Business Cycles, Credit Cycles, and Asymmetric Effects of Credit Fluctuations Evidence From Italy For The Period of 1861-2013

Journal of Macroeconomics 61 (2019) 103130
Contents lists available at ScienceDirect
Journal of Macroeconomics
journal homepage: www.elsevier.com/locate/jmacro
Business cycles, credit cycles, and asymmetric effects of credit

T
fluctuations: Evidence from Italy for the period of 1861–2013
Silvana Bartolettoa, Bruno Chiarinib, , Elisabetta Marzanob,c, Paolo Pisellid
⁎
a
University of Naples “Parthenope” and CNR-ISSM, Naples, Italy
b
University of Naples Parthenope, Naples, Italy
c
CESifo, Munich, Germany
d
Bank of Italy, DG Economics, Statistics and Research, Rome, Italy
ARTICLE INFO ABSTRACT
Keywords: We propose a joint dating of Italian business and credit cycles on a historical basis by applying a
Business fluctuations local turning-point dating algorithm to the level of the variables. In addition to short cycles
Financial cycle corresponding to traditional business cycle fluctuations, we also investigate medium cycles be-
Bank credit cause there is evidence that financial booms and busts are longer and more persistent than
Medium-term fluctuations
business cycles. The results show that medium-term cycles account for the largest part of fluc-
tuations of the Italian credit cycle. Second, we find evidence that credit and business cycles are
JEL classification:
weakly synchronized in the medium term, whereas they steadily comove in the short term, when
E32
E44 the GDP cycle is leading the bank credit cycle. However, the study found that, over the medium
N13 cycles, economic downturns associated with credit crunches and financial disruption are more
N14 severe than other types of economic downturns. Finally, by modelling interaction between credit
and business cycles in a simple threshold VAR model, we confirm that GDP response to credit
shock is much more intense in the recessionary phase of credit cycle.
1. Introduction
Following the international crises of 2007–2008, there has been renewed interest in the connection between real and financial
variables. Historically, the study of the business cycle has focused on the behavior of macroeconomic data, with cycles lasting no
more than 8 years on average (A'Hearn and Woitek, 2001; Zarnowitz, 1992). Although the idea of long swings in economic activity
stems back to the ideas of economists such as Kuznets, Abramovitz and Schumpeter, and the idea of long financial swings was
discussed in Minsky (1964), recent studies have shown that real and financial cycles interact at lower frequencies than those of the
traditional business cycle (Aikman et al., 2015; Drehmann et al., 2012).
In this paper, using a local turning-point dating algorithm based on the level of variables (or the NBER approach), we propose a
joint dating of Italian business and credit cycles during the last 150 years (1861–2013). Our study is innovative in two ways. First, we
focus on two types of cyclical patterns, namely those that have the same periodicity as the business cycle (short cycles) and those that
have considerably longer periodicity (medium cycles). With regard to the latter, we find support for our hypotheses using spectral
analysis. Second, while previous works on Italy have considered either only the credit (De Bonis and Silvestrini, 2014) or business
cycle (Clementi et al., 2014), this study examines both cycles together. To our knowledge, so far only one study has proposed a joint
dating of business and credit cycles with a specific focus on Italy (Bulligan et al., 2017), but the authors analyze only recent years,
⁎
Corresponding author.
E-mail address: bruno.chiarini@uniparthenope.it (B. Chiarini).
https://doi.org/10.1016/j.jmacro.2019.103130
Received 12 September 2017; Received in revised form 31 May 2019; Accepted 4 June 2019
Available online 05 June 2019
0164-0704/ © 2019 Elsevier Inc. All rights reserved.
S. Bartoletto, et al. Journal of Macroeconomics 61 (2019) 103130
using quarterly data from 1970 to present day.

The dating of the turning points is a preliminary step to investigate, using historical narratives and econometric estimates, how
comovement changes with the definition of the cycle and the real impact of the credit cycle. In this regard, we first examine the
degree of synchronization between the credit cycle and GDP cycle (short and medium) using parametric and non-parametric methods
(Harding and Pagan, 2002 and 2006) as well as Granger causality. Then, we shift our attention to the rate of growth of GDP to
investigate whether the short and medium credit cycles can explain the short-term economic performance observed during the period
1861–2013. Specifically, we first examine how the probability of a recession varies conditional on the credit cycle and to what extent
the rate of growth of GDP in recession decreases conditional on a credit contraction. Second, we analyze the real impact of credit
cycles in a simple VAR model to detect possible nonlinear responses of GDP dynamics to the phases of credit cycles.
The results of our study show that medium-term fluctuations account for the largest part of the Italian credit cycles. Second, we
find evidence that credit and business cycles are weakly synchronized in the medium term, whereas they steadily comove in the short
term, when the GDP cycle is leading the bank credit cycle. However, we found that, over the medium cycles, economic downturns
associated with credit crunches and financial disruption are more severe than other types of economic downturns. Finally, we found
strong evidence that the examined cycles express a non-negligible amount of non-linearity. By exploiting a simple asymmetric VAR
model, we find that while GDP is totally unaffected by credit expansions, it is significantly downsized by credit contractions.
The rest of the paper proceeds as follows: Sections 2 and 3 describe the data and the dating methodology and provide evidence of
long-lasting fluctuations in the series under scrutiny. Section 4 presents the results of the turning-point analysis for the business and
credit cycles. Section 5 details descriptive evidence regarding the business and credit cycles and presents results regarding syn-
chronization/comovement1 and Granger causality. Section 6 discusses the real impact of credit cycles as estimated by univariate and
multivariate analysis, which considers possible asymmetric responses of GDP to credit shocks (threshold VAR model). Section 7 draws
conclusions from the study's results.
2. Data
We analyze the behavior of GDP as the representative variable for the business cycle. One of the novelties of our work is that we
explored the main features of the fluctuations in GDP using the annual time series reconstructed by the Bank of Italy (Baffigi, 2013),
which is a unique series for the entire period 1861–20132.
Also for the analysis of the credit cycle, we relied on a new long-term series of bank loans (De Bonis et al., 2012). In this series,
loans include lending granted to households, non-financial corporations and government.3 Hence our bank loans are not strictly
“credit to the private sector” (the usual benchmark aggregate in the macroeconomic analysis), because they include loans to the
public sector. This latter series cannot be excluded from the loans series since data on loans by Institutional sector are not available
before WWII.
Loans have been reconstructed net of bad debts because of the difficulty of finding data in the past. Credit aggregate, i.e. total
bank loans, is the key variable in our analysis. In addition, we compile the aggregate sum of loans and public bonds held by banks,
which for convenience of exposition, we will call “bank claims”.4 As regards public bonds held by banks, we combine some series
from the Bank of Italy's Statistics Data Base. For the period 1861–1993, we use the series “General Government: securities held by
other monetary financial institutions” (other than the Bank of Italy) from the Bank of Italy (2014a). This is the longest updated series
of government bonds in banks’ balance sheets, and largely coincides with amounts drawn from other sources (Cotula et al., 1996;
Garofalo and Colonna, 1999). For the remaining years, we draw on the series “Public sector securities held by banks” from the
Bank of Italy (2014b) after excluding the holdings of Cassa Depositi e Prestiti.5
We look at bank loans and bank claims series only, mainly because of a lack of long time series for other assets. Nonetheless, this is
not a serious limitation to our analysis for at least two reasons. First, Italy's financial system has traditionally been more dominated by
credit institutions than those of other advanced economies (Ciocca and Biscaini Cotula, 1994; Rajan and Zingales, 2003). Second,
recent studies have emphasized the role played by credit aggregates in providing information about the likelihood of future financial
crises (Schularick and Taylor, 2012; Borio, 2012).
Often the financial cycle has been dated and analyzed starting from the series of credit to GDP ratio (Drehmann et al., 2012; De
Bonis and Silvestrini, 2014). Beyond this, the level of stocks of financial assets (bonds, stock market capitalization, loans) relative to
GDP may help to investigate many far-reaching economic issues. For instance, they are a measure of financial development or
financial deepness often employed in cross-country comparisons (e.g. Mendoza and Quadrini, 2009) and in a historical perspective
(Schularick and Taylor, 2012). Moreover, by reporting the stock of the financial variable, namely credit, to the GDP has been proven
to be an important source of information for macro prudential purpose, since it is a reliable indicator of financial distress (Drehmann
1
These two terms are used interchangeably throughout this paper.
2
In a previous work on Italian business cycle Delli Gatti et al. (2005) used data from Ercolani (1975) for the years 1861–1890; from
Rossi et al. (1993) for the period 1890–1990; and from ISTAT (2001) for the 1990s.
3
Although in the first few years after unification a significant proportion of total lending to the economy was granted by the banks of issue, we do
not include their loan activity in our aggregates (see, e.g., De Bonis and Silvestrini, 2014). Interbank loans are excluded.
4
This is only a definition of convenience to make it clear that we are including something other than loans in this credit aggregate.
5
Cassa Depositi e Prestiti SpA (CDP), the Italian development bank, is a joint-stock company under public control, and is usually excluded from
private credit series. See the Bank of Italy's Financial Stability Report (Bank of Italy, 2013).
2
and Tsatsaronis, 2014; Bartoletto et al., 2019).

