ArticlePDF Available

Three Approaches to Macrofinance: Evidence From 140 Years of Data

Authors:
  • Systematic Portfolios LLC

Abstract and Figures

We interrogate 140 years of macrofinancial data from three directions. The first approach pays attention to the slow buildup of financial imbalances that threaten financial stability and isolates medium-term movements in credit and property prices to identify national financial cycles. We show that the data reveals the reemergence of outsized financial cycles since the 1980s. In particular, we document the existence of an outsized financial cycle in the United States during the interwar period. Further, we show that virtually all financial booms are accompanied by housing-finance booms. The second approach pays attention to the predictive information in the consumption to wealth ratio. We show that the consumption to wealth ratio predicts not only global real rates (as has already been shown) but also property excess returns and the term spread although not stock excess returns. The third approach pays attention to the risk appetite of financial market intermediaries whose marginal value of wealth prices a broad class of financial assets. We create a metric for intermediary risk appetite for an earlier period not covered by the extant literature (1920-1970) and show that intermediary risk appetite predicts stock excess returns. We conclude with a discussion of our findings and the path forward for marcrofinance.
Content may be subject to copyright.
Three Approaches to Macrofinance: Evidence From 140 Years
of Data
Anusar Farooqui
December 1, 2016
Abstract
We interrogate 140 years of macrofinancial data from three directions. The first
approach pays attention to the slow buildup of financial imbalances that threaten
financial stability and isolates medium-term movements in credit and property prices
to identify national financial cycles. We show that the data reveals the re emergence of
outsized financial cycles since the 1980s. In particular, we document the existence of
an outsized financial cycle in the United States during the interwar period. Further,
we show that virtually all financial booms are accompanied by housing-finance booms.
The second approach pays attention to the predictive information in the consumption
to wealth ratio. We show that the consumption to wealth ratio predicts not only global
real rates (as has already been shown) but also property excess returns and the term
spread although not stock excess returns. The third approach pays attention to the
risk appetite of financial market intermediaries whose marginal value of wealth prices
a broad class of financial assets. We create a metric for intermediary risk appetite
for an earlier period not covered by the extant literature (1920-1970) and show that
intermediary risk appetite predicts stock excess returns. We conclude with a discussion
of our findings and the path forward for marcrofinance.
1 Introduction
The period since the Western financial crisis has witnessed a dramatic return of the financial
sector to the center of scholarly attention. Three approaches in particular have been
especially fruitful. The first of these pays attention to the build-up of financial imbalances
that can threaten financial stability (Claessens et al. (2010), Drehmann et al. (2012),
Borio (2014), Aikman et al. (2015), and Galati et al. (2016)). In the absence of financial
repression, the excess elasticity of the financial sector generates outsized financial booms.
Financial crises of domestic origin tend to occur at, or near, the peak of financial booms.
Indian Institute of Management—Udaipur, Balicha campus, Udaipur 313001, India. Email:
anusar.farooqui@iimu.ac.in.
1
Recessions that follow financial crises are deep, prolonged and characterized by substantial
lost output. Moreover, even in the absence of financial crises, recessions that follow credit
booms are more painful than run-of-the-mill business cycle recessions. Gaining a better
understanding of financial cycles is therefore of considerable importance.
Drehmann et al. (2012) have shown that the financial cycle is best captured by medium-
term fluctuations in credit aggregates and property prices. Specifically, they find that joint
fluctuations in raw credit, credit-to-GDP and property prices are a strong predictor of
financial crises. Our first goal in the present study is to adapt and extend their analysis to
a much longer time horizon.1We exploit the publication of the macroeconomic database
of Jord`a et al. (2016) and Knoll et al. (2017) to examine national financial cycles for
seventeen OECD countries over 1870-2013. Extending their empirical work beyond the
Second World War reveals the existence of an outsized financial cycle in the United States
during the interwar period. The overall picture is that of emergence, suppression and
re emergence of the financial cycle at the center of the world economy.
We also examine the comovement of national financial cycles. We find that there is
indeed some evidence that national cycles move together. Especially since the 1980s, there
is strong evidence that national financial cycles have become increasingly synchronized.
Further, we emphasize the dominant role played by housing-finance in financial booms.
We develop a measure for housing-finance cycles which we show is contemporaneously
correlated with national financial cycles. We show that the centrality of housing is not
a peculiarity of the United States but rather characterizes all seventeen industrialized
economies for which we have data. Remarkably, we find that every financial boom is
attended by a housing-finance boom. We believe that this is the closest thing we have to
an empirical law in macrofinance.
A second approach seeks to identify a global financial cycle by paying attention to
the predictive information contained in the consumption to wealth ratio of select center
countries. Gourinchas and Rey (2016) have shown that the consumption to wealth ratio
predicts global real rates over the long horizon. We go back in time before the Great War
and examine the period 1870-2010 as a whole.2We show that the consumption to wealth
predicts not only the global real rate but also property excess returns and term spreads.
We also show that it does not predict stock excess returns.
A third approach pays attention to the risk appetite of financial market intermediaries
(Adrian et al. (2011), He and Krishnamurthy (2013), Adrian et al. (2013a) and Adrian
et al. (2013b)). Adrian et al. (2014) and He et al. (2016) have shown that the marginal
value of intermediary wealth prices a broad class of financial assets. We create a metric
for intermediary risk appetite for an earlier period, 1920-1970, not covered by the extant
literature and show that intermediary risk appetite predicts stock excess returns.3
We proceed as follows. Section 2 takes up the first approach, that of Drehmann et al.
