb up periods, and deeper respective to the
trend than the booms stand above it. This behavior is consistent with the findings
of Nefti (1984), Sichel (1993), Ramsey and Rothman (1996), Verbrugge (1997)
or Razzak (2001), who conclude that business cycles present deepness (i.e.
recessions tend to fall deeper than expansions are tall respective to the trend) and
steepness (i.e. the fall into recession is steeper than the climb up back to
expansion).
One is reminded at this point of Simon Johnsons views on crises in emerging
economies, which he largely based on his experience as chief economist of the
IMF (e.g. Johnson 2009 or Johnson and Kwak 2010). Johnson explains that,
despite the wide diversity of their triggering events, economic crises always look
depressingly similar, because they all result from powerful, privileged elites
overreaching in good times to maximize their rents, but resisting the pressure to
cut back on them when their excessive risk taking results in a credit crisis. Johnson
also makes a very strong case that the U.S. 2008 credit crisis presents exactly the
same profile, with the U.S. financial sector playing the role of the privileged elite.
This is, of course, precisely the sort of mechanism portrayed in Appendix 2 as well
as in Gracia (2005).
Furthermore, these crises are usually associated to low productivity growth
rates during as well as in the years before the recession just as Figure 1 would
suggest. This correlation was first noticed when applying growth accounting to the
Soviet economy (e.g. Powell 1968 or Ofer 1987): post-war growth rates looked
impressive for a long while but, despite the progress of Soviet technology,
productivity barely grew at all from around 1950, and actually fell from 1970
onwards, leading to the system collapse in 1991. Except perhaps for a few die-hard
Marxists, the causal link between the decline and fall of the Soviet economy and
its massive, rent-seeking state apparatus is just beyond question. This famously led
Krugman (1994) to state, on the basis of the observation of comparably low
productivity growth patterns among the so-called East Asian Tiger economies in
the 80s and early 90s (e.g. Young 1992, 1994 and 1995, or Kim and Lau 1994),
that the East Asian rapid economic expansion trend was unsustainable. When, as
predicted, the 1997-98 Asian financial crisis hit and brought the growth period to
an abrupt halt, its root cause was, unsurprisingly, also found largely in the rent-
www.economics-ejournal.org
22
www.economics-ejournal.org
23
centralization, for example. As discussed in Section 3 of this paper, one could even
assume that rational agents will introduce such changes only up to the point where
they neither add nor subtract productive efficiency and no further. Nevertheless, as
long as these changes are rent-generating, they add rigidity to the process and, in a
stochastic world, the higher the rigidity of the productive process (i.e. the heavier
the weight of rents over total output), the greater its exposure to unexpected
shocks, and therefore the higher the probability that a small stochastic shock lead
to a major crash.
The first objective of this paper was to highlight a serious fallacy (the Fallacy of
Diffusion Symmetry) that underpins much of the debate around the Rational
Expectations Hypothesis by fostering a belief that, if markets are efficient,
predictable bubbles and crashes cannot exist. This, it has been shown, is only valid
for symmetric probability diffusion processes, which are rarely, if ever, posed in
modern financial models although the fallacy itself all-too-frequently goes
unacknowledged.
The second objective was to develop a rational-expectations model of periodic
financial bubbles driven by an agency conflict between producers and investors
along the lines of Gracia (2005), test its likelihood against U.S. stock market data,
and then extend it to macroeconomic business cycles by combining it with the
aggregate production function put forward in Gracia (2011).
Both objectives have substantial modeling implications. If the rational
expectations debate is vitiated by the fallacy of diffusion symmetry, it may be
unnecessary to resort to bounded rationality or price stickiness to explain many of
the phenomena put forward as evidence of market inefficiency. On the contrary, it
may make sense to review some classical rational expectations propositions (e.g.
money neutrality, Ricardian equivalence, even perfect market clearing) to assess
under what conditions they would still hold along the median path. Ultimately, the
mean path be as it may, if the median trajectory is the one with the highest
likelihood to match observations, then it arguably makes sense to target it instead
of the mean as the primary objective of economic policy which might, in turn,
www.economics-ejournal.org
24
Analyze the role of credit along the observed path (where it is not necessarily
neutral) and its implications for monetary policy.