In this paper, in order to date the financial cycle, we inspect the pattern of the credit series, a stock variable, like in
Claessens et al. (2012). The main reason for this choice is that by employing the credit-to-GDP ratio would prevent to separate real
and financial fluctuations. As a result, this would make more difficult to investigate the different timing of turning points (a drop in
GDP might result in a peak in credit/GDP ratio), and might engender a major flaw in the analysis, since the main goal of our article is
to compare business and credit cycles.
All series are in real terms since we have deflated them using a GDP implicit deflator (Baffigi, 2013).6
3. Short and medium cycle
3.1. Methodology
To identify business and financial cycles, we use the classical definition of a business cycle used at NBER and introduced by
Burns and Mitchell (1946).
This approach to business cycle analysis is quite new in the literature on the Italian economy over a long time span. In fact, an
official dating of the Italian business cycle is not available over such a long period. The cyclical chronology according to the NBER
approach has been maintained by ISCO since 1945 (ISCO, 1962; Bruno and Otranto, 2004) and today by Istat (2011). Moreover, there
is no dating of the Italian business cycle according to the classical approach based on annual data, except for the recent work by
Jordà et al. (2013). Their analysis is similar to ours, but their historical sources are different and not as homogeneous as the recent
reconstructions of GDP and bank loans series used here and described in Section 2. Clementi et al. (2014) put forward a business cycle
analysis based on Baffigi's (2013) reconstruction, but they identify turning points in the deviation from trend series (growth cycles).
In Appendix A.1, we evaluate our business cycle dating by comparing it with other historical analyses of the Italian business cycle
using other approaches.
The advantage of using the classical methodology arises from the fact that it is a “local” procedure. As a result, the turning points
that are identified are robust to the inclusion of newly available data with respect to other filter-based procedures. More importantly,
turning point identification relies on data around the date considered, not involving observations over a long time, as is the case with
filter-based or detrending procedures (Canova, 1998). In fact, this can be questionable when data span different historical periods,
because their quality inevitably changes over time, and the evaluation of magnitudes in the distant past at constant prices gives rise to
distortions. Although these shortcomings are somehow unavoidable when long historical periods are involved, we still think that a
classical approach to business cycle dating can mitigate them.
The specific cycle-dating algorithm we use is that introduced by Harding and Pagan (2002), who extend the so-called BB algo-
rithm developed by Bry and Boschan (1971) to identify the turning points in the (log) level of a series. Here, we modify Harding and
Pagan's procedure by adapting it to annual data and considering two different cycles, a short cycle and a medium cycle.
For the short cycle, we follow Watson (1994), who first adapted the BB algorithm to annual data: contractions are defined as
sequences of absolute declines in the series, and expansions are defined as sequences of absolute increases. One year being the
minimum time unit, the minimum duration of a phase is 1 year and that of a complete cycle (peak to peak) is 2 years. To capture
cycles that are longer than those typically considered in business cycle analysis, we also define a medium7 cycle by adjusting
Harding and Pagan's (2002) method and applying it to yearly data. We define peaks (troughs) at time t if they fulfill two conditions:
(i) each cycle has a minimum length of 4 years; and (ii) each phase (expansion or contraction) is at least 2 years long.
It is important to note that these two definitions of a cycle are seemingly arbitrary. Not only is the procedure consistent with the
practice at the NBER, but given the frequency of the data the short cycle is also the shortest that can be identified using annual data.
Similarly, once the short cycle is defined, the medium cycle is the closest one in terms of the minimum length of the phase. In other
words, given these two cycles, it is not possible to identify a cycle in between them. We purposely avoid to set an upward threshold
for cycles’ length since, differently from a growth-cycle approach, here we want the data lead us to document the existence of major
swings in the real activity and in lending, to detect in a narrative approach, historical episodes that originated them.
Finally, one or more short cycles are exactly nested within each medium cycle as a result of the relative minimum/maximum
criterion: a turning point of a medium cycle is always also the turning point of a short cycle. Although the two cycle categories are
mutually exclusive when considering a specific medium cycle, this does not occur when considering the full sample: it can happen
that a medium cycle identified in a certain historical period may have a briefer duration than a short cycle pertaining to a different
medium cycle. Technically speaking, this is due to the absence of a censoring rule for the maximum length of the short cycle. As well
pointed out by Pagan (1997) and Harding and Pagan (2002) the expected duration of a classical cycle depends upon the size of the
trend growth (direct relationship) and the size of the shocks (inverse relationship). Of course, when comparing long historical periods
the stochastic structure is almost certainly not invariant, and we can expect a variegated cycle length.
6
Data at constant prices and GDP deflator are reported in Appendix A.4.
7
As is made clear in the next paragraph, we follow the literature by using the label “medium” to denote a cycle that is longer than the usual
business cycle.
3
3.2. Evidence of medium-term fluctuations
Comin and Gertler (2006) and Blanchard (1997) pointed out that over the postwar period, many industrialized countries, Italy
included, have tended to oscillate between phases of robust growth and relative stagnation, displaying longer oscillations than those
typically considered in conventional business cycle analysis. They proved the importance of the medium-term component of fluc-
tuations in GDP exceeding the short-term component.8 Finally, they underlined that conventional business cycle detrending methods
tend to sweep these kinds of oscillations into a linear trend, which instead exhibit considerable variation, reflecting the presence of
significant cyclical activity at the medium frequencies.
Aikman et al. (2015) presented some empirical evidence across countries (Italy among them) over the last 150 years stressing how
credit cycles (measured by variations in the ratio of bank lending to GDP) are distinct from business cycles in their frequency and
amplitude because fluctuations in credit to output operate over the medium term, that is, beyond the business cycle frequency, with
peak-to-trough cycles completed over the course of a decade or more. Finally, Drehmann et al. (2012) showed that the medium-term
component of fluctuations is more important in the joint behavior of credit and property prices than GDP.
With reference to our real and credit series, we provide some evidence of the role played by medium-term fluctuations in our data.
Following Aikman et al. (2015), we compute the spectral density of GDP and the credit series available to us. This analysis, based on a
simple tool, evaluates the weight of the fluctuations at different frequencies in accounting for variability (variance) over time of our
series. However, spectral analysis requires a stationary process, and this raises the issue of detrending, which, as mentioned above, is
not irrelevant for the cyclical properties of resulting series. As is standard in the literature, we analyze the spectrum of difference-
stationary series (growth rates) in order to emphasize the differences in the cyclical properties of real and credit variables, bearing in
mind that our cyclical dating is drawn from the (log) level of the series.9
Fig. 1 plots the estimates of the spectral density of growth rates of GDP and the two credit series, namely loans and bank claims.
The horizontal axis shows the frequency normalized between 0 and 1, and the vertical axis shows the weight of each frequency. A
roughly flat spectrum means that every frequency contributes to the variance of the series (absence of cycles), whereas when a
specific frequency or range of frequencies accounts for the spectrum more than others, it features a peak at those frequencies, which
defines the period of the underlying cycle.
Despite the limited sample size, the spectral density is suggestive of the empirical relevance of medium-term variations in our
series. In all cases, peaks lie close to zero frequency, which implies the presence of long-lasting cyclical components, while the
contribution of short cycles is generally very low. In particular, the peak in GDP spectrum is at frequency f = 0.17, which entails a
period of 2(π)/ f = 13 years.10 Overall, credit variables show even longer cyclical components, with peaks in the spectra at fre-
quencies less than 0.1. In these cases, the main cyclical components are all about 30–35 years.
From this preliminary analysis of cyclical properties, three main regularities emerge. First and foremost, long-lasting fluctuations
are the main cyclical component in our annual centenarian time series in both real and financial series, and they are much longer than
those studied in standard business cycle analysis. This evidence underlies our focus on medium-term cycles along with the usual
business cycles. Second, business and financial cycles feature different cyclical regularities. Credit fluctuations are generally longer
than business cycles, and the contribution of shorter fluctuations, 4 years or less (f = 0.5 or more), is modest compared with GDP
fluctuations. Finally, despite the fact public bonds held by banks respond to different (shorter-term) investment needs compared with
loans, the two credit aggregates display a very similar spectral density distribution.11
4. Business and credit cycles dating
4.1. The business cycle
Table 1 reports the main cyclical features of the Italian business cycle over the period 1861–2013. Medium cycles range from 4 to
68 years, while short cycles last between 3 and 35 years. Upturns are generally much longer than downturns, both in the short and in
the medium term: expansions last up to 62 years, as in the post-WWII cycle, while recessions do not exceed 6 years. While there is no
clear tendency in medium-cycle duration, it seems that the length of the short cycles has increased, as witnessed by the fact that in the
80 years before WWII, the Italian economy went through 12 different cycles, but only five in the next 70 years.
The first medium cycle runs from 1866 to the year prior to the outbreak of World War I, with a minimum peak in 1872 and an
expansion lasting until 1913. In the period considered, there is a huge potential for growth: during the 41 years of expansion
(1872–1913), real GDP grew by 107% at an annual rate of 1.8%. This growth shows surprising stability when analyzed in terms of the
five short cycles that make up this long period (see Appendix A.3 for the key events underlying the turning points).
8
In a growth cycle approach, short-term cycles are those lasting up to 8 years, while medium cycles are those lasting between 8 and 50 years
(Comin and Gertler, 2006).
9
A useful reference in the case of historical time series is A'Hearn and Woitek (2001). As a robustness check, we also carried out an analysis on the
transformed binary series of expansions and recessions, which are stationary, although in this case, the periodicity of fluctuations is partly imposed
by construction. The results confirm the main results of the growth rate analysis.
10
We neglect the peak at zero frequency, because it simply refers to the variance of the white noise error term (stationary) of the differenced I(1)
component of the series (Cochrane, 1988).
11
In the working paper version of this article (Bartoletto et al., 2017) it is possible to find a complete description of the main features of the series
“public bonds held by banks”.
4
Fig. 1. Standardized spectral density (growth rates).

Note: Standardized spectral density, obtained by smoothing the sample periodogram of growth rates. Smoothing with a Parzen window with lag
parameter = 35.
Table 1
GDP: basic features of medium and short cycle.
Cycles Turning points Duration (years) (1) Amplitude (2) Slope (3)
Peak 1 Peak 2 Trough Downturns Upturns Cycle Downturns Upturns Downturns Upturns
Medium cycle 1 1866 1913 1872 6 41 47 −4.1 106.9 −0.7 1.8

short 1 1866 1870 1867 1 3 4 −7.8 7.5 −7.8 2.4
short 2 1870 1875 1872 2 3 5 −3.2 6.7 −1.6 2.2
short 3 1875 1883 1876 1 7 8 −1.9 15.6 −1.9 2.1
short 4 1883 1888 1884 1 4 5 −0.8 9.1 −0.8 2.2
short 5 1888 1913 1889 1 24 25 −2.5 62.3 −2.5 2.0
Medium cycle 2 1913 1917 1915 2 2 4 −8.9 9.5 −4.5 4.6
short 6 1913 1917 1915 2 2 4 −8.9 9.5 −4.5 4.6
Medium cycle 3 1917 1929 1921 4 8 12 −9.0 43.7 −2.3 4.6
short 7 1917 1920 1919 2 1 3 −8.7 2.7 −4.4 2.7
short 8 1920 1926 1921 1 5 6 −2.9 31.2 −2.9 5.6
short 9 1926 1929 1927 1 2 3 −1.9 11.6 −1.9 5.7
Medium cycle 4 1929 1939 1931 2 8 10 −5.7 23.0 −2.9 2.6
short 10 1929 1932 1931 2 1 3 −5.7 2.1 −2.9 2.1
short 11 1932 1935 1934 2 1 3 −1.4 5.4 −0.7 5.4
short 12 1935 1939 1936 1 3 4 −3.5 20.1 −3.5 6.3
Medium cycle 5 1939 2007 1945 6 62 68 −43.9 1522.3 −9.2 4.6
short 13 1939 1974 1945 6 29 35 −43.9 721.9 −9.2 7.5
short 14 1974 1992 1975 1 17 18 −2.1 61.3 −2.1 2.9
short 15 1992 2002 1993 1 9 10 −0.9 18.2 −0.9 1.9
short 16 2002 2007 2003 1 4 5 0.0 6.7 0.0 1.6
Medium cycle 6 2007 2011 2009 2 2 4 −6.6 2.2 −3.3 1.1
short 17 2007 2011 2009 2 2 4 −6.6 2.2 −3.3 1.1
Notes: (1) The duration of the full cycle is measured from peak to peak. (2) Percentage change from trough to peak (downturns) of peak to trough
(upturns). (3) The slope is the average (geometric) growth rate in the phase.
Furthermore, the average intensity of downturn, less than 1%, is the lowest value recorded over all the medium cycles examined.
Particularly interesting is the short cycle 5 which is characterized by a very long duration (25 years of expansion), due to several
factors, including the growth of industrial sector, that showed a regular pattern from the last decade of XIX century, with a strong
acceleration during the Giolitti period (1898–1913), fostered by the international economic recovery. So, our cycle captures the new
thesis of a wave growth of Italian industry (Fenoaltea, 2006) rather than a take-off in 1897 singled out in a different cyclical
framework by Delli Gatti et al. (2005).
The second historical period, which includes medium cycles 2, 3, and 4, is particularly complex for a number of reasons such as
the economic effects of World War I, the rise of fascism, the spread of the Great Depression, and the massive intervention of the
government in the economy. During this period there is no continuity of growth because many shocks hit the economy. The duration
of the cycles is reduced and their slopes (downturns and upturns alike) become more violent. Italy experienced a severe economic
recession during WWI and the years 1917–1920 (short cycle 7), with a recovery during the first fascist period (short cycle 8). In the
late 1920s and early 1930s, the Great Depression also affected Italy, with a large fall in industrial production, investments, a per-
sistent decline in international trade and wage rigidities (Perri and Quadrini, 2002). There are different interpretations of the role
played by real and financial factors in the widespread of Great Depression in Italy (Ferri, Garofalo, 1994). Our analysis of business
and credit cycles confirms that the causes of crisis were above all of real nature, since we don't find a trough in the medium term
credit cycle. Already Ciocca and Toniolo (1983) argued that the causes of the Great Depression were mainly of a real nature and that
for Italy would not fit the monetarist interpretation of Friedman and Schwartz (1963).
Medium cycle 5 (1939–2007) starts with the deepest recession in our history, as a consequence of the massive destruction of
capital and labor during WWII (the GDP fell by 44% in 1939–1945, short cycle 13). However, the most striking evidence about
5
medium cycle 5 is the exceptional expansion (1945–2007) that began at the end of World War II and continued for 62 years, during
which time real GDP grew by 15 times, at an average annual rate of 4.6% (Table 1).
Of course, several interruptions to GDP growth occurred during this period, although consistent with our definition of a medium
cycle, none of them signified a medium-cycle turning point. None of the recorded recessions, the oil recession of 1974–1975, the
1992–93 financial crisis, and the 2002–2003 stagnation, lasted for more than 1 year. Relevant political events and real and financial
aspects interacted to shape this very long cyclical phase. Clementi et al. (2014) find that two structural breaks occur, in 1974 and
2001, in GDP growth, which coincide with turning points of our dating (short cycles 14 and 16). The first break put an end to the
strong recovery that began with the postwar capital reconstruction (short cycle 13), supported by substantial international aid, which
had led to the so-called “economic miracle.” Very interesting is the long duration (29 years) of the expansion phase of short cycle 13
(1945–1974), corresponding to the Bretton Woods era and the international economic expansion, that supported the greatest growth
of industrial production in Italy.
After the oil recession, the Italian economy experienced two long-lasting expansionary phases (of 17 and 9 years, respectively,
short cycles 14 and 15), although at a slower pace of growth. During these years, a steep increase in the debt-to-GDP ratio com-
menced, leading to debt-fueled growth, which was followed, after the currency crisis of 1992, by a restrictive fiscal stance, with a
large primary surplus but also stagnant economic performance.
Hence, what has really shaped this unusually long-lasting medium-term expansion is not an exceptional rate of growth (apart
from short cycle 13), but rather the unusually short-lived recessions, to which further research should be devoted. The joint inter-
action of fiscal and monetary policies and the active role of the central bank and the government in stabilizing the economy provide a
possible explanation, especially during the years preceding the austerity plans introduced by the European Monetary Union (early to
mid-1990s). The role of fiscal policy was completely different in the early 1990s, when Italian governments were strongly committed
to pursuing primary surpluses following the currency crisis of 1992, although once again, the devaluation of the lira helped the
recovery by raising foreign demand. In the early years of the 21st century, there was no support from either fiscal or monetary policy
to mitigate the impact of the recession experienced in 2001–2002; therefore, a further explanation of the brief duration of the
recessions recorded during the period 1945–2007 might be the role of wealth effects originating from real and/or financial assets,
since during the entire period of 1950–2007 private consumption contributed about 60% to GDP growth, more if we exclude the
period 1950–1974. This issue is certainly worthy of further investigation because, as we note in the next section, while this long
period of expansion occurred, some medium-term contractions adversely affected the financial side of the economy.
4.2. The credit cycle
We use two measures to define and depict the credit cycle, total bank loans and bank claims, as described in Section 3.
In Table 2, we summarize the cyclical properties of total bank loans, while Fig. 2 plots the level of the series along with turning-
point years. In the figure, the shaded parts correspond to the periods of recession in economic activity as calculated by GDP medium
cycles. Credit medium cycles range from 17 to 40 years, while short cycles last between 2 and 15 years, except for short cycle 12,
Table 2
Total bank loans: basic features of medium and short cycle.
short 1 1872 1879 1873 1 6 7 −13.2 79.6 −13.2 10.2