1Drehmann et al. (2012) consider the period 1960-2011.
2Gourinchas and Rey (2016) consider the period 1920-2010.
3Intermediary asset pricing has so far been empirically tested of US data from 1968 onwards.
2
(2012), refining and extending it back to the late-nineteenth century. In Section 3, we
develop a measure of the housing-finance cycle and show that it tracks the national financial
cycles of most OECD countries. In Section 4, we construct the consumption-wealth ratio
in the spirit of Gourinchas and Rey (2016), and show that it predicts global real rates, the
term spread and property excess returns but not stock excess returns. In Section 5, we
construct a metric for intermediary risk appetite for the period 1920-1970 and show that
it predicts stock returns. Section 6 concludes with a discussion of the main takeaways and
some future lines of investigation.
2 National financial cycles
Drehmann et al. (2012) identify the national financial cycle with the medium-term compo-
nent in the joint fluctuations of credit and property prices. They show that financial cycle
peaks coincide with financial crises and that the length and amplitude of the financial cycle
have increased markedly since the mid-1980s. Their empirical analysis begins in 1970. Our
goal in this section is two-fold. First, we extend their analysis to the period 1870-2013;
data permitting. In particular, we document the existence of an outsized financial cycle
in the United States during the interwar period. Second, we investigate the comovement
of national financial cycles. We find that, with important exceptions, national financial
cycles do display significant comovement.
Concretely, Drehmann et al. (2012) use a bandpass filter to isolate movements with
a frequency between 8 and 30 years in the average of indices for total credit, credit-to-
GDP ratios and house prices. We adopt their methodology modulo a single change. Since
we are interested in investigating a long period in which there was considerable financial
development, we find it more appropriate to use a log transformation for the raw aggregate
real credit series. Figure A.1 displays the raw and log transformed real credit series for the
United States as well as their bandpass filtered counterparts.4We see that using the raw
series amplifies the more recent and numerically larger movements while tempering earlier
ones. This is not a problem for the shorter period investigated by Drehmann et al. (2012).
But it becomes problematic when looking at a much longer period over which there was
considerable financial development. Put another way, given that the capacity of the real
economy to sustainably absorb a credit expansion has grown with financial development,
medium-term movements in aggregate credit of equal magnitude were more destabilizing
in the early twentieth century than in the early twenty-first century.
Figure 2.1 displays the US financial cycle obtained using the author’s and Drehmann
et al. (2012)’s methodology. The shaded region corresponds to the period 1950-1980 when
there was substantial financial repression. Extending the investigation back beyond the
second world war reveals an outsized financial cycle during the interwar period. The
amplitude of the interwar financial cycle is comparable to that of the 2000s’. In other
4Figure A.2 displays the other two components of the US financial cycle.
3
Figure 2.1: US financial cycle (author’s vs. Drehmann et al. (2012)’s definition)
words, Figure 2.1 tells the story of the re emergence of outsized financial cycles at the
center of the world economy. In the next section, we will see that a very similar story
emerges from a completely dierent approach.
Figures A.3 and A.4 display the national financial cycles for sixteen OECD countries; all
juxtaposed with the US financial cycle to facilitate comparison. Visual inspection reveals
that some countries’ cycles comove with the US cycle, some are loosely correlated, while
others are asynchronous. In the first group, we find the United Kingdom, Denmark, and
Spain. In the second, much larger group, we have Portugal, Netherlands, Norway, France,
Italy, Australia, Finland, Sweden and Switzerland. Finally, in the third group, that are
largely asynchronous with the US, we find Germany, Japan, Belgium and Canada. The fact
that the Canadian financial cycle is asynchronous with its giant neighbor is quite surprising
and no doubt a consequence of Canada’s stringent financial repression.
Out of our sample of 17 advanced economies, 12 had a financial boom peaking around
1989 and 11 had a boom peaking around 2006. This suggests that national financial
cycles exhibit significant comovement. In order to investigate this question we now carry
out a principal component analysis. There are two challenges in such an undertaking.
The first is the paucity of data for many countries; especially for the first half of the
twentieth century. The second challenge is the problem of detrending. Most (unfiltered)
financial cycle data series exhibit a strong positive trend until World War I, violent up-
down movements between and during the world wars, and again a strong positive trend in
the post-war era.
We get around the problem of detrending by using 5-year log returns defined by
Rf
i,t+5 := log (fi,t+5/fi,t ),(2.1)
where fi,t is the average of nation i’s indices for log real credit, credit-to-GDP ratio, and
property prices in year t. We carry out three separate analyses: (a) For 5 advanced
economies we label the G55over 1910-2013; (b) for 12 advanced economies that we label
5US, Denmark, Finland, Sweden, Switzerland.
4
the G126over 1960-2013; (c) for 16 advanced economies that we call the G167for 1981-
2013. This allows us to extend the analysis as far as data allows while letting us check the
consistency of the narrower but longer series against the broader but shorter series in their
periods of overlap.
The first principal component explains 65, 37 and 48 per cent of the variation for (a),
(b) and (c) respectively. Figure 2.2 displays the first principal components (“G12-fpc” and
“G16-fpc”) for (b) and (c). We see that the two are tightly coupled over their period of
overlap; a fact that is not all that remarkable. What is remarkable is that G16-fpc explains
nearly two-thirds of the variation for a broad group of sixteen advanced economies in the
neoliberal era (1980-2013), even though G12-fpc explains roughly a third of the variation
of twelve advanced economies in the postwar period (1960-2013). The implication is that
national financial cycles have become increasingly synchronized over the course of the late
twentieth century. Moreover, the amplitude of the common component of national financial
cycles has increased dramatically since the 1980s.