Review the models that supported the successful predictions discussed in
Section 2 to identify their key insights, and analyze whether they might be
compatible with a rational expectations framework where the fallacy of
diffusion symmetry has been cast away.
Consider the impact of the production function in Gracia (2011) on the
assessment of fiscal policy, redistribution and long-term growth.
Conduct more empirical work to develop a working model for cycle statistical
inference (subject of course to the limitations of forecasting the median path).
www.economics-ejournal.org
25
APPENDICES
APPENDIX 1
The purpose of this appendix is to show how an empirical analysis of a time series
generated by a geometric Gauss-Wiener diffusion process (i.e. a geometric
Brownian motion) should be expected to yield an observed growth rate closer to
the median, not the mean, path of the distribution.
Consider an asset whose market value Pt follows a geometric Brownian
motion i.e.:
dPt
= dt + dZ t
Pt
(1.1)
dP
dZ
E 0 t = + E 0 t =
dt
Pt dt
(1.2)
= t
d (ln Pt ) =
dt
Pt
Pt
Pt
2 Pt
2
2
2
dPt
2
dt = dt + dZ t
= d (ln Pt ) +
Pt
2
d (ln Pt ) = dt
Pt = P0 e
2
dt + dZ t ln Pt =
2
2
t +Z t
www.economics-ejournal.org
t + Z t + constant
(1.3)
26
2
2
ln Pt = a + bt + u t
(1.4)
a = ln P0
b=
2
u t = Z t
(1.5)
e gT
ln PT ln P0
PT
g
P0
T
(1.6)
www.economics-ejournal.org
27
2
T + Z T
2
ln PT ln P0
g
=
T
T
2
E 0 [g ] =
2
2
+ E 0 [Z T ] =
2
T
(1.7)
Which is, once again, the median growth rate of the distribution, not the mean.
In sum, the observed path may, on average, be very different from this
expected equilibrium without posing any challenge to the efficient markets
hypothesis. 24
Q.E.D.
APPENDIX 2
The purpose of this appendix is to develop in analytical form the reasoning that in
Section 3 of the main text was presented in an intuitive, discursive way.
0< <
2
2
> 0 , whereas the median (and therefore the most likely observed path) would
2
fall at a rate
< 0 . In other words, it might be perfectly rational for investors to buy
2
and hold the asset (as they expect its value to grow at a rate > 0 ) whilst an external observer
grow at a rate
wonders why they keep throwing their money at an investment that consistently loses value.
www.economics-ejournal.org
28
enterprise has two types of stakeholders: investors (i.e. principals) who hold
ownership of the assets, and producers (i.e. agents) who control and manage those
assets on their behalf. Hence, the net value added Yt the entity generates (i.e. the
sum of the return it generates for labor and capital or, what is the same, the
difference between its sales revenue and the market price of its non-labor inputs)
may be broken down into four portions:
Yt mt Lt + Rt + Ct +
dK t
dt
(2.1)
dK
t
Of course this means the investors profit ( t ) is equal to t C t +
.
dt
*
We now designate as K t the capitalized value of the producers control of the
productive process i.e. of the asset represented by their ability to extract rents for
themselves: 25 Therefore, the firms overall rate of return rt is:
Yt mt Lt t + Rt
(2.2)
K t + K t*
K t + K t*
~
We also represent as ~
rt the cost of opportunity of the assets K t i.e. the return they
rt
would yield if invested outside the structure of the company in a risk-free form
~
e.g. lending them for a fixed rent. In this context, C t represents the cash investors
would extract from those assets while invested at a risk-free rate i.e.:
~
~
~ dK t
Ct ~
rt K t
dt
(2.3)
Finally, we define Tobins q ratio as the market value of the company weighted by
the replacement market value of its assets:
_________________________
25 The asset that would be represented by the marginal cost of labour
in this framework, as its cost would always equal its income.
www.economics-ejournal.org
mt Lt
29
K
qt ~t
Kt
(2.4)
And the producer rent ratio t as the value of the producers rents divided by the
value remaining for the investors in the company:
K t*
t
Kt
(2.5)
dK t
E T [rt K t ] = E T Ct +
dt
(2.6)
Where E T [] indicates the mean value per the information available at instant T.
rt dt = dt + dZ t
(2.7)
~
rt = ~
r
(where ~
r constant )
www.economics-ejournal.org
(2.8)
30
The market values K t , K t and K t* are observable at the instant t where they take
~
place, i.e., K t , K t , K t* I t (where I t is the set of information available) i.e.
t T :
lim E T [K t ] = K t
T t
and
[ ]
~
~
lim E T K t = K t
T t
and
[ ]
lim E T K t* = K t*
T t
(2.9)
Comment: This is quite intuitive, as these are all stock variables i.e. they are
defined as values at a point in time (instead of flows between a point in time and
the next).