short 2 1879 1888 1880 1 8 9 −1.7 125.1 −1.7 10.7
Medium cycle 1 1888 1913 1891 3 22 25 −14.9 129.6 −5.2 3.8
short 3 1888 1892 1891 3 1 4 −14.9 8.4 −5.2 8.4
short 4 1892 1897 1895 3 2 5 −5.0 0.8 −1.7 0.4
short 5 1897 1906 1898 1 8 9 −3.0 66.9 −3.0 6.6
short 6 1906 1910 1907 1 3 4 −1.2 30.2 −1.2 9.2
short 7 1910 1913 1911 1 2 3 −0.5 6.8 −0.5 3.3
Medium cycle 2 1913 1933 1918 5 15 20 −41.1 319.1 −10.0 10.0
short 8 1913 1924 1918 5 6 11 −41.1 108.7 −10.0 13.0
short 9 1924 1933 1925 1 8 9 −3.3 107.8 −3.3 9.6
Medium cycle 3 1933 1973 1945 12 28 40.0 −81.3 3697.7 −13.0 13.9
short 10 1933 1939 1938 5 1 6 −20.8 6.7 −4.6 6.7
short 11 1939 1941 1940 1 1 2 −6.5 3.0 −6.5 3.0
short 12 1941 1973 1945 4 28 32 −77.0 3697.7 −30.8 13.9
Medium cycle 4 1973 1993 1982 9 11 20.0 −16.0 77.8 −1.9 5.4
short 13 1973 1976 1975 2 1 3 −3.9 1.6 −2.0 1.6
short 14 1976 1979 1978 2 1 3 −7.1 2.4 −3.6 2.4
short 15 1979 1993 1982 3 11 14 −9.6 77.8 −3.3 5.4
Medium cycle 5 1993 2010 1996 3 14 17.0 −5.5 99.4 −1.9 5.1
short 16 1993 2008 1996 3 12 15 −5.5 93.3 −1.9 5.6
short 17 2008 2010 2009 1 1 2 −3.5 6.9 −3.5 6.9
6
Fig. 2. Turning points of business and credit medium cycles (GDP and loans measured at 2005 prices, logs), 1861–2013.
Note: Turning points based on the methodology described in Section 3.1.
which includes WWII, after which an exceptionally long credit expansion commenced (1945–1973). Differently from business cycles,
financial cycles are quite homogeneous in terms of their length, with the exception of medium cycle 3, particularly long due to the
very long credit expansion occurred in the period 1945–1973 (short cycle 12).
Cycle characteristics have been quite homogeneous over the examined period, although the intensity of the downturns in medium
cycles seems to have declined since the 1970s. In medium cycles, the duration of the contraction phase (downturn) is always shorter
than that of the expansionary one, and similar evidence emerges in relation to amplitude and slope. However, when looking at the
short cycles, there are some exceptions to these regularities (namely short cycles 3, 4, 10, 11, 13 and 14).
The first medium bank lending cycle, 1888–1913, is quite procyclical, and therefore consistent with the medium GDP cycle of
1866–1913, with an increase in total loans of 130% and GDP growth of about 107%. To better understand the dynamics of the credit
cycle during this period, it is important to consider the effect of the gold standard on interest rates and international investments
(Fenoaltea, 2006). In Italy, the gold standard was resumed in 1883, and together with a better organization of the financial system
thanks to the new banking law of 1874, it had a positive impact on the inflow of foreign capital, which financed an important surge in
economic growth, especially in the construction sector (James and O'Rourke, 2013). In addition, during this medium cycle, a major
reform of the banking system, in 1893, followed the Banca Romana scandal, and allowed to cope with the bail-out of two largest
commercial banks, Credito Mobiliare and Banca Generale, failed in 1893 and 1894, respectively. The reform reduced the number of
issuing banks from six to three and led to the foundation of the Bank of Italy, which had a monopoly over note issuing, together with
Banco di Napoli and Banco di Sicilia. Later, with the 1926 banking law, the Bank of Italy gained the monopoly of issue
(Gigliobianco, 2006).
Medium cycle 2 lasts from 1913 to 1933, and includes only two short cycles. This period shows an increased amplitude in both
contractions and expansions of total loans. In the 15 years of upturn from the end of World War I to 1933, the average annual growth
rate of credit to the economy was 10%. The requirements of war and reconstruction between the two wars prompted a substantial
injection of credit into the economy, in particular to firms which, in various phases of the First World War and the fascist period, were
able to increase their profit margins.
During medium cycles 1 and 2 (Table 2), the prevailing bank business model was a relationship model, the so called “banca mista”
(universal banks). It is likely that this model contributed to shaping the credit cycle. In particular, relationship lending alleviated
credit constraints during the period of the Great Depression, as can clearly be seen from medium cycle 2 and short cycle 9. These
results are in line with the main finding in the international literature on banking relationships over the credit cycle, namely that
firm-bank relationships provide continued access to finance during downturn phases (D'Auria et al., 1999).
The beginning of medium cycle 3 corresponds with an important change in banks’ lending practices, that is, the end of “banca
mista” and the 1931–1936 banking reform that established a distinction between short-term and long-term credit institutions as a
consequence of the financial and economic crisis of the early 1930s (Bartoletto and Garofalo, 2014; Barbiellini Amidei et al., 2012). In
1931, the Istituto Mobiliare Italiano (IMI) was created. This was a state-owned credit institution specializing in the provision of long-
term credit to industry. After the establishment of the Institute for Industrial Reconstruction (IRI) in 1933, the state took control of
the major banks and a significant proportion of the industrial system.
7
Interestingly, the medium cycle is characterized by three strong credit contractions in 1933–1938 (short cycle 10), 1939–1940
(short cycle 11), and 1941–1945 (short cycle 12), followed by a very long expansionary phase in 1945–1973. The first two credit
contractions were more intense than the corresponding expansion, while the third, which occurred during World War II, as is clearly
shown in Fig. 2, was the most intense contraction of the last 150 years. The decline averaged more than 30% per year over the 4-year
period.
After the WWII, there was a long phase of 28 years of upturns (short cycle 12). We should consider that the institutional reform of
the financial system might have contributed to this economic boom (see Table 1). Battilossi et al. (2013), among others, argue that
the Italian banks supported the real economy effectively in the period 1948–1970 when industrial production increased annually by
about 8% and Italy joined more advanced economies, recording the highest growth rates of loans to GDP in Italy's history (6.9% per
year on average). The medium cycle during the period 1973–1993 is characterized by rather divergent fiscal and monetary policies.
On one hand, nominal income grew, driven by public spending, consumption, and inflation, while on the other hand, monetary
policy, which was based on the intermediate target of total domestic credit, became restrictive. In the context of wage indexation to
prices and strong depreciation of the lira, governments have continued to create budget deficits pushing consumption and inflation
(Bartoletto et al., 2014). Comparing Tables 1 and 2, it appears that during this medium cycle, a negative comovement between credit
and economic activity prevailed. As already noted in Section 4.1, apart from the 1975 recession, GDP had been growing steadily since
1945. Conversely, short credit cycles 13, 14 and 15 showed sharp loan contractions during the period 1973–1978 (with the exception
of 1976) and 1979–1982.
There were two reasons for the contraction in the stock of real credit experienced during the period of 1973–1982: first, the
administrative controls on credit expansion, and second, the very strong rise in prices. The credit ceiling was in place from 1973 to
1975, and then, after a suspension during the period March 1975–October 1976, a new ceiling for private sector loans was in place
until 1983.
The last medium cycle (1993–2010) is characterized by an initial sharp contraction of credit during the period 1993–1996, a
particularly fragile period for the Italian economy, which faced the imminent establishment of the Eurozone, along with a crisis
linked to fiscal imbalance and the exchange-inflation relationship. Moreover, during this period, the Italian banking system was
affected by massive losses from loans stemming from the deterioration of corporate finances in 1992 and 1993 and the long stag-
nation of the southern economy following the termination of the program of extraordinary measures for this area.
The beginning of credit cycle 5 corresponds with another important reform of the banks’ business model, namely the 1993
banking law (Testo Unico Bancario), which prescribed the elimination of any distinction between short-term and long-term credit
institutions and a return to a universal banking model. Thanks to the significant reforms introduced during the 1990s, the banking
system succeeded in overcoming these difficulties, and in the following years of upturn, credit growth was excellent, with an average
of 5% per year (Messori, 2002).
Table 3 shows the cyclical properties of the banks’ claims series. The series differs from that commented on Table 2 because here,
credit includes government bonds held by the banking system. The two series are almost perfectly synchronized, especially with
Table 3
Bank claims: basic features of medium and short cycle.
Medium cycle 1 1888 1913 1891 3 22 25 −11.1 124.8 −3.8 3.8