Figure 2.2: First principal components of the 5-year log returns on the national financial
cycles for the G12 and G16.
Figure A.5 displays three subfigures. Subfigure (a) shows that G5-fpc tracks the much
broader G12-fpc rather closely. Subfigure (b) shows that the G12-fpc loosely tracks the
US cycle in the postwar period. And Subfigure (c) shows that the G5-fpc tracks the US
cycle over the entire century for which data is available. Note that since the G5-fpc is
unweighted, the correlation between it and the US cycle is not due to the US’ weight in
the G5.
The upshot is that there is indeed some evidence for significant comovement in national
financial cycles. Especially since the 1980s, there is strong evidence to suggest that national
financial cycles have become increasingly synchronized. We now turn to the extraordinary
role played by housing-finance in financial booms.
6US, UK, Japan, France, Australia, Belgium, Denmark, Finland, Netherlands, Norway, Sweden, and
Switzerland.
7US, UK, Germany, Japan, France, Australia, Belgium, Canada, Denmark, Finland, Italy, Netherlands,
Norway, Spain, Sweden, Switzerland.
5
3 Housing-finance and the national financial cycle
The extraordinary boom in US mortgage lending of the mid-2000s is widely held to be a
major culprit behind the US financial crisis of 2007-2008. Alongside the mortgage lending
boom, we also saw a sustained rise in house prices. We now show that this was far from
an aberrant episode: Housing-finance has always played an extraordinary role in financial
booms; both in the United States and elsewhere.
As a measure of the housing-finance cycle, we isolate medium frequency movements in
mortgage credit-to-GDP and house prices. Specifically, we use a bandpass filter to capture
movements of frequency between 8 and 30 years in the average of indices for mortgage
credit-to-GDP and national house prices.8Figure 3.1 displays the US housing-finance
cycle and the US financial cycle over 1890-2013. We see that the two are essentially phase-
locked, so to speak. In other words, the broader, multi-variate measure of the financial
cycle contains little additional information beyond what is already available by considering
the property sector in isolation. An obvious question immediately arises: Is the United
States peculiar in this regard?
Figure 3.1: US financial cycle and US housing-finance cycle.
Figures A.6 and A.7 display the housing and financial cycles for sixteen OECD coun-
tries. We see that the United States is far from exceptional. With few exceptions, the two
are indeed tightly coupled. For more systematic analysis we posit the simple linear relation-
ship between the 3-year log return on national financial cycles and national housing-finance
cycles:
Rf
i,t,t+3 := 0+1RHF
i,t,t+3 +"i,t+3.(3.1)
Figure 3.2 displays the R2from estimating the model (3.1) for 17 OECD countries. We note
that in 12 of 17 advanced economies, housing-finance explains more than two-thirds of the
variation in national financial cycles. Germany and Sweden are least dependent on housing-
finance. Although even here, housing-finance explains a third of the variation. We thus
have two kinds of national financial systems in the OCED countries under investigation:
Those where housing-finance is a big part of the story and those where housing-finance is
most of the story.
8Both indices have been set to unity in 1990.
6
Figure 3.2: Proportion of variation in national financial cycles explained by housing-finance.
Visually examining Figures A.6 and A.7, we fail to identify a single financial boom that
is not associated with a property boom. In order to quantify this observation we undertake
a straightforward Boolean analysis. We vary the threshold in ten steps for a financial boom
from slightly below the level the 1980s US boom to slightly below the level of the the US
boom in the 2000s. For each of these ten thresholds we ask what is the proportion of
booms years for all 17 OECD economies under consideration according to the threshold.
If both the housing-finance cycle and the financial cycle clear the threshold we encode
that as a correct prediction. If there is a housing-finance boom unattended by a financial
boom, we encode that as a Type I error. And if there is a financial boom unattended by a
housing-finance boom we encode that as a Type II error. Figure 3.3 displays the results.
Remarkably, even if we take the lowest threshold, only 14 of the 1288 nation-years for
which we have data witnessed a financial boom unattended by a housing-finance boom. And
as we increase the threshold for what constitutes a boom, the number of false negatives
rapidly drops to zero. The available evidence suggests that financial booms are almost
invariably attended by housing-finance booms.9This is the closest thing we have to an
empirical law in macrofinance.
4 Consumption to wealth ratio and asset returns
Gourinchas and Rey (2016) have shown that the consumption-wealth ratio (henceforth
9The converse is not true. Finland had a substantial property boom in the interwar period that was not
accompanied by a broader credit boom.
7
Figure 3.3: Proportion of nation-years spent in financial booms and housing-finance booms
for dierent thresholds.
cw) predicts the global real rate over the long horizon. In what follows we shall refine and
extend their work. We will show that cw predicts the global real rate over the entire period
1870-2010. Furthermore, we shall demonstrate that cw also predicts the term spread and
property excess returns but not stock excess returns.
We construct national consumption to wealth ratios, cwi, via the equation,
cwi
t=wyi
t/cyi
t,(4.1)
where wyiis the wealth-to-GDP ratio of country iobtained from Zucman and Piketty
(2013) and cyiis the consumption-to-GDP ratio obtained from Jord`a et al. (2016). Instead
of taking the average of the G410 as in Gourinchas and Rey (2016), we define cw as
cwt=wtcwUS
t+(1wt)cwUK
t,(4.2)
where wtis ratio of private real US wealth to the sum of US and UK private real wealth
in US dollars at the sterling-dollar market exchange rate.