The cash flows paid to investors ( C t and C t ) and to producers ( Rt ) are known at
~
the instant t in which they take place i.e. C t , C t , Rt I t (where I t represents the
set of information available at time t) or, what is the same, t T :
lim E T [C t ] = C t
T t
and
[ ]
~
~
lim E T C t = C t
T t
and
lim E T [Rt ] = Rt
(2.10)
T t
Comment: In other words, the risk of the return being different from expected at
any given point in time is borne by the retained earnings
www.economics-ejournal.org
dK t
.
dt
31
~
Ct Ct
= ~
Kt Kt
(2.11)
Comment: There is quite a wide range of utility functions that would produce this
result. As an example, Appendix 3 shows its derivation from a standard functional
form (a time-additive discounted expected utility function with unity time
elasticity).
d t = (qt 1) t dt
(2.12)
Liquidation is not an instant process (hence is assumed finite) but, the more
investors find they are losing by not liquidating (i.e. the smaller Tobins q), the
more companies will be reorganized or liquidated to cut down their rents.
On the flip side, of course, the larger Tobins q the least likely are investors to
liquidate or to impose heavy controls, so more opportunities will pop up over
time for producers to increase the rents they extract.
On balance, liquidations will dominate when qt < 1 (i.e. when it is more
profitable for investors to liquidate), whereas producers will have more room
to expand their rents when qt > 1 .
Although the functional form in (2.12) has been chosen primarily because of
its simplicity, it can be justified intuitively if we assume that both the probability
www.economics-ejournal.org
32
Analytical Development:
Combining Assumption 1 (i.e. expression 2.6) with Definition (2.1) we find,
t T :
E T [rt K t ] = E T [ t ]
E T rt K t* = E T [Rt ]
(2.13)
Which, per Assumptions 4 and 5 (i.e. expressions 2.9 and 2.10), becomes, for the
special case T = t :
E t [rt ]K t* = Rt
(2.14)
Rt + C t +
dK t
rt K t + K t* rt (1 + t )K t
dt
(2.15)
E t [rt ] t K t + C t +
dK t
= rt (1 + t )K t
dt
dK t
= rt K t Ct + (rt E t [rt ]) t K t
dt
(2.16)
By simple inspection, we can see that, along the expected path, the impact of qt
will be fully discounted out, for, if we write the expected value of (2.16) at point t
and then apply Assumption 2 (i.e. expression 2.7), we obtain:
dK
E t t = E t [rt K t C t ] + (E t [rt ] E t [rt ]) t K t
dt
=0
dK
E t t = E t [rt ]K t C t = K t C t
dt
www.economics-ejournal.org
(2.17)
33
Which, after integration, yields the familiar Net Present Value formula i.e., t 0
:
E 0 [K t ] = Ae t + E 0 [C t ]e t dt
(2.18)
2
then the median path of expression (2.16) is as follows:
2
2
dK
Median t t =
2
dt
2
K t C t
t Kt
2
(2.19)
At the same time, by combining Definition (2.3) with Assumptions 3, 4 and 5 (i.e.
~
with expressions 2.8, 2.9 and 2.10) we obtain the (deterministic) path of K t i.e.:
~
dK t ~ ~ ~
= r K t Ct
dt
(2.20)
~ 2
~
dqt dK t dK t 1 dK t
=
~ + ~
qt
Kt
2 Kt
Kt
(2.21)
=0
_________________________
26 Note that the model would work just as well if one imposed, as is common in efficient markets
asset valuation models, a transversality condition such that the expected value always equal the
net present value of future cash flows (so that A = 0). This represents a key difference respective
the rational bubbles model Blanchard and Watson (1982) put forward, which can only work if
no such transversality condition exists.
www.economics-ejournal.org
34
~
Ct Ct
dqt
2 ~
2
r dt
~ dt
t dt
Median t
=
2
2
Kt
t
qt
K
=0
dq
2 ~
2
t dt
r dt
Median t t =
2
2
qt
(2.22)
For simplicity, we will designate by t the equity premium, which, for the median
path in expression (2.22), will be the constant value =
2
2
~
r.