short 1 1872 1888 1873 1 15 16 −9.9 314.4 −9.9 9.9
short 2 1888 1894 1891 3 3 6 −11.1 13.0 −3.8 4.2
short 3 1894 1906 1895 1 11 12 −1.6 60.8 −1.6 4.4
short 4 1906 1910 1907 1 3 4 −1.9 24.5 −1.9 7.6
short 5 1910 1913 1911 1 2 3 −1.9 4.9 −1.9 2.4
Medium cycle 2 1913 1933 1918 5 15 20 −38.3 304.4 −9.2 9.8
short 6 1913 1919 1918 5 1 6 −38.3 30.1 −9.2 30.1
short 7 1919 1924 1920 1 4 5 −0.6 44.7 −0.6 9.7
short 8 1924 1933 1925 1 8 9 −5.6 129.1 −5.6 10.9
Medium cycle 3 1933 1941 1938 5 3 8 −22.9 10.9 −5.1 3.5
short 9 1933 1939 1938 5 1 6 −22.9 6.0 −5.1 6.0
short 10 1939 1941 1940 1 1 2 −3.6 8.6 −3.6 8.6
Medium cycle 4 1941 1979 1945 4 34 38 −73.2 2887.3 −28.0 10.5
short 11 1941 1973 1945 4 28 32 −73.2 2587.4 −28.0 12.5
short 12 1973 1975 1974 1 1 2 −3.6 3.2 −3.6 3.2
short 13 1975 1979 1976 1 3 4 −0.8 12.7 −0.8 4.1
Medium cycle 6 1979 1994 1981 2 13 15 −7.3 69.6 −3.7 4.1
short 14 1979 1994 1981 2 13 15 −7.3 69.6 −3.7 4.1
Medium cycle 7 1994 2012 1996 2 16 18 −5.0 75.3 −2.5 3.6
short 15 1994 2008 1996 2 12 14 −5.0 61.4 −2.5 4.1
short 16 2008 2010 2009 1 1 2 −1.2 8.6 −1.2 8.6
short 17 2010 2012 2011 1 1 2 −0.2 1.4 −0.2 1.4
8
regard to the medium cycles, but with some major differences. First, the credit downturn in the 1970s is much shorter, since the
crunch only commences in 1980 (bank loans to the private sector declined in 1974) during the second oil shock. Moreover, contrary
to Table 2, here, upturn phases are always longer than contraction phases, with the only exceptions being the short-term contractions
during WWI (short cycle 6) and the Great Depression (short cycle 9).
5. Comparing business and credit cycles
In the aftermath of the global financial crisis, the financial cycle has taken centre stage to detect financial imbalances and pursue
macroeconomic stability (Assenmacher-Wesche and Gerlach, 2010; Liang, 2017). Several studies have documented that recessions
associated with financial downturns (and/or banking crises) are more severe than simple GDP recessions (Claessens et al., 2012;
Bordo and Haubrich, 2010; Schularick and Taylor, 2012; Jordà et al., 2013) and that credit cycles based on credit or the credit-to-
GDP ratio are among the main macroprudential tools for achieving macro-financial stability (Babecky et al., 2014). However, there is
mixed evidence regarding the synchronization of business and financial cycles. Some studies find that credit is lagging (Haavio, 2012;
Haavio et al., 2013; Runstler and Vlekke, 2017), while other studies find that credit is leading (Gomez-Gonzales et al., 2014);
moreover, in some studies the results vary according to the specific country under examination (Tamirisa et al., 2009). Stylized facts
reported for EU in ECB (2013) and Giannone et al. (2012) show that real private sector loans have been lagging over the business
cycle although there are differences across the main components. In this section, we show how business and credit cycles interact in
Italy.
In so doing we keep in our dataset the years involving the two World Wars, that, as we have just seen in Tables 1–3, exert an
important effect in shaping business and credit cycles, both the short and the medium ones. As a matter of fact, Table 2 shows a
perfect synchronization between the troughs of the medium cycle of credit and the end of the belligerent events, respectively 1918 for
medium cycle 2 and 1945 for medium cycle 5. As to the business cycle, the WWI was coincident with an economic expansion
(medium cycle 2) whereas the WWII caused a severe contraction (medium cycle 5). Our approach is consistent with Meller and Metiu
(2017) who find that downturns in the medium-term component of real bank loans are associated to “rare economic disasters, such as
wars and systemic banking crises”. In other terms, the exclusion from the analysis of the years related to the World Wars events,
would bias the dating of the medium cycles.
5.1. Measures of synchronization
Having defined the cycle in terms of the turning points of a series, this section uses parametric and nonparametric methods to
assess the degree of synchronization between credit aggregates and GDP cycles (short and medium). First, we use the concordance
index developed by Harding and Pagan (2002, 2006). This is a measure of how often two cyclical variables, in our case GDP (Yt) and
credit (Ct) are in the same phase of the cycle. Denote a binary variable SY,t such that SY,t = 1 when the series Yt is in expansion and
SY,t = 0 during the time spent in contraction. A similar definition applies to the binary variable SC,t denoting the credit cycle. Then,
the concordance CIYC between the credit and the business cycles is defined as
1
CIYC = [ I (SY , t = 1; SC, t = 1) + I (SY , t = 0; SC , t = 0)]
n (1)
where I(.) is an indicator function taking a value of 1 when the expression in parentheses is true. Two series are perfectly procyclical
(countercyclical) if the index is equal to unity (zero). This concordance index is the sample analog of Prob (SY,t = SC,t). An advantage
of the index is that it is a well-defined quantity even if the variables Yt and Ct are integrated series, while standard comovement
measures are defined between covariance stationary variables.
Although CIYC = 1 implies perfect synchronization of phases and CIYC = 0 implies perfect countercyclicality, CIYC does not take a
unique value in the case of perfect nonsynchronization or independence because its value depends on the asymmetry of the phases. In
fact, CIYC = 0.5 occurs under the independence hypothesis only if expansions and recessions last for the same period of time.
However, Harding and Pagan suggest that if the two series are independent, then the expected concordance is:
E (CIYC ) = E (SY , t ) E (SC, t ) + (1 E (SY , t ))(1 E (SC , t )) (2)
Hence, an actual concordance that is higher than the expected one indicates procyclicality, while a lower concordance indicates
countercyclicality.
The concordance index is linked to other measures of synchronization such as the common (Pearson's) coefficient correlation.
Harding and Pagan (2006) showed that, for two variable X and Y, the following relationship holds:
CIxy = 1 + 2 (µ (Sx )(1 µ (Sx ))1/2 (µ (Sy )(1 µ (Sy ))1/2 + 2µ (Sx ) µ (Sy ) µ (Sx ) µ (S y ) (3)
where m stands for the sample average, r is the estimated correlation coefficient between Sx and Sy and we neglect the index t for
simplicity. This regression-based approach for r provides us with a standard error for the significance of correlation between credit
and business cycles.
Finally, being SC and SY binary variables, we also report the corrected contingency coefficient suggested by Artis et al. (1997). For
two binary cyclical variables of N observations each, the corrected contingency coefficient is:
9
Table 4
Comovements of business and credit cycles, 1861–2013.
Periods a) 1861–2013 b) 1861–1939 c) 1939–2013
Measures of synchronization with GDP cycle Total loans Bank claims Total loans Bank claims Total loans Bank claims
Medium cycle Medium cycle Medium cycle
CI 0.75 0.80 0.71 0.73 0.80 0.88

E(CI) 0.68 0.74 0.70 0.71 0.66 0.78
CC 32.1*** 31.9*** 5.6 8.1 58.3*** 58.8***
Correlation 0,20 0.23* 0,03 0.05 0.34* 0.47**
Correlation at t-1 (1) 0.21* 0.14 0,05 0.05 0.31* 0.28
Correlation at t+1 (2) 0,09 0.19 −0.05 −0.04 0.24 0.49**
Short cycle Short cycle Short cycle
CI 0.73 0.76 0.65 0.69 0.81 0.82