Our exclusive attention to the Atlantic center countries is motivated by the extraordi-
nary instability of Continental economies between 1914-1945. Subjectively it is clear that
the UK should be given considerably greater weight at the end of the nineteenth than at the
end of the twentieth century. An unweighted average of the UK and the US would therefore
be inappropriate. And since we are interested in wealth booms rather than production,
we choose wealth weights instead of output weights. In other words, instead of making
a judgement on when the baton crossed the Atlantic we let the wealth do the talking.11
10The United States, France, Germany and the United Kingdom.
11Economic historians disagree on when exactly the baton crossed the Atlantic. Some point to the unheard
8
Figure 4.1: The consumption to wealth ratio.
Unless stated otherwise, all quantities refer to the wealth-weighted average of the UK and
the US.
Figure 4.1 displays cw over 1870-2010. We are eectively using the ratio of consumption
to wealth as a method of detrending wealth since consumption is much more stable than
wealth. The ratio cw is mean-reverting over the long horizon. Specifically, cw strongly
predicts the 10-year forward return on cw. To see this consider the simple linear model,
Rcw
t,t+10 =0+1cwt+"t,(4.3)
where Rx
t,t+kis the k-year return on xdefined by
Rx
t,t+k:= xt+k/xt1.(4.4)
Table 4.1 reports the model estimates and Figure A.9 displays the fitted and actual values
for the 10-year return on cw. We see that the estimated slope coecient is negative and
economically and statistically significant. Our interpretation is that the relationship (4.3)
captures the stochastic periodicity of cw over the long horizon.
Table 4.1: 10 year return on cw (1870-2010)
Variable Estimate Std Error t-statistic p-value
Constant 0.889 0.094 9.487 0.000
cw -4.294 0.453 -9.479 0.000
R20.411 adj. R20.406
More generally, there is a strong long-term cycle in cw. Figure 4.2 displays the cycle
obtained by isolating movements with a frequency of between 20 and 120 years in the
of scale of private American fortunes at the turn of the century and place it as early as 1898. Adam Tooze
would perhaps prefer 1916 when European statesmen went hat in hand to Washington. Fernand Braudel
was partial to 1929, when in his view New York replaced London as the focal point of the world economy.
Others, arguing that Britain maintained its central position until the Second World War, would place it
much later, say 1945.
9
consumption to wealth ratio. We use an inverted scale to facilitate comparison with the
US financial cycle. Just as Figure 2.1 revealed the existence of a financial-credit boom in
the interwar years comparable in amplitude to the credit boom in the 2000s, Figure 4.2
reveals the existence of a financial-wealth boom during the interwar period comparable in
amplitude to the extraordinary wealth boom that peaked in 2006.
Figure 4.2: Long-term cycle in the consumption to wealth ratio (inverted scale).
We now document that cw predicts asset returns. Figure A.10 displays stock excess
returns, real-estate excess returns, and the global real rate over 1870-2013.
We begin with the global real rate. Let arm,n
tbe the m-year forward, n-year average
of global real rates,
arm,n
t:= 1
n
n
X
k=1
rt+m+k2.(4.5)
We propose the linear relationship,
ar5,10
t=0+1cwt+2GS
t+",(4.6)
where GS is a dummy for the Gold Standard (1870-1914). Table 4.2 displays the estimates
and Figure A.11 displays the fitted and actual values of the 5-year forward, 10-year global
real rate average. We see that cw is a robust predictor of global real rates that accounts
for nearly three-fifths of the variation in 10-year global real rate average.
Table 4.2: 5-year forward, 10-year global real rate average (1870-2010)
Variable Estimate Std Error t-statistic p-value
Constant -0.255 0.023 -10.876 0.000
cw 1.257 0.112 11.268 0.000
GS 0.040 0.004 10.443 0.000
R20.585 adj. R20.578
The parameter estimate of the dummy for the Gold Standard is economically large and
statistically significant. We submit that a dummy is merited due to persistent deflation
during the Gold Standard that mechanically pushed up real rates. In the last thirty years
of the nineteenth century prices fell by a wealth-weighted average of 21 percent in our
10
center countries. No fewer than 17 out of 30 years saw the price level falling. The only
other vaguely comparable period of multi-year deflation is the initial phase of the Great
Depression (1929-1933). Figure A.12 displays the wealth-weighted average inflation in
1870-2010 in the US and the UK.
Table 4.3: 5-year forward, 10-year global real rate average (1920-2010)
Variable Estimate Std Error t-statistic p-value
Constant -0.277 0.025 -11.209 0.000
cw 1.361 0.117 11.627 0.000
R20.640 adj. R20.635
In order to facilitate comparison with Gourinchas and Rey (2016), we restrict our
sample to the period 1920-2010 and reestimate the model. Table 4.3 displays the estimates.
We see that our G2-based cw is an even stronger predictor than their measure.12 The
restriction to the post-1920 period also yields a better fit than the full sample. But not by
all that much. We will return to real rates in Section 6.
We now turn to property excess returns. We propose the linear relationship,
RH
t+10,t =0+1cwt+2GS
t+",(4.7)
where RH
t+10,t is the 10-year return on house prices in excess of short rates. Table 4.4
displays the model estimates. We see that cw predicts half the variation in property excess
returns. The slope is large, negative, and statistically significant. Also note that as opposed
to real rates, the coecient of the dummy for the Gold Standard is negative. Figure A.13
displays the actual and fitted values of 10-year excess returns on house prices.