On this basis it is now possible to close the dynamic system representing the
median path by combining expression (2.22) with (2.12) in the following final
expression: 27
dqt
2
=
t dt
qt
2
d t
= (qt 1)dt
t
(2.23)
www.economics-ejournal.org
35
APPENDIX 3
The purpose of this appendix is to show how Assumption 6 in Appendix 2 can be
derived from a standard representative consumer utility function within the
parameters most usually applied in mainstream literature.
In the following example, we will assume that the representative consumer
intends to maximize a von Neumann-Morgenstern time-additive discounted utility
function with unity inter-temporal elasticity of substitution, i.e.:
max E 0 (ln C t )e t dt
0
(3.1)
_________________________
29 In Gracia (2005) the assumption was that, instead of ~
rt being constant, it had the same
~
variability as rt so that the difference rt rt represented a mere agency premium
which had the same value for the mean and the median and therefore, if investors were risk
averse, would be guaranteed to be positive also in the median path. Here, conversely, we have
adopted Assumption 2 instead, which is a bit simpler without making much of a difference.
www.economics-ejournal.org
36
K 0 = E 0 C t t dt
0
M0
(3.2)
dM t
2
(r dt + dZ t ) M t e
Mt
t Z t
(3.3)
Thus, the first-order condition for the resolution of this problem is, t 0 :
U (C t ) t e t
2
e =
= e
C t
Ct
t Z t
2
r t +Z t
2
1
=C e
(3.4)
r t +Z t
Ct
dC t
2
= e
= (r )dt + dZ t
C0
Ct
(3.5)
If we now use this to replace into the budget constraint (3.2) we obtain:
C
K 0 = E 0 C t t dt = E 0 C 0 e tt dt = 0
0
0
M0
C0
=
K0
(3.6)
Hence, for any point in time t taken as a reference, under this utility function the
Ct
equals the constant irrespective of the rate of return of the underlying
Kt
~
asset, and therefore the ratio will apply all the same if the asset is K t , so that:
ratio
www.economics-ejournal.org
37
~
Ct Ct
= ~ =
Kt Kt
(3.7)
Q.E.D.
www.economics-ejournal.org
38
References
Acemoglu, Daron and Robinson, James (2012). Why Nations Fail: The Origins of Power,
Prosperity, and Poverty. Crown Business.
Bezemer, Dirk J. (2009). No One Saw It Coming: Understanding Financial Crisis
through Accounting Models. MPRA Paper 15892, University Library of Munich,
Germany. http://mpra.ub.uni-muenchen.de/15892/
Blanchard, Olivier J. and Watson, Mark W. (1982). Bubbles, Rational Expectations, and
Financial Markets. In P. Wachtel (Ed.), Crisis: The Economic and Financial
Structure. Lexington, MA: Lexington.
Boix, Carles (2003) Democracy and Redistribution. Cambridge University Press.
http://www.princeton.edu/~cboix/book-sample.pdf
Burns, Arthur F. and Mitchell, Wesley C. (1946). Measuring Business Cycles. NBER Book
Series Studies in Business Cycles. National Bureau of Economic Research,
Cambridge, MA.