E(CI) 0.64 0.67 0.62 0.64 0.66 0.70
CC 35.2*** 35.6*** 11.7 18.8 59.1*** 53.8***
Correlation 0.23** 0.25*** 0.07 0.12 0.40** 0.40**
Correlation at t-1 (1) 0.10 0.18* −0.05 0.06 0.30* 0.36**
Correlation at t+1 (2) 0.20** 0.34*** 0.14 0.27** 0.30* 0.45***
Notes: Business cycle is represented by real GDP cycle. CI = concordance index; E(CI) = expected concordance index; CC = corrected contingency
coefficient. *,**,***: significant at 10%, 5% and 1%, respectively. (1) The variable is leading if this correlation is the highest; (2) The variable is
lagging if this correlation is the highest.
N+ 2
C= *100
0.5
(4)
which lies between 0 and 100 and is equal to 0 (χ2 = 0) in the case of independence and 100 in the case of perfect association. In our
context, independence indicates that there is no contemporaneous relationship between the two cycles.
Table 4 reports the unconditional concordance index (CI), the expected concordance index (E(CI)), the corrected contingent
coefficient (CC), and the Pearson's correlation between credit and GDP cycles.12 Table 4a refers to the entire sample, but synchro-
nization has been also assessed for two sub-periods, 1861–1939 (Table 4b) and 1939–2013 (Table 4c); the threshold year, 1939, is
derived from GDP turning points.
Table 4 lists several elements that deserve attention. First, loans and bank claims are pro-cyclical for both short-term and medium-
term cycles: the concordance index is always higher than the expected value.
Second, looking at the contingency index CC and Pearson's correlation coefficient, the table shows that bank loans and bank
claims are significantly synchronized with the business cycle although for medium cycles the correlation is only significant for bank
claims. However, business and credit cycles are poorly synchronized in the first part of the sample.
Third, Table 4 also shows correlations between GDP and the other variables 1 year earlier (t – 1) and 1 year ahead (t + 1). In the
medium cycles, the cross-correlations are generally lower than the contemporaneous ones, and no lead/lag relationship arises be-
tween GDP and credit. However, in the short cycles, bank claims show a significant and higher correlation with past GDP, which
means that GDP leads this broader credit aggregate.
To sum up, two main conclusions can be drawn from Table 4. First, synchronization between real and credit cycles has changed
over time. Although the degree of concordance is similar between the two periods, the CC and correlation are much stronger and
more significant in recent decades than in the past. The finding that business and credit cycles are poorly synchronized in the first
part of the sample is a new result in analyses that focus on Italian data, whereas the evidence of strong correlation during the second
part of the sample is consistent with available studies that use quarterly data from 1970 onwards (Tamrisa et al., 2009; Classens et al.,
2012; Runstler and Vlekke, 2017).13
A second conclusion concerns the timing of synchronization, which changes with the definition of a cycle. At medium frequencies,
the credit cycle indicator (loans and bank claims alike), is a coincident indicator of the business cycle.14 Conversely, at higher
frequencies (short cycles), the credit cycle of bank claims is led by the business cycle.
12
In Table 4, we refer to the standard likelihood-ratio chi-squared test for independence although we are aware that cyclical variables are not
independent and identically distributed random variables but are serially correlated. By doing so, we are underestimating the probability of in-
dependence. Further, a response of independence is a lower-bound probability for the null.
13
In Appendix A.2 we show that qualitatively similar results are obtained when using the Christiano and Fitzgerald (2003) filter to date business
and credit medium cycles.
14
See Appendix A.3 for a summary of the major events that impinge on the turning points of the medium cycles.
10
5.2. Business and credit cycles: a Granger causality analysis
In the previous section, we found some evidence that the business cycle is leading for the bank claims cycle; in other words, SY,t-1
and SC,t are significantly correlated (Table 4, last row). In this section, we run a Granger causality test, which, in our case, tests
whether lagged cycle states of GDP, as denoted by the binary variable SY,t, help predict the probability of future credit cycle, as
denoted by the binary variable SC,t conditional on past SC,t:
SC, t = F[ + SC, t 1 + SY , t 1] + t, (5)
15
where F[] is a logit link function, and Eq. (5) is estimated by the Maximum Likelihood estimator as usual. In this setting, if lagged
business cycles have some predictive power for credit cycles, then the business cycle is a leading indicator of the credit cycle (see,
e.g., Breitung and Candelon, 2006, who recently used this interpretation in the frequency domain setting). In exploring the dynamic
relationship between credit and business cycles, we consider both short and medium cycles. Moreover, we test whether credit cycles
Granger cause the business cycle:
SY , t =F[ + SY , t 1+ SC, t 1] + t (6)
To the best of our knowledge, despite the importance this issue may assume in light of the present crisis, these questions have
rarely been asked in classical analyses of the business cycle. The pairwise Granger causality tests are shown in Table 5.
Along with the estimated coefficients, we also report the marginal effects (MEs) when they are significant. The pairwise Granger
causality tests provide support for the results reported in Table 4. In the medium cycles, there is no causality in both directions
between the credit and business cycles. However, in the short cycles, the probability of upturns in credit is significantly increased by a
turn in the business cycle although the reverse is not true. In particular, when GDP turns from recession to expansion, the probability
of an expansionary credit phase increases by 26% with regard to bank loans and by 37% in relation to the bank claims.
This evidence agrees with the earlier correlation analysis and other studies, according to which, especially in the post-WWII era,
GDP leads bank credit (EBF, 2011; ECB, 2012), at least at the usual business cycle frequencies (our short cycle).
6. The real impact of financial cycles
6.1. Descriptive evidence from univariate analysis
In this section, we investigate what happens when recessionary phases of credit and GDP are synchronized. Similar to
Claessens et al. (2012) and Bordo and Haubrich (2010), we seek to determine whether the coincidence of recessions makes recessions
more intense.
As in the aforementioned studies, we do not mean to model Italian GDP but simply measure its cyclical properties conditional on
the cyclical phase of financial variables. We first examine how the likelihoods of downturns change conditional on having a negative
financial phase.
Table 6 shows the following: i) the unconditional probability of a GDP downturn measured as the fraction of years of recession
observed in the sample; ii) the conditional probability of a GDP downturn, which refers to the fraction of years in which recessions are
associated with credit contractions. In Table 6, the unconditional probability of being in a downturn phase is 16% in the medium
cycle and 20% in the short cycle. These percentages almost double (to 31 and 37%, respectively) under loan recession, and the effect
is even stronger when bank claims are declining.
Another way to recognize the impact of a financial recession on a real, ongoing recession is to measure the intensity of conditional
GDP recessions, as in Bordo and Haubrich (2010). However, because their sample included few observations (15), equal to the phases
detected, we slightly modify the regression by using GDP growth rates, and we compare the average GDP growth rates across
different business and financial cycles.16 A simple way to run this test is by regressing GDP growth rates on the cyclical variables in
order to single out the effects of i) a GDP recession; ii) a loans contraction and iii) the joint presence of a GDP recession and a financial
contraction.17 If we denote the yearly GDP growth rate as gt, then the estimated equations are as follows:
gt = a + b (1 SY , t ) + t (7)
gt = a + b (1 SY , t ) + c (1 SY , t )(1 SC , t ) + d (1 SC , t ) + t (8)
where SY,t and SC,t were defined in Section 5 such that SY,t = 1 when the series Yt is in expansion and SY,t = 0 during the time spent in
contraction, and the same is true for the credit series.
We repeat the regression for both the short cycle (SC) and the medium cycle (MC). To control for autocorrelation of residuals, we
provide robust standard errors using the Newey-West estimator.
15
We use one lag, which is the standard specification with annual data. However, more lags are collinear because of the use of binary variables.
Standard errors are robust to account for autocorrelations of residuals.
16
It is worth noticing that this amounts to running this test on short cycles of GDP, given the 1:1 correspondence between annual growth rate and
cyclical phase in our definition of a short cycle (see Section 3.1).
17
Instead of comparing average growth rates between phases using a standard mean difference t-test, the regression approach allows us to account
for autoregressive structures in the data (e.g. samples are not independent) and to adjust standard error accordingly.
11
Table 5
Pairwise causality between credit and business cycles (estimated coefficients).
Cycle Bank loans Bank claims
From credit to GDP (b)

Medium cycle 0.996 0.034
Short cycle 0.306 0.743
From GDP to credit (g)
Medium cycle −0.868 0.637
Short cycle 0.829* 1.596***
ME 0.26*** 0.37***
Notes: Logit estimates; robust Newey s.e. for autocorrelation of residuals. The marginal
effect (ME) is the change in probability of expansion when the predictor X goes from
recession (X = 0) to expansion (X = 1). *,**,***: significant at 10%, 5% and 1%, re-
spectively.
Table 6
Probability of GDP recession conditional on the financial cycle.
State of financial cycle Medium cycle Short cycle
a) Probability of GDP recession

Unconditional probability 0.16 0.20
Conditional on:
Bank Loans upturns 0.11 0.14
Bank claims upturns 0.12 0.14
Bank Loans downturns 0.31 0.37
Bank claims downturns 0.36 0.39
Note: The unconditional probability of a downturn is the fraction of time in which a downturn occurs
during the sample. The conditional probability is the fraction of time in which there is a downturn in
GDP given a specific financial cyclical phase.
In Table 7, columns 1 and 2 report the estimated coefficients for Eq. (7), whereas columns 3–6 show the estimated coefficients for
Eq. (8). Specifically, we estimate Eq. (8) either for the bank loans (SC,t = 0 when loans are contracting, coefficients shown in columns
3 and 4) and for the bank claims (SC,t = 0 when bank claims are in recession, columns 5 and 6).
As for Eq. (7), the constant coefficient in columns (1) and (2) provides us with growth rates in expansions (3.5% in the MC and
3.9% in the SC). By adding the estimated b coefficient (second row), we can estimate the average growth rate in recessions:
−7.4 + 3.5 = −3.9 in MC and −7.8 + 3.9 = −3.9 in SC. To determine the effect of financial contractions (Eq. (8)), we look at
columns (3) and (4) and columns (5) and (6). The constant term still represents the growth rate in expansion, while the estimated b′
coefficient is the decline in the GDP growth rate in recession, conditional on the credit variable being in expansion (that is, SC,t = 1;
hence c′ = 0 and d′ = 0). As expected, the estimated module for the b′ coefficient decreases compared with columns (1) and (2): if
credit is not declining during a business cycle downturn, then the recession is less intense.
Table 7
Impact of financial cycle contraction on GDP growth rate in recessions.
Dep. variable: GDP growth rate SC (1) MC (2) SC (3) MC (4) SC (5) MC (6)
Const 3.9*** 3.5*** 4.2*** 3.7*** 3.9*** 3.7***

(a, a′ in Eqs. (7) and (8)) (0.5) (0.5) (0.6) (0.7) (0.6) (0.5)
GDP recession −7.8*** −7.4*** −6.5*** −5.4*** −6.3*** −5.4***
(b, b′ in Eqs. (7) and (8)) (1.3) (1.6) (6.8) (1.02) (0.75) (0.9)
GDP & Total loans recession −1.8 −3.6**
(c′ in Eq. (8)) (1.9) (2.1)
GDP & Bank claims recession −3.1 −4.9**
(c′ in Eq. (8)) (2.1) (2.7)
Total loans recession (d′ in −1.4** −1.1
Eq. (8)) (0.7) (0.8)
Bank claims recession −0.6 −1.3
(d′ in Eq. (8)) (0.8) (0.9)
obs. 152 151 152 151 152 151
Average GDP growth in recession (a + b) −3.9 −3.5
Average GDP growth in recession and credit expansion (a′ + b′) −2.3 −1.7 −2.4 −1.7
Average GDP growth in real and financial recession (a′ + b′ + c′) −4.1 −5.3 −5.5 −6.6
Note: SC = short cycles; MC = medium cycles. Standard error in parentheses, robust for autocorrelation of residuals (Newey-West correction).
*,**,***: significant at 10%, 5% and 1%, respectively.
12
Conversely, and more importantly, when credit and real downturns coincide, GDP recessions are more severe. In column (4),
when the negative effect of loans contraction is significant, GDP drops by 3.7−5.4−3.6 = −5.3% (a′ + b′ + c′), and the impact is
even stronger when bank claims drop (−6.6%); this negative change is much higher than the unconditional growth rate in recession.
Finally, d′ provides us the effect of credit contraction per se, which, although negative does not affect significantly the GDP growth
rate in 3 out of 4 cases, with the notable exception of contraction of bank loans during short cycles: in this case a reduction in loans
would worsen economic performance by - −1.4% (the d′ coefficient in Eq. (8)).
The estimates in Table 7 suggest three conclusions. First, the financial cycle can worsen economic downturns, but only when we
look at medium-term fluctuations, which is when downturns are associated with more severe credit crunches and financial dis-
ruption. In this regard, as we have already pointed out, the war episodes, which have been included purposely in the analysis after a
reasoned choice, contribute to shape the dating of medium cycles, certainly affecting the results of Table 7. This leads to our second
remark, namely war events as well as severe financial crises, such as the one of 2008–09, are alike in their amplifying effect of
economic downturns. Albeit this does not allow us to conclude that financial shocks are the origin of the deeper recession (our
Granger causality analysis is not supportive of this), we do find evidence that the overlap of medium term cycles has a huge disruptive
potential. This is not a novel interpretation; Minsky (1964) reports in the very first lines of his article the testimony of Professor
Abramowitz classifying wars among the episodic events that can trigger long-swing expansions and contractions. In addition, the
financial instability hypothesis, which is a linchpin in the Minsky's analysis of long waves in financial relations, can be extended to
examine, in addition to speculative and Ponzi financial postures, also the consequences of war events. As a matter of fact, the
transition from a stable to an unstable financial system is depending upon the intensity of the shock occurring to the real economy
(the shortfalls of income-financial default plane reported in Minsky, 1964), which is obviously very ample in presence of the war
events.
Finally, and this is our third conclusion, the broader the financial aggregate, the larger the effect on the business cycle. This
suggests that when banks also restrict public bond holdings, the effect of a credit crunch on the real economy is increasingly intense.
6.2. Evidence from multivariate analysis: asymmetric responses to credit shocks
The previous section showed that business cycle recessions are more likely to occur during financial downturns (Table 6) and that,
when financial disruptions and economic downturns coincide, business cycle recessions are more severe than simple GDP recessions
(Table 7). In this section, we investigate the asymmetric response of the business cycle to credit shocks with respect to the credit cycle
phase in a formalized econometric model. Our analysis employs a nonlinear threshold vector autoregressive model (TVAR) of two
variables (GDP and Credit), which separates observations into different regimes based on a threshold variable. In our case, regimes
are credit expansions and recessions in the short as well as in the medium cycle.
Our starting point is a simple linear VAR model of GDP and Credit (credit stands, alternatively, for loans or bank claims) in growth
rates:
K K
gdpt = a1i gdpt i + b1i creditt i + ut
i=1 i=1 (9.1)
K K
creditt = a2i creditt i + b2i gdpt i + vt
i=1 i=1 (9.2)
where K = 3 lags are enough to minimize the information criteria and rule out autocorrelation in the residuals. According to this
linear specification, we get significant evidence of the effects of credit shocks on the GDP growth rate and vice versa.
Fig. 3 shows the impulse response functions using a standard Cholesky procedure. The ordering used in Fig. 3, credit after GDP,
according to Mumtaz et al. (2018), is adopted in most of the empirical works examining the real effects of credit shocks.18 Granger
causality analysis in our bivariate VAR confirms that the causal relation between Credit and GDP is bidirectional.19
However, in a linear VAR, the interlink between business and credit cycle overlooks the cyclical phase in which a shock occurs. In
the next specification, we evaluate the effect of a credit shock on GDP (and vice versa), conditional on the phase of the credit cycle the
economy is going through. To do so, we estimate a nonlinear threshold VAR model that is defined as follows:
K K
yt = c1 + 1j yt j + vt SC , t 1 + c2 + 2j yt j + ut (1 SC , t 1)
j =1 j=1 (10)
where Δyt is a vector including the first differences of GDP and credit (credit stands, alternatively, for loans or bank claims), and SC,t
was defined in Section 5 such that SC,t = 0 when the credit is in recession (Regime 1) and SC,t = 1 during credit expansions (Regime
2). Our threshold variable SC,t is defined both in the short and in the medium cycles. This amounts to estimating two standard
18
We have also checked for the sensitivity of the results displayed in Figs. 3–5 to the alternative chain ordering (Juselius, 2006). Results are robust
and available upon request.
19
From here on, in order to simplify the presentation of the results, although the variable credit stands for both loans and bank claims, we only
show and comment on the results that refer to the loans. The results that refer to the cyclical component of bank claims are qualitatively similar and
available upon request.
13
Fig. 3. Impulse response of the linear VAR.