Table 4.4: 10-year property excess returns (1870-2010)
Variable Estimate Std Error t-statistic p-value
Constant 0.302 0.030 10.219 0.000
cw -1.451 0.141 -10.314 0.000
GS -0.027 0.007 -4.112 0.000
R20.509 adj. R20.500
Recall that the term spread is a significant predictor of US recessions. We therefore
include a measure of real activity and ask whether 5-year lagged cw adds to our predictive
ability. Our model is given by
spreadt=0+1yt+1 +2cwt4+3GS
t+"t,(4.8)
12Gourinchas and Rey (2016) find an R2of 0.43 when they regress 0-year forward, 10-year average real
rates on their G4-cw.
11
where yt+1 is the (wealth-weighted UK-US) growth rate in year t+ 1. Table 4.5 displays
the estimates.
Table 4.5: Term spread (1870-2010)
Variable Estimate Std Error t-statistic p-value
Constant 0.068 0.014 4.852 0.000
GS -0.009 0.002 -3.793 0.000
yt+1 0.101 0.024 4.217 0.000
cwt4-0.303 0.066 -4.573 0.000
R20.264 adj. R20.247
We can explain a quarter of the variation in the term spread in the full sample (1870-
2010). Remarkably, all independent variables are significant at the 1 percent level.
The term spread may reflect future real rates and it could very well be the case that
cw predicts the spread through the level of the yield curve. In order to test this possibility
we project the spread on 1-year forward growth rate and 10-year real rate average. We
decompose the latter into the sum of two components:
ar0,10
tarcw
tar?
t,(4.9)
where ˆartare the predicted values of the 10-year real rate average in model (4.6) while
ˆar?
tare the residuals orthogonal to the prediction.13 We drop the dummy since that has
already been incorporated into the fitted values. We also include 1-year forward inflation
since that information should be reflected in the spread. Our alternate model for the term
spread is thus given by
spreadt=0+1it+1 +2yt+1 +3ˆarcw
t+4ˆar?
t+"t,(4.10)
where it+1 is wealth-weighted average of inflation in the US and the UK in year t+ 1.
Table 4.6 reports the estimates. We see that variation in the 10-year real rate average
orthogonal to that predicted by cw is statistically insignificant even at the 10 percent level.
Even more surprisingly, 1-year forward inflation is also insignificant and carries the wrong
sign. Meanwhile the fit is a bit poorer than using than the previous model. Figure A.14
displays the actual and fitted values. Note the excess volatility of the term spread.
We now show that cw is a poor predictor of stock excess returns. We estimate the
simple linear model,
Rstock
t,t+10 =0+1cwt+"t+10.(4.11)
Table 4.7 presents the estimates and Figure A.15 displays the actual and fitted values of
10-year stock excess returns. In the next section, we will see that we need to take an
entirely dierent approach to account for stock excess returns.
13Using both lagged cw and the 10-year real rate average results in multicollinearity. Using predicted
12
Table 4.6: Term spread (1870-2010)
Variable Estimate Std Error t-statistic p-value
Constant 0.006 0.002 3.645 0.000
it+1 -0.030 0.023 -1.330 0.186
yt+1 0.092 0.024 3.772 0.000
ˆarcw -0.186 0.050 -3.684 0.000
ˆar?-0.058 0.060 -0.964 0.337
R20.221 adj. R20.195
Table 4.7: 10-year stock excess returns (1870-2010)
Variable Estimate Std Error t-statistic p-value
Constant -0.051 0.053 -0.953 0.342
cw 0.268 0.258 1.038 0.301
R20.008 adj. R20.001
5 Intermediary risk appetite and stock excess returns
The market price of risk has been tied to risk-bearing capacity of leveraged intermediaries.
Specifically, the marginal value of wealth to US securities broker-dealers provides an in-
formative pricing kernel for a broad class financial assets (Adrian et al. (2011), He and
Krishnamurthy (2013),Adrian et al. (2013a), Adrian et al. (2013b), Adrian et al. (2014),
and He et al. (2016)). Empirical tests for intermediary pricing theories have so far relied
on data from the period 1968-2016.
In this section, our goal is to show that intermediary risk appetite is also an important
predictor of stock returns in the period 1920-1970. We define our measure of intermediary
risk appetite, t,by
t+1 := log (BSCt+1/BSCt),(5.1)
where BSCtis the aggregate balance sheet capacity (total assets) of the New York City
Member Banks in year t. This group of intermediaries most closely corresponds to se-
curities broker-dealers who play the starring role in aforementioned analyses. We obtain
annual data on total assets of New York City Member Banks from Banking and Monetary
Statistics, 1914-41 and Banking and Monetary Statistics, 1941-1970, and use the stock
market index from Jord`a et al. (2016). Figure 5.1 displays our measure of intermediary
risk appetite.
Consider first the simple linear relationship,
Rstock
t=0+1t+"t+1,(5.2)
and residual values of the 10-year real rate average allows us to get around the problem of multicollinearity
while still examining both the real rate and cw for predictive information.
13
Figure 5.1: Intermediary risk appetite
where Rstock
tis the log return on US stocks in excess of the risk-free rate. Table 5.1 displays
the estimates and Figure A.16 displays the actual and fitted values. The coecient is
economically large and statistically significant.