Davidson, Paul (2009). Can Future Systemic Financial Risks Be Quantified? Ergodic vs
Nonergodic Stochastic Processes. Brazilian Journal of Political Economy 29(4): 324
340. http://www.scielo.br/pdf/rep/v29n4/01.pdf
Fama, Eugene (1965). The Behavior of Stock Market Prices. Journal of Business 38: 34
105. http://stevereads.com/papers_to_read/the_behavior_of_stock_market_prices.pdf
Fama, Eugene and French, Kenneth (1988). Permanent and Temporary Components of
Stock Returns. Journal of Political Economy 96(2): 243273.
http://ideas.repec.org/a/ucp/jpolec/v96y1988i2p246-73.html
Foldvary, Fred (1997). The Business Cycle: A Georgist-Austrian Synthesis. American
Journal of Economics and Sociology 56(4): 521541.
http://www.jstor.org/stable/3487330
Gaffney, Mason (2011). An Award for Calling the Crash. Econ Journal Watch 8(2): 185
192. http://econjwatch.org/articles/an-award-for-calling-the-crash
Gal, Jordi (1996) Technology, Employment and the Business Cycle: Do Technology
Shocks Explain Aggregate Fluctuations? NBER Working Paper No. 5721, National
Bureau of Economic Research, Cambridge, MA.
http://ideas.repec.org/p/nbr/nberwo/5721.html
www.economics-ejournal.org
39
Gill, Indermit S. and Raiser, Martin (2012). Golden Growth: Restoring the Luster of the
European Economic Model. The World Bank, Washington, D.C.
Goodwin, Richard M. (1967). A Growth Cycle. In C. Feinstein (Ed.), Socialism,
Capitalism and Economic Growth: Essays Presented to Maurice Dobb. Cambridge
University Press, Cambridge, UK.
Gracia, Eduard (2005) Predator-Prey: An Alternative Model of Stock Market Bubbles and
the Business Cycle. Intervention Journal of Economics 2(2): 77105.
http://econpapers.repec.org/article/intjournl/v_3a2_3ay_3a2005_3ai_3a2_3ap_3a77105.htm
Gracia, Eduard (2011) Uncertainty and Capacity Constraints: Reconsidering the Aggregate
Production Function. Economics: The Open-Access, Open-Assessment E-Journal,
Vol. 5, 2011-19.
http://www.economics-ejournal.org/economics/journalarticles/2011-19
Harrison, Fred (1983). The Power in the Land: An Inquiry into Unemployment, the Profits
Crisis, and Land Speculation. Universe Books, New York.
Harrison, Fred (1997). The Coming Housing Crash. In F.J. Jones and F. Harrison, The
Chaos Makers. Othila Press, London.
Hoyt, Homer (1933). One Hundred Years of Land Values in Chicago. Arno Press, New
York.
Johnson, Simon (2009). The Quiet Coup. The Atlantic Online (May).
http://www.theatlantic.com/magazine/archive/2009/05/the-quiet-coup/307364/
Johnson, Simon and Kwak, James (2010). 13 Bankers: The Wall Street Takeover and the
Next Financial Meltdown. Pantheon, New York.
Juglar, Clment (1862). Des crises commerciales et leur retour periodique en France, en
Angleterre, et aux Etats-Unis. Guillaumin, Paris.
Keen, Steve (2006). The Lily and the Pond. Interview reported by the Evans Foundation.
Available at http://evatt.org.au/news/445.html
Kim, Jong-Il and Lawrence Lau (1994). The Sources of Economic Growth of the East
Asian Newly Industrialized Countries. Journal of the Japanese and International
Economies 8(3): 235271.
http://ideas.repec.org/a/eee/jjieco/v8y1994i3p235-271.html
www.economics-ejournal.org
40
Kitchin, Joseph (1923). Cycles and Trends in Economic Factors. Review of Economics and
Statistics 5(1): 1016. http://www.jstor.org/stable/1927031
Kiyotaki, Nobuhiro and Moore, John (1997). Credit Cycles. Journal of Political Economy
105(2): 211248. http://ideas.repec.org/a/ucp/jpolec/v105y1997i2p211-48.html
Kondratiev, Nikolai (1926). Die langen Wellen der Konjunktur. Archiv fuer
Sozialwissenschaft und Sozialpolitik 56(3): 573609.
Krugman, Paul (1994). The Myth of Asias Miracle. Foreign Affairs 73(6): 6278.
http://www.foreignaffairs.com/articles/50550/paul-krugman/the-myth-of-asiasmiracle
Kuznets, Simon (1930). Secular Movements in Production and Prices. Their Nature and
their Bearing upon Cyclical Fluctuations. Houghton Mifflin, Boston.