Note: Orthogonalized Impulse Response Function to a standard deviation shock, based on VAR model (9.1 and 9.2). The Credit variable is Bank
Loans. Cholesky identification; ordering of the variables is GDP Credit.
separate linear VARs for GDP and credit for each regime identified by credit cycle phases, i.e. the threshold is exogenous. The
advantage of this approach is that, once a regime is fixed, the model is linear, namely we are implicitly assuming that the system
remains “for a while” in a given regime after the shock. In other words, impulse responses are not functions of history.
First of all, we test for the linear specification against the nonlinear one.20 In a standard threshold autoregressive VAR, the
distribution under the null are non-standard, because they depend on the nuisance parameters (the value of the threshold and the lag
of the threshold variable), which are not present under the null (Lo and Zivot, 2001). However, when the threshold variable and the
lag are exogenously given, the linearity test is a standard Likelihood ratio test, where the Likelihood under the restricted model (the
SSR (NL) T
linear one, L) is compared to the non-linear specification NL (Tong, 1990). The LR test is –2ln[ SSR (L) ]( 2 ) is distributed as χq+1 where
q is the number of parameters in the linear specification and T the number of observations. The linear specification is strongly
rejected in either credit cycle regimes: the LR statistic is 17.00 (p-value 0.00) in the short cycle,21 55.84 (p-value 0.00) in the medium
cycle.
This simple nonlinear model allows to go one step further compared to the univariate analysis of the previous section, providing a
reliable evidence about the asymmetric effect of credit shocks on the growth rate of GDP and vice versa. The plots in Fig. 4a display
the responses of the GDP rate of growth to one standard deviation shock to the rate of growth of loans during short-cycle credit
contractions (Regime 1, left-hand side panel) and expansions (Regime 2, right-hand side panel). Similarly, the plots in Fig. 4b show
the response of the loans rate of growth to the rate of growth of GDP.
After a shock in growth rate of credit, the response of GDP is substantially different in the two regimes. During credit cycle
downturns, there is a significant negative response of GDP after a credit negative shock, occurring when a credit recession is in
progress,22 peaking one year later at about −4%. Differently, GDP response after a positive credit shock (when credit cycle is in
20
Here we comment only the main results of interest for the article. The full set of estimates and tests are available upon request. The results
commented in the next pages are robust to the inclusion, in Equation (10), of a dummy variable controlling for the years of the WWII, namely 1940-
45.
21
Even assuming that threshold is unknown in the credit cycle, the linearity assumption is strongly rejected. To run the test, we used the
procedure TVAR-LRtest in the R package Package ‘tsDyn’.
22
In the standard linear impulse response (for given regime the TVAR is linear), the shock is assumed to be positive. However, our definition of
14
Fig. 4. Impulse response in the asymmetric VAR, regimes based on short cycles.
Note: Orthogonalized Impulse Response Function to a standard deviation shock, based on threshold VAR model (10). The Credit variable is Bank
Loans, and the asymmetric effect is investigated under short credit cycles. Cholesky identification; ordering of the variables is GDP Credit.
expansion) is considerably smoother, reaching a high of about 1%. This asymmetric response to credit shocks is only to a limited
extent explained by the smaller standard deviation of the shocks in the expansionary regime. Indeed, the estimated standard de-
viations of credit shock in the diagonalized error matrix is 0.11 in Regime 1 and 0.07 in Regime 2, whereas the estimated response in
Regime 1 is about 4 times as large as those in Regime 2.
Hence, when comparing these effects with those of the linear model, it is clear that the positive correlation between credit and
GDP growth rates in the linear model are mostly driven by the positive correlation between negative growth rates during recessions
(Fig. 4a), as the expansionary effects of credit shock are very modest. Differently, looking at the linear specification only, one might
think that the positive effect of credit expansion on GDP growth are important. This asymmetry is confirmed by Granger causality
analysis, supporting the idea that there is Granger causality from credit to GDP only during credit downturns23. In addition, in order
to reinforce the evidence about the asymmetric effects of shocks, Fig. 4 finally shows that regime asymmetries are important in
(footnote continued)
short-credit cycles implies that in Regime 1 (credit recession) growth rate of credit is always negative. Hence, we plot the IRF for negative shock.
Given the linearity of the model, it simply amounts to multiply the original IRF by (−1).
23
Absence of Granger causality is strongly rejected in VAR under credit downturns (χ2(3) = 25.6, p-value 0.00), whereas is not rejected at 5%
under credit expansions (χ2(3) = 7.51, p-value 0.057).
15
shaping the response of credit to GDP too (panel 4b). Albeit the effect of GDP shocks on credit is apparently more intense during
credit recessions than during credit expansions, Granger causality analysis does not support the idea that past values of GDP con-
tribute to explain current values of credit during credit contractions. Differently, credit is Granger-caused by GDP during credit
expansion regimes.24
Moving to a threshold VAR specification based on medium credit cycle, the previous results are confirmed and reinforced (Fig. 5).
The differences between the recessions and the credit expansions are much stronger in medium cycles compared to short cycles, in
line with the results of Table 7: the scale of the response under Regime 1 is of one order of magnitude larger than the response under
the Regime 2, whereas the differences in the standard error of shocks are much smaller (0.11 and 0.06 for credit shocks in Regime 1
and 2 respectively; 0.05 and 0.03 for GDP shocks). As to Granger causality, it is reinforced the evidence that GDP is Granger caused by
credit only during credit contractions, whereas in both the regimes credit is Granger-caused by GDP.25
The remarkable differences between the regime-based impulse responses is not unusual in macroeconomic literature examining
asymmetric response of output to exogenous shocks (Auerbach and Gorodichenko, 2012), and is certainly affected by the assumption,
implicit in the exogenous modeling of the threshold, that the regimes themselves don't change. Actually, by using a threshold variable
exogenous to the system and examining separately the impulse response functions in the two regimes, we are forcing the system to
remain permanently in a given regime. This restriction is less counterintuitive in medium cycle analysis, provided that, as we have
shown in Tables 1 and 2, medium cycles display strong persistence. Nonetheless, as well pointed out by Auerbach and
Gorodichenko (2012), in such situations it is advisable to interpret reported magnitudes of the response for the two regimes as
“bounds from polar settings” acknowledging that “more realistic situations will fall between the extremes”. In addition, the authors
leave to understand that this approach may be more suitable than resorting to a generalized impulse response (Koop et al., 1996;
Pesaran and Shin, 1998). Indeed, calculating full dynamic impulse response functions that include internally consistent regime shifts,
would imply computing the exogenous threshold variable and evaluating its effects of the endogenous ones at each date along the
trajectory simulated by the generalized impulse response, a task which is beyond of the scope of the present analysis.
The results of this section provide further support to the preliminary evidence drawn from Section 6.1 and reinforce the hy-
pothesis that credit disruptions significantly contribute to business cycle recessions. In fact, while in Table 7 the credit contractions
that occurred during short credit cycles did not significantly impact the rate of growth of GDP, such contractions did play a role in
non-linear analysis. As shown, when considering asymmetries in credit cycles, there is evidence of a real, significant impact of the
credit cycle both at short and at medium-term frequencies. However, these crucial differences between financial expansions and
recessions and their asymmetric effects on the business cycle are completely overlooked in the more common linear analyses of
economic fluctuations.
7. Concluding remarks
In this paper, we investigated the link between business and credit cycles in Italy over the period of 1861–2013. We examined
both the standard business cycle frequencies and longer fluctuations, and we paid special attention to medium-term frequencies,
which identify the most significant events of financial turmoil and macroeconomic disruption.
In accordance with Aikman et al. (2015) and Drehmann et al. (2012), we show that long-lasting fluctuations are the main cyclical
component of the business and, notably, financial cycles. At the usual business cycle frequencies, all measures of comovement are
significant for both loans and bank claims aggregates, and we find some evidence that the business cycle leads the credit cycle. As for
the medium term, credit cycles appear to be weakly associated with business cycles. However, comovement increases and measures
become significant when the second part of the sample is considered. The fact that credit and business cycles are weakly synchronized
is only an apparent contradiction. In fact, two cycles may be weakly synchronized and still influence each other: when their negative
phases overlap, as in the recent crisis and in a few other cases in the past, the recession is deeper than a simple GDP recession.
An interesting result of the study concerns the relationship between the GDP growth rate during recessions and medium-term
financial cycles, as estimated in the univariate analysis. As we have shown, medium cycle dating is severely affected by exceptional
events, such as World Wars and financial crises, who contribute to shape long waves in financial as well as real aggregates. Under
such circumstances, the coincidence with a medium-term credit downturn significantly worsens the recession, as the GDP growth rate
declines from −1.7 to −5.3 and peaks at −6.6% when broad credit aggregate is considered. Differently, when looking at short-term
credit contractions, the overlapping of credit downturn and real recession does not produce any additional real effects, whereas credit
contraction mitigates the rate of growth of GDP during expansionary phases.
The important linkage between medium-term financial cycles and economic performance is also confirmed by a non-linear VAR
analysis, which indicates that there are important asymmetric effects on GDP depending upon the credit cycle phase considered, both
in short and in medium- term cycles. Precisely, we prove that the response of GDP to (a negative) credit shock is much more intense in
credit downturns. Although the WWII plays a crucial role in shaping medium-term downturns, the asymmetric response of GDP to
24
Absence of Granger causality is strongly rejected in VAR under credit expansions (χ2(3) = 18.95, p-value 0.00), whereas is not rejected at 10%