Table 5.1: Stock excess returns (1920-1970)
Variable Estimate Std Error t-statistic p-value
Constant -0.034 0.031 -1.085 0.284
Risk appetite⇤⇤⇤ 1.093 0.299 3.657 0.001
R20.225 adj. R20.208
We now show that intermediary risk appetite remains a significant predictor of stock
excess returns even after we control for output growth, credit growth, future recessions and
lagged stock excess returns. Table 5.2 displays the results of this exercise. Recession is a
forward-looking Boolean variable that takes the value 1 in year tif real output in year t+1
is below the real output in year tand zero otherwise; credit is aggregate bank loans; and
the last variable is 1-year lagged stock excess returns.
Apart from the forward-looking recession indicator, our intermediary risk appetite vari-
able is the only statistically significant predictor of stock returns in the period 1920-1970.
14
Table 5.2: Stock excess returns (1920-1970)
Variable Estimate Std Error t-statistic p-value
Constant 0.045 0.050 0.913 0.367
Risk appetite⇤⇤ 0.767 0.362 2.119 0.040
Recession⇤⇤⇤ -0.194 0.076 -2.569 0.014
log(credit) 0.005 0.300 0.016 0.987
log y-0.506 0.539 -0.940 0.353
Rstock(lagged) 0.083 0.164 0.505 0.617
R20.335 adj. R20.256
6 Conclusion
In this paper we have presented some preliminary results from our investigations into 140
years of macrofinancial data using three dierent approaches. We now summarize our
results and discuss future lines of investigation.
The broad story that emerges is one of emergence, suppression and re emergence of the
financial cycle at the center of the world economy. In particular, we have documented the
existence of an outsized financial cycle during the interwar period in the United States
that comparable in amplitude to the US cycle that peaked in 2006. While we remain in a
period of prolonged economic stagnation, automatic stabilizers and policy action by hard-
currency issuing central banks have contained the worst of the fallout. In light of interwar
macroeconomic history, we should count this as a policy success.
We have shown that national financial cycles tend to move together. Especially since the
1980s, national cycles have become increasingly synchronized. This is dramatically evident
from our finding that the first principal component for the financial cycle of sixteen OECD
countries accounts for two-thirds of their variation since 1980. While we have not explored
the mechanism behind this phenomena in this paper, we believe that cross-border (gross)
financial flows are an important part of the explanation (Agrippino and Rey (2013); Bruno
and Shin (2013); Cerutti et al. (2016); Claessens and van Horen (2016)). For instance,
cross-border flows from northern creditors played an important role in amplifying the
extraordinary credit boom in the eurozone periphery in the lead up to the eurozone crises.
We have shown that housing-finance has always played an extraordinary role in financial
booms. The record shows that virtually all financial booms are accompanied by housing-
finance booms. What accounts for this empirical law? We believe that a good place
to begin is the mutually-reinforcing interaction between mortgage lending and property
prices. A rise in property prices increases the collateral value of property justifying further
lending which in turn pushes up property prices and so on ad terrore. The upshot is that
housing-finance is especially susceptible to the excess elasticity of the financial sector. This
is good news because macroprudential measures can be designed to rein in housing-finance
15
relatively easily. Such narrowly-targeted regulation can be expected to be more eective
than counter-measures designed to push back against financial booms across a broad front.
We have shown that the consumption to wealth ratio predicts not only global real
rates (as has already been shown) but also property excess returns and the term spread.
Gourinchas and Rey (2016) have shown that, under minimal assumptions, today’s average
propensity to consume out of wealth encodes information about expected future consump-
tion growth, expect future risk-free rates and expected future risk premia. Both term
spreads and property excess returns have a substantial risk premium component. In light
of their theoretical argument, it makes sense that the consumption to wealth ratio predicts
future spreads and excess returns on property.
We have shown that the consumption to wealth ratio does not predict stock excess
returns and by implication the equity risk premium. This is quite surprising since stock
excess returns are contemporaneously correlated with returns on aggregate wealth and the
consumption to wealth ratio predicts future returns on wealth.14 In order to explain stock
excess returns we turned to an entirely dierent approach.
Intermediary asset pricing pays attention to the risk-bearing capacity of intermediaries
who act as market markers in financial markets. It has been shown that balance sheet
quantities of market-making intermediaries provide an informative pricing kernel that prices
a broad class of financial assets. We construct a metric for intermediary risk appetite for a
period not covered in the extant literature. We show that the metric predicts stock excess
returns over the period 1920-1970 even after controlling for credit growth, output growth,
recessions and lagged stock excess returns. The takeaway is that intermediary risk appetite
is an important predictor of asset returns over a much longer period than has hitherto been
recognized. In particular, it provides a powerful explanation of asset returns even during
the period of otherwise substantial financial repression.
This study has barely scratched the surface of macrofinance. Much empirical and
theoretical work remains ahead. Macrofinance is likely to remain an active area for research
for some time to come.
14Projecting log returns on US real wealth onto US stock excess returns yields an R2of 40 per cent, while
projecting 10-year log returns on wealth on 10-year lagged cw yields and R2of 19 per cent.
16
References
Tobias Adrian, Erkko Etula, and Tyler Muir. Broker dealer leverage and the cross section
of stock returns. Federal Reserve Bank of New York StaReports, 464, 2011.
Tobias Adrian, Emanuel Moench, and Hyun Song Shin. Leverage asset pricing. Federal
Reserve Bank of New York StaReports, (625), 2013a.
Tobias Adrian, Emanuel Moench, and Hyun Song Shin. Dynamic leverage asset pricing.
Federal Reserve Bank of New York StaReports, 625, 2013b.
Tobias Adrian, Erkko Etula, and Tyler Muir. Financial intermediaries and the cross-section
of asset returns. The Journal of Finance, 69(6):2557–2596, 2014.