Kydland, Finn E. and Prescott, Edward C. (1982). Time-to-Build and Aggregate
Fluctuations. Econometrica 50(6): 13451370.
http://ideas.repec.org/a/ecm/emetrp/v50y1982i6p1345-70.html
Kydland, Finn E. and Prescott, Edward C. (1991). Hours and Employment Variation in
Business Cycle Theory. Economic Theory 1(1): 6381.
http://ideas.repec.org/a/spr/joecth/v1y1991i1p63-81.html
Long, John B. and Plosser, Charles I. (1983). Real Business Cycles. Journal of Political
Economy 91(1): 3969.
http://www.economics.harvard.edu/files/faculty/40_Real_Business_Cycles.pdf
Lotka, Alfred J. (1925). Elements of Physical Biology. Williams & Wilkins Co., Baltimore,
MD.
McNeill, William H. (1982). The Pursuit of Power: Technology, Armed Force, and Society
since A.D. 1000. University of Chicago Press, Chicago, IL.
http://www.press.uchicago.edu/ucp/books/book/chicago/P/bo5975947.html
Nefti, Sahlin N. (1984). Are Economic Time Series Asymmetric over the Business Cycle?
Journal of Political Economy 92(2): 307328.
http://ideas.repec.org/a/ucp/jpolec/v92y1984i2p307-28.html
North, Douglass C. (1981). Structure and Change in Economic History. W.W. Norton,
London.
North, Douglass C. (1990). Institutions, Institutional Change and Economic Performance.
Cambridge University Press.
www.economics-ejournal.org
41
Ofer, Gur (1987). Soviet Economic Growth, 1928-1985. Journal of Economic Literature
25(4): 17671833. http://www.jstor.org/stable/2726445
Olson, Mancur L. (1982). The Rise and Decline of Nations: Economic Growth, Stagflation,
and Social Rigidities. Yale University Press, New Haven.
Olson, Mancur L. (2000). Power and Prosperity: Outgrowing Communist and Capitalist
Dictatorships. Oxford University Press.
Ostrom, Elinor C. (2005). Understanding Institutional Diversity. Princeton University
Press.
Poterba, James M. and Summers, Lawrence H. (1988). Mean Reversion in Stock Returns:
Evidence and Implications. Journal of Financial Economics 22(1): 2759.
http://www.sciencedirect.com/science/article/pii/0304405X88900219
Powell, Raymond P. (1968). Economic Growth in the U.S.S.R. Scientific American 219(6):
1723.
Ramsey, James B. and Rothman, Philip (1996). Time Irreversibility and Business Cycle
Asymmetry. Journal of Money, Credit, and Banking 28: 121.
http://ideas.repec.org/a/mcb/jmoncb/v28y1996i1p1-21.html
Razzak, Weshah A. (2001). Business Cycle Asymmetries: International Evidence. Review
of Economic Dynamics 4(1): 230243.
http://ideas.repec.org/a/red/issued/v4y2001i1p230-243.html
Roll, Richard (1992). A Mean/Variance Analysis of Tracking Error. Journal of Portfolio
Management 18(4): 1322.
http://www.anderson.ucla.edu/documents/areas/fac/finance/1992-2.pdf
Samuelson, Paul (1965). Proof That Properly Anticipated Prices Fluctuate Randomly.
Industrial Management Review 6(2): 4149.
Shea, John (1998). What Do Technology Shocks Do? NBER Working Paper No. 6632,
National Bureau of Economic Research, Cambridge, MA.
http://ideas.repec.org/p/nbr/nberwo/6632.html
Shiller, Robert J. (2005). Irrational Exuberance. (2nd Edition) Princeton University Press.
Sichel, Daniel E. (1993). Business Cycle Asymmetry: A Deeper Look. Economic Inquiry
31: 224236. http://ideas.repec.org/a/oup/ecinqu/v31y1993i2p224-36.html
www.economics-ejournal.org
42
www.economics-ejournal.org
43
Please note:
You are most sincerely encouraged to participate in the open assessment of this article. You
can do so by either recommending the article or by posting your comments.
Please go to:
http://dx.doi.org/10.5018/economics-ejournal.ja.2012-41
The Editor
Author(s) 2012. Licensed under a Creative Commons License - Attribution-NonCommercial 2.0 Germany