under credit recession (χ2(3) = 5.91, p-value 0.12).
25
The results of the Granger causality tests are, in the medium cycles regime, more clear-cut compared to the short cycle. The null hypothesis, i.e.
no Granger causality from credit to GDP, is rejected in the regime identified by credit cycle downturn (χ2(3) = 39.3, p-value 0.00), whereas in
Regime 2, credit does not Granger causes GDP (χ2(3) = 3.73, p-value 0.29). Differently, the test statistics prove that credit is Granger caused by GDP
in both regimes: Regime 1 (χ2(3) = 12.56, p-value 0.00); Regime 2 (χ2(3) = 20.39, p-value 0.00).
16
Fig. 5. Impulse response in the asymmetric VAR, regimes based on medium cycles.
Note: Orthogonalized Impulse Response Function to a standard deviation shock, based on threshold VAR model (10). The Credit variable is Bank
Loans, and the asymmetric effect is investigated under medium credit cycles. Cholesky identification; ordering of the variables is GDP Credit.
credit shocks in either cycle is robust to the inclusion of appropriate controls to for WWII years.
The existence of asymmetric effects in shaping the real impact of financial cycles is particularly meaningful because it helps to
partly reconcile the presence of a financial cycle dominated by long upturn phases that are not necessarily related to (shorter)
business cycle fluctuations and sudden intense credit downturns, which instead amplify recessions.
Acknowledgements
This paper benefited from many comments and suggestions by colleagues and scholars: Federico Barbiellini Amidei, Stefano
Battilossi, Riccardo de Bonis, Rui Pedro Esteves, Alfredo Gigliobianco, Matteo Gomellini, Kilian Rieder, Solomos Solomou, Gianni
Toniolo, and Francesco Vercelli. We also thank the participants at the Third CEPR Economic History Symposium (19–20 June 2015;
OSLO, Norges Bank, the seminar at the University of Crete (May 2015), the 56th Annual Conference of The Italian Economic
Association, Naples), and the FED Conference on Economic and Financial History (May 23, 25, 2016, Richmond). As usual, all
remaining errors are ours. The views expressed do not necessarily reflect those of the Bank of Italy. Bartoletto, Chiarini and Marzano
gratefully acknowledge the University of Naples Parthenope for funding provided (Competitive Research Programme, 2016–2018).
17
Appendix
A.1. The Italian business cycle: a comparison
In this work, we take real GDP as the representative variable for the business cycle, as this is common practice in the empirical
literature and in other papers on Italian business cycles (Gallegati and Stanca, 1998; Delli Gatti et al., 2005; Baffigi et al., 2013).
Although there is no official dating to examine, because our dating procedure is new, and because of the length of the period studied,
we compare our turning points to some of those found in the literature. Baffigi (2013) estimated peaks in Italian GDP over the last
150 years using the methodology introduced by Bai and Perron (1998, 2003). As a benchmark, he also drew up a list of business cycle
peaks by combining the results of some previous studies on different periods. Clementi et al. (2014) put forward a business cycle
analysis based on Baffigi's (2013) reconstruction, but they identify turning points in the deviations from trend series (growth cycles).
Finally, Jordà et al. (2013) provide a classical dating of the Italian business cycle for the last 140 years using the Bry and
Boschan (1971) algorithm, collecting data on GDP from different sources26. In contrast to our dating method, they use real GDP per
capita to determine turning points in economic activity. In Table A.1, we contrast the peaks of these cycles with those of our short
GDP cycles.
Table A.1
Business cycle dating: a comparison of peaks.
Short cycles peaks Baffigi et al. (2013) (2) Baffigi et al. (2013) (3) Clementi et al. Jordà et al. (2013) (5) Difference Difference
(this work) (1) (2013) (4) (1)–(4) (6) (1)–(5) (7)
1862
1866 1866 1867 1865
1870 1870 1870 1870 0 0
1875 1874 1874 1874 1 0
1878 1879 0
1883 1883 1881 1883 2 0
1888 1887 1887 1887 1 1
1891 1891
1896 1897 1897
1900
1907
1913 1914 1913 1913 0
1917 1918 1917 1918 0 −1
1920 1923 – −3
1926 1925 1922 1925 1925 1 1
1929 1929 1930 1929 1929 0 0
1932 0
1935
1939 1939 0
1943 1941 1942
1947 1948 1947
1951
1957
1963 1963
1970 1969 1970
1974 1974 1974 1974 0 0
1980 1981 1980
1992 1990 1991 1989 1992 3 0
1995
2002 2000 2002 2001 2002 1 0
2007 2004
2007 2007 2007 0 0
Notes: (1) Year of peak from our dating. We wittingly excluded 2011 peak, because the most recent years are not included in the other series. (2)
Year of peak worked out as the union of different sources: Ciccarelli e Fenoaltea (2007), Delli Gatti et al. (2005), ISTAT (2001, 2011). (3) Year of
peak based on Bai and Perron (1998, 2003). (4) Year of peak based on Baffigi et al. (2013) data and turning point analysis of growth cycles. (5) Year
of peak based on levels of GDP per capita.
26
As to the sources of data, Jordà et al. (2013) refer to Schularick and Taylor (2012), who generically cite official statistical publications and the
work of individual economic historians.
18
Although the combination of previous studies provides a large number of cycles (peaks in column 2), the most recent works over a
long time span (column 3, 4, and 5) detect less cycle phases. In particular, we obtain the same numbers of peaks (17) as
Jordà et al. (2013), and roughly the same number as Clementi et al. (2014), who found 18. In a few cases (3 out of 17 and 5 out of 18,
respectively), our dating skips peaks pointed out in the other procedures. However, the distance (in terms of years) between the
commonly identified peaks is often equal to zero (columns 6 and 7). Moreover, we do not anticipate or postpone the turning point
systematically.
A.2. business cycle: results from band pass filter
In this section, we illustrate the dating of business and credit cycles we obtain by applying the filtering methodology suggested by
Christiano and Fitzgerald (2003), which is referred to as a CF filter hereinafter . This exercise complements the analysis conducted in
the main text with the NBER approach to show that, although they produce a different dating of the medium-cycle turning points for
GDP (Fig. A.1 right panel) and loans (Fig. A.1 left panel), the statistics based on the band pass filter (bands are set to 8–50 years),
provide some similar qualitative results to NBER approach.
If we compare the GDP and credit cycles turning points reported in Tables 1 and 2 in the main text with the dating illustrated in
the graphs of Fig. A.1, we observe that the number of medium cycles is roughly the same until 1940, whereas major differences in
dating arise in the post-WWII era. Indeed, until 1940, for the financial cycle we identify three medium cycles with the CF filter, which
is the same number we obtained using the NBER approach for bank claims (Table 3) and one more than for bank loans (Table 2). As
for the business cycle, the CF filter identifies three medium cycles compared to the four we obtained using the NBER approach
(Table 1). Differently, after 1940, the number of medium cycles detected with the CF filter is significantly higher compared to the
number found using the NBER approach. Precisely, the medium cycle 4 illustrated in Table 1 for GDP (1939–2007) has no
equivalence in the CF-filtered series, as, by using this methodology, we find five cycles of length that vary from 9 to 19 years. As for
the credit cycles, three medium cycles (1933–73; 1973–93; 1993–2010) were found using the NBER approach, as shown in Table 2,
whereas five medium cycles were found using the CF filter (1940–53; 1953–63; 1963–73; 1973–92; 1992–2008).
Despite these differences, it is clear from Fig. A.1 that synchronization increases in the second part of the sample compared to the
first sub sample. The concordance index for the period up to 1940 is 0.63, whereas in the post-WWII era the index increases to 0.70.
Major differences arise in the post-WWII era because of the statistical properties of the examined time series, GDP and credit,
which in the decades following WWII, have been characterized by a sustained trend growth and low volatility. The sustained path of
growth of the Italian economy is consistent with the evidence for developed countries (Crafts and Toniolo, 2010). Similarly,
Baffigi et al. (2013) have pointed out that Italian GDP has been characterized, in the period following the WWII, by a considerably
lower volatility compared to in previous decades. Both of these reasons play an important role in explaining the different dating
results obtained using the NBER and growth cycle approaches. As indicated by Pagan (1997) and Harding and Pagan (2002), the
expected duration of a classical cycle depends upon the size of the trend growth (direct relationship) and the size of the shocks
(inverse relationship). Accordingly, in our analysis, we did not find any medium-term contractions in GDP in the post-WWII era, as no
one of the occurred recessions were of an intensity that could interrupt the output growth for more than one year. Conversely, when
economic fluctuations are measured according to the growth cycle approach and the trend contribution to growth is ruled out, we
find a considerable number of medium cycles in the post-WWII period because a recession simply means that the pace of growth
slows down.
Fig. A.1. Business (right panel) and credit (left panel) cycle dating according to CF filter.
19
A.3. Key events underlying the turning points in medium-term cycles
In Table A.2, we provide brief descriptions of the main historical events that characterized business and credit cycles in Italy.
Table A.2
Business and credit cycles turning points: the main historical events.
Business cycles (Peak-Trough-Peak) Credit cycles (Peak-Trough-Peak)
1) 1866–1872–1913 1) 1888–1891–1913
•1866- War against Austria; currency crises, suspension of banknotes • 1888–1894- Banks run into difficulties for the burst of real estate bubble;
convertibility. Italy experienced the 1890 Baring crises, which caused a massive flight of
• 1872–1873- Italy shared the international crises. foreign capital.
• 1876- Right-wing was defeated and left-wing came to power. •1892- Scandal and failure of Banca Romana (one of the six banks of issue).
•1876–1899- Slow economic growth and price deflation. •1893- New banking law that established the Bank of Italy and reduced the
•1883- Resumption of convertibility; expansion of credit to the real estate issuing banks from six to three; failure of Credito Mobiliare and Banca
sector. Generale.
•1887–1894- Burst of real estate bubble; international Baring crises. •1907- Stock speculation bubble burst; the main universal banks were
•1900–1913- Giolitti period. Real GDP growth rate (2.6%), inflation below struck, especially Società Bancaria Italiana, which was on the verge of
2%; reduction of interest spending and public debt/GDP ratio. bankruptcy.
2) 1913–1915–1917 2) 1913–1918–1933
•WWI caused severe economic and financial crises; high military expenses; •1914 crises- The slowdown of real economy triggered a drop in the credit
suspension of gold standard, high inflation (23% on average), currency to GDP ratio; universal banks, especially the Banco of Roma, suffered from a
crises (devaluation of lira, 1916–18). high level of nonperforming loans.
•1921 crises- The banking system suffered a severe liquidity crisis when the
war ended; one of the main universal banks, the Banca Italiana di Sconto,