Silvia Miranda Agrippino and H´el`ene Rey. Funding flows and asset prices in carry trade
economies. 2013.
David Aikman, Andrew G Haldane, and Benjamin D Nelson. Curbing the credit cycle.
The Economic Journal, 125(585):1072–1109, 2015.
Claudio Borio. The financial cycle and macroeconomics: What have we learnt? Journal
of Banking & Finance, 45:182–198, 2014.
Valentina Bruno and Hyun Song Shin. Capital flows, cross-border banking and global
liquidity. NBER Working Paper, (w19038), 2013.
Eugenio Cerutti, Stijn Claessens, and Lev Ratnovski. Global liquidity and cross-border
bank flows. 2016.
Stijn Claessens and Neeltje van Horen. The role of foreign banks in local credit booms.
The Future of Large, Internationally Active Banks, 55:273, 2016.
Stijn Claessens, M Ayhan Kose, and Marco E Terrones. Financial cycles: what? how?
when? In NBER International Seminar on Macroeconomics 2010, pages 303–343. Uni-
versity of Chicago Press, 2010.
Mathias Drehmann, Claudio EV Borio, and Kostas Tsatsaronis. Characterising the finan-
cial cycle: don’t lose sight of the medium term! 2012.
Gabriele Galati, Irma Hindrayanto, Siem Jan Koopman, and Marente Vlekke. Measuring
financial cycles with a model-based filter: Empirical evidence for the united states and
the euro area. 2016.
Pierre-Olivier Gourinchas and H´el`ene Rey. Real interest rates, imbalances and the curse
of regional safe asset providers at the zero lower bound. 2016.
17
Zhiguo He and Arvind Krishnamurthy. Intermediary asset pricing. The American Eco-
nomic Review, 103(2):732–770, 2013.
Zhiguo He, Bryan T Kelly, and Asaf Manela. Intermediary asset pricing: New evidence
from many asset classes. NBER Working Paper, (w21920), 2016.
`
Oscar Jord`a, Moritz Schularick, and Alan M Taylor. Macrofinancial history and the new
business cycle facts. In NBER Macroeconomics Annual 2016, Volume 31. University of
Chicago Press, 2016.
Katharina Knoll, Moritz Schularick, and Thomas Steger. No price like home: Global house
prices 1870-2012. American Economic Review, 2017.
18
Appendix A Figures
Figure A.1: Real credit in the United States. The raw real credit series (top-left) and the log
transformed series (top-right) behave very dierently when the bandpass filter is applied.
The filter attenuates earlier fluctuations since they are of relatively smaller magnitude
(bottom-left). The filter acts more uniformly on the log transformed series where no such
attenuation is evident.
19
Figure A.2: US credit-to-GDP and house price index. The credit-to-GDP series is very
hard to detrend since there is no secular trend. There is a strong upward trend until the
outbreak of World War I; a major crash during the war; rapid growth in the 1920s followed
by a dramatic crash during the Great Depression and World War II. The rapid postwar
recovery lasts until the 1960s and from there on the series appears not to exhibit a secular
trend. This is a common feature of all countries examined in this study.
20
Figure A.3: National financial cycles (blue) and the US financial cycle (dotted, red). The
German, Japanese, Belgian and Canadian cycles are manifestly asynchronous with the US
cycle. On the other hand, British, Australian and Danish cycles exhibit strong comovement
with the US cycle.
21
Figure A.4: National financial cycles (blue) and the US financial cycle (dotted, red). Note
the scale of the mid-2000s Spanish boom.
22
Figure A.5: First principal components (fpc) of the 5-year log returns of national finan-
cial cycles. G5 includes US, Denmark, Finland, Sweden and Switzerland. G12 includes
the G5 countries as well as the UK, Japan, France, Australia, Belgium, Norway and the
Netherlands. All series have been standardized to facilitate visual comparison.
(a) G5-fpc tracks the much broader G12-fpc rather closely.
(b) G12-fpc loosely tracks the US cycle.
(c) G5-fpc tracks the US cycle over a much longer period.
23
Figure A.6: National financial cycle (blue) and housing-finance cycle (dotted, red). Note
the absence of financial booms unattended by housing-finance booms.
24
Figure A.7: National financial cycle (blue) and housing-finance cycle (red). Note the
absence of financial booms unattended by housing-finance booms.
25
Figure A.8: 10-year returns on wealth and cw are contemporaneously correlated.
Figure A.9: cw predicts 10-year forward, 10-year return on cw. It is therefore strongly
mean-reverting.
26
Figure A.10: Stock and real-estate excess returns, term spread and global real rate.
Figure A.11: cw predicts 5-year forward, 10-year global real rate average.
27
Figure A.12: Inflation in the center countries (1870-2010)
Figure A.13: cw predicts 10-year returns on property.
28
Figure A.14: cw predicts 5-year forward term spread.
Figure A.15: cw is a poor predictor of 10-year stock excess returns.
29
Figure A.16: Intermediary risk appetite predicts stock excess returns.
30

Supplementary resource (1)

... Shin (2010, 2014),Adrian et al. (2011Adrian et al. ( , 2014a,Adrian et al. ( , 2013Adrian et al. ( , 2014b,Bruno and Shin (2013),Etula (2009Etula ( , 2013, Wojnilower (2015),He and Krishnamurthy (2013),He et al. (2016) andFarooqui (2016). ...
Article
Full-text available
We examine the pricing performance of various measures of dealer risk appetite. We construct a novel measure of risk appetite that captures the provision of leverage to the clients of broker-dealer firms. We also construct a factor mimicking portfolio that tracks shocks to the leverage of broker-dealers. We show that intermediary pricing models convincingly outperform benchmark models for a class of test assets which the benchmark factors were specifically designed to price.
Thesis
Full-text available
The post-GFC environment of depressed interest rates and growth in the United States and the euro area is a cause for concern. Drawing on a large body of literature, this paper synthesizes the research on secular stagnation, decomposes the multi-decade downtrend in natural interest rates into structural and cyclical components, and briefly outlines policy prospects for both economies. The results suggest that subdued trend growth and investment explain most of the decline in natural interest rates since the 1980s. In this respect, a forecast conditional on the outlook for both drivers diverges from a strict interpretation of secular stagnation but not from a new normal subject to significant risks.
Article
Full-text available
Credit cycles have been a characteristic of advanced economies for over 100 years. On average, a sustained pick-up in the ratio of credit to GDP has been highly correlated with banking crises. The boom phases of the cycle are characterised by large deviations in credit from trend. A range of mechanisms can generate these effects, each of which has strategic complementarity between banks at its core. Macro-prudential policy could curb these credit cycles, both through raising the cost of maintaining risky portfolios and through an expectations channel that operates via banks’ perceptions of other banks’ actions. This article is protected by copyright. All rights reserved.
Article
In advanced economies, a century-long, near-stable ratio of credit to GDP gave way to rapid financialization and surging leverage in the last forty years. This “financial hockey stick” coincides withshifts in foundational macroeconomiCrelationships beyond the widely noted return of macroeconomiCfragility and crisis risk. Leverage is correlated withcentral business cycle moments, whichwe can document thanks to a decade-long international and historical data collection effort. More financialized economies exhibit somewhat less real volatility, but also lower growth, more tail risk, as well as tighter real-real and real-financial correlations. International real and financial cycles also cohere more strongly. The new stylized facts that we discover should prove fertile ground for the development of a new generation of macroeconomic models witha prominent role for financial factors. © 2017 by the National Bureau of Economic Research. All rights reserved.
Article
We investigate global factors associated with cross-border capital flows. We formulate a model of gross capital flows through the international banking system and derive a closed form solution that highlights the leverage cycle of global banks as being a prime determinant of the transmission of financial conditions across borders. We then test the predictions of our model in a panel study of 46 countries and find that global factors dominate local factors as determinants of banking sector capital flows.
Article
The financial cycle captures systematic patterns in the financial system and is closely related to the concept of procyclicality of systemic risk. This paper investigates the characteristics of financial cycles using a multivariate model-based filter. We extract cycles using an unobserved components time series model and applying state space methods. We estimate financial cycles for the United States, Germany, France, Italy, Spain and the Netherlands, using data from 1970 to 2014. For these countries, we find that the individual financial variables we examine have medium-term cycles which share a few common statistical properties. We therefore refer to these cycles as 'similar'. We find that overall financial cycles are longer than business cycles and have a higher amplitude. Moreover, such behaviour varies over time and across countries. Our results suggest that estimates of the financial cycle can be a useful monitoring tool for policymakers as they may provide a broad indication about when risks to financial stability increase, remain stable, or decrease.
Article
How have house prices evolved over the long run? This paper presents annual house prices for 14 advanced economies since 1870. We show that real house prices stayed constant from the nineteenth to the mid-twentieth century, but rose strongly and with substantial cross-country variation in the second half of the twentieth century. Land prices, not replacement costs, are the key to understanding the trajectory of house prices. Rising land prices explain about 80 percent of the global house price boom that has taken place since World War II. Our findings have implications for the evolution of wealthto- income ratios, the growth effects of agglomeration, and the price elasticity of housing supply.
Article
We adopt an unobserved components time series model to extract financial cycles for the United States and the five largest euro area countries over the period 1970–2014. We find that financial cycles can parsimoniously be estimated by house prices and total credit or the credit-to-GDP ratio. We show that these medium-term cycles are longer and have larger amplitudes than business cycles, and that their length and amplitude vary over time and across countries.
Article
The literature traditionally uses US monetary and financial factors as indicators of global financial conditions. This paper explores whether similar European factors also need to be considered when studying the behaviour of cross-border bank flows. Using a longer time series and broader country sample than previous studies, we confirm that flows vary with uncertainty (VIX), US monetary policy (real interest rate and term spread), and US exchange rate. In contrast to the existing literature, we find that US bank conditions are insignificant in explaining flows outside the global financial crisis. European bank conditions (euro-area and UK large bank leverage, or TED spread, the three-month interbank rate minus three-month government bond yield) are, however, important throughout the 2000s, even outside the crisis and when controlling for commonality in global conditions. Taken together, our results suggest that global financial conditions are best captured by US monetary conditions and exchange rate dynamics and European bank conditions. This finding is consistent with the important role of European banks in intermediating cross-border credit, including dollar-denominated credit.
Article
It is high time we rediscovered the role of the financial cycle in macroeconomics. In the environment that has prevailed for at least three decades now, it is not possible to understand business fluctuations and the corresponding analytical and policy challenges without understanding the financial cycle. This calls for a rethink of modelling strategies and for significant adjustments to macroeconomic policies. This essay highlights the stylised empirical features of the financial cycle, conjectures as to what it may take to model it satisfactorily, and considers its policy implications. In the discussion of policy, the essay pays special attention to the bust phase, which is less well explored and raises much more controversial issues.