failed; the Banco di Roma was on the verge of collapse.
•1926- New banking law; Bank of Italy monopoly over note issuance.
•1930–31 crises- Italy shared the effect of the 1929 crises; the main
universal banks, Credito Italiano and Banca Commerciale, had serious
solvency problems; the role of the government and the Bank of Italy was
crucial for overcoming the crises; industrial assets of the main banks were
transferred to the IRI.
3) 1917–1921–1929 3) 1933–1945–1973
•Huge growth of domestic and foreign debt (peak of 158% in 1920); high •1934–38- Strong contraction of loans; 1931–1936 banking reform that
inflation rate (43% between 1917 and 18). established the end of universal banks and state control over the major
•1919–20- Social unrest, high unemployment levels. banks.
•1922- Fascism rose to power, robust GDP growth (5,6% between 1922 and •WWII-hyperinflation; Einaudi credit crunch (1947); Italy joined the IMF;
26). growth of industrial credit institutions (IMI, Mediobanca, Efi, Mediocrediti
•1927- Restrictive monetary policies of fascism (gold standard and quota regionali) that supported industrial expansion.
’90) caused a slowdown of GDP growth rates; public debt to GDP reached
104%.
4) 1929–1931–1939 4) 1973–1982–1993
•1929–1935- Deep economic and financial crises. •Italy was struck by the 1970 oil crises; end of Bretton Wood system;
•1931-large contraction of domestic and foreign demand; price reduction. inflation; growth of public debt.
To address the industrial and banking crises, the IMI and IRI were •1981- Divorce between the Bank of Italy and the Treasury.
established. •1990- Privatization of the banking system (Amato banking law)
•1935- The invasion of Ethiopia drove the recovery of the Italian economy. •1993- New banking reform that abolished the functional and temporal
specialization of the Italian credit institutions.
•1990–1996- Severe credit contraction.
5) 1939–1945–2007 5) 1993–1996–2010
•1939–45- WWII caused a sharp reduction of GDP per capita, huge budget •1993–1996- Increase in bad loans/loans ratio; large losses in banks’
deficit, hyperinflation (143% in 1944). securities portfolios.
•1945- Production reached minimum levels. •2008–2010- Drying up of international liquidity; the sovereign debt crises
•1946–1974- Strong GDP growth, low interest rates, trade expansion. affected the credit sector; deterioration of loan quality; worsening of
•1950–61- Industrial added value increased on average by 9%. balance sheets of Monte dei Paschi di Siena (the third Italian bank in terms
•1974–1992- Slowdown of GDP rates of growth; large deficits. of assets) and many smaller banks.
•1970s- Italy was hit by wage, oil and public finance shocks; end of Bretton
Woods system, devaluation of lira.
•1980s- Worsening of the fiscal imbalance.
•1992–2002- Very modest economic growth (almost zero in several years);
disintegration of European Monetary System, Italy joined the Economic and
Monetary Union, currency and financial crises in 1992.
6) 2007–2009–2011
Italy shared the international economic and financial crises; crisis of the
sovereign debt; the budget rules imposed by the EMU have severely affected
economic growth.
20
A.4. Data
Fig. A.1
References
A'Hearn, B., Woitek, U., 2001. More international evidence on the historical properties of business cycles. J. Monet. Econ. 47, 321–346.
Aikman, D., Haldane, A., Nelson, B., 2015. Curbing the credit cycle. Econ. J. 125 (585), 1072–1109.
Artis, M.J., Kontolemis, Z.G., Osborne, D.R., 1997. Business cycles for G7 and European countries. J. Busin. 70, 249–279.
Assenmacher-Wesche, K., Gerlach, S., 2010. Monetary policy and financial imbalances: facts and fiction. Econ. Policy 25 (63), 437–482.
Auerbach, A.J., Gorodnichenko, Y., 2012. Measuring the output responses to fiscal policy. Am. Econ. J. Econ. Policy 4 (2), 1–27.
Babecký, J., Havránek, T., Matějů, J., Rusnák, M., Šmídková, K., Vašíček, B., 2014. Banking, debt and currency crisis: early warning indicators for developed countries.
J. Financ. Stabil. 15, 1–17.
Bai, J., Perron, P., 1998. Estimating and testing linear models with multiple structural changes. Econometrica 66, 47–78.
Bai, J., Perron, P., 2003. Computation and analysis of multiple structural change models. J. Appl. Economet. 18, 1–22.
Baffigi, A., 2013. National accounts, 1861-2011. In: Toniolo, G. (Ed.), The Oxford Handbook of the Italian Economy Since Unification, editor. OUP.
Baffigi, A., Bontempi, M.E., Golinelli, R., 2013. Output potenziale, gap e inflazione in Italia nel lungo periodo (1861-2010): un'analisi econometrica, Bank of Italy.
Econ. Hist. Work. Papers(29).
Bank of Italy (2013), Financial Stability Report, 6, November.
Bank of Italy (2014a), Supplements to the statistical bulletin-monetary and financial indicators-the public finances, borrowing requirement and debt, XXIV, 14 Apr
2014, No. 68.
Bank of Italy (2014b), Supplements to the statistical bulletin-monetary and financial indicators-the financial market, XXIV, 14 November 2014, No. 62.
Barbiellini Amidei, F., Giordano, C., Gigliobianco, A., 2012. Credit to whom? Selective credit policy and economic development in post-WWII Italy. In: Bonin, H.,
Haueter, N.V., Gigliobianco, A., James, H. (Eds.), Public Policies and the Direction on Financial Flows. Studies in Banking and Financial History EABH, June.
Bartoletto, S., Garofalo, A., 2014. Il Ruolo del credito nell'economia Italiana 1861-2013. Mondadori Università, Milano.
Bartoletto, S., Chiarini, B., Marzano, E., 2014. The sustainability of fiscal policy in Italy (1861-2012). Ital. Econ. J. 3, 301–327 XXI.
Bartoletto, S., Chiarini, B., Marzano, E., Piselli, P., (2017), Business cycles, credit cycles and bank holdings of sovereign bonds: historical evidence for Italy (1861-
2013), Quaderni di Storia economica-Banca d'Italia, n. 43.
Bartoletto, S., Chiarini, B., Marzano, E., Piselli, P., 2019. Banking crises and business cycle: evidence for Italy (1861-2016). J. Financ. Econ. Policy 11 (1), 34–61.
https://doi.org/10.1108/JFEP-03-2018-0055.
Battilossi, S., Gigliobianco, A., Marinelli, G., 2013. Resource allocation by the banking system. In: Toniolo, G. (Ed.), The Oxford Handbook of the Italian Economy Since
Unification, editor. Oxford University Press.
Blanchard, O.J., 1997. The medium run. Brook. Papers Econ. Activ. 0 (2), 89–141.
Bordo, M., Haubrich, J.G., 2010. Credit crises, money and contractions: an historical view. J. Monet. Econ. 57, 1–18.
Borio, C., 2012. The financial cycle and macroeconomics: what have we learnt? In: BIS Working Papers No.395.
Bruno, G., Otranto, E., 2004. Dating the Italian business cycle: a comparison of procedures, ISAE. In: Working paper No. 41, January.
Breitung, J., Candelon, B., 2006. Testing for short- and long-run causality: a frequency-domain approach. J. Economet. 132 (2), 363–378.
Bry, G., Boschan, C., 1971. Cyclical analysis of economic time series: selected procedures and computer programs. In: NBER Technical Working Paper No. 20.
Bulligan, G., Burlon, L., Delle Monache, D., Silvestrini, A., 2017. Real and financial cycles: estimates using unobserved component models for the Italian economy. In:
Bank of Italy Occasional Papers, n.382.
Burns, A.F. and Mitchell, W.C. (1946), Measuring business cycles NBER.
Canova, F., 1998. Detrending and business cycle facts. J. Monet. Econ. 41 (3), 475–512.
Christiano, L.J., Fitzgerald, T.J., 2003. The band pass filter. Int. Econ. Rev. 44 (2), 435–465.
Ciccarelli, C., Fenoaltea, S., 2007. Business fluctuations in Italy, 1861–1913: the new evidence. Explor. Econ. Hist. 44, 432–451.
Ciocca, P., Biscaini Cotula, A.M., 1994. Italian financial structure: long-term quantitative aspects (1870-1970). In: Federico, G. (Ed.), The Economic Development of
Italy since 1870, edited by. Elgar.
Ciocca, P., Toniolo, G., 1983. Industry and finance in Italy, 1918-1940. Eur. Econ. Hist. Rev Special issue.
Claessens, S., Kose, M.A., Terrones, M., 2012. How do business and financial cycles interact? J. Int. Econ. 87 (1), 178–190.
Clementi, F., Gallegati, M., Gallegati, M., 2014. Growth and cycles of the Italian economy since 1861: the new evidence. Ital. Econ. J.. https://doi.org/10.1007/s40797-
014-0005-0.
Cochrane, J.H., 1988. How big is the random walk in GNP? J. Polit. Econ. 96 (5), 893–920.
Comin, D., Gertler, M., 2006. Medium-term business cycles. Am. Econ. Rev. 96 (3), 523–551 June.
Cotula, F., Raganelli, T., Sannucci, V., Alieri, S., Cerrito, E., 1996. I Bilanci Delle Aziende Di Credito 1890-1936. Collana storica della Banca d'Italia – Statistiche,
Laterza, Roma-Bari, pp. 3–56.
Crafts, N., Toniolo, G., 2010. Aggregate growth, 1950-2005 in The Cambridge Economic History of Modern Europe 2. Cambridge University Press, Cambridge, pp.
1870 to the Present.
D'Auria, C., Foglia, A., Marullo Reedz, P., 1999. Bank interest rates and credit relationships in Italy. J. Bank. Finance 23.
De Bonis, R., Farabullini, F., Rocchelli, M., Salvio, A., 2012. A quantitative look at the Italian banking system: evidence from a new dataset since 1861. In: Economic
History Working Papers, No 26. Bank of Italy.
De Bonis, R., Silvestrini, A., 2014. The Italian financial cycle: 1861-2011. Cliometrica 8 (3), 301–334.
Delli Gatti, D., Gallegati, M., Gallegati, M., 2005. On the nature and causes of business fluctuations in Italy, 1861-2000. Explor. Econ. Hist. 42, 81–100.
Drehmann, M., Borio, C., Tsatsaronis, K., 2012. Characterizing the financial cycle: don't lose sight of the medium term!. In: BIS Working Papers, No 380.
Drehmann, M., Tsatsaronis, K., 2014. The credit-to-GDP gap and countercyclical capital buffers: questions and answers. In: BIS Quarterly Review, pp. 55–73.
EBF, 2011. Credit cycles and their role for macro-prudential policy. Eur. Bank. Feder November 2011.
ECB, 2012. Money and credit growth after economic and financial crisis – a historical global perspective. ECB Month. Bull. 69–85 Feb.
ECB, 2013. Stylised facts of money and credit over the business cycle. ECB Month. Bull. 18–22 Oct.
Ercolani, P., 1975. Documentazione statistica di base. In: Fua`, G. (Ed.), Ed. Lo Sviluppo Economico in Italia III Franco Angeli, Milano.
Ferri, G., Garofalo, P., 1994. La crisi finanziaria nella grande depressione in Italia. Ricerche Per La Storia della Banca d'Italia V Laterza editore.
Fenoaltea, S. (2006), L'economia italiana dall'Unità alla grande guerra, Roma-Bari, Laterza.
Friedman, M., Schwartz, A.J., 1963. A Monetary History of the United States 1867-1960. Princeton University Press, Princeton.
Garofalo, P., Colonna, D., 1999. Statistiche creditizie. In: Cotula, F. (Ed.), Stabilità e sviluppo negli anni Cinquanta, vol.3 Collana storica della Banca d'Italia –
Statistiche, Laterza, Roma-Bari.
Gallegati, M., Stanca, L.M., 1998. Le Fluttuazioni Nell'economia italiana, 1861-1995. Giappichelli.
Giannone, D., Lenza, M., Reichlin, L., 2012. Money, credit, monetary policy and the business cycle in the euro area. In: CEPR Discussion Paper No. DP8944.
Gigliobianco, A., 2006. Via Nazionale, Banca d'Italia e Classe Dirigente. Cento anni di storia, Donzelli.
Gomez-Gonzales, J.E., Ojeda, J., Tenjo-Galarza, F., Zarate, H.M., 2014. Testing for causality between credit and real business cycle in the frequency domain: an
21
illustration. Appl. Econ. Lett. 21 (10), 697–701.

Haavio, M., 2012. Financial Cycles and Business Cycles: Some Stylized Facts. Bank of Finland BoF on line, 1.
Haavio, M., Mendicino, C., Punzi, M.T., 2013. Financial and economic downturns in OECD countries, Financial and economic downturns in OECD countries, Bank of
Finland. In: Research Discussion Papers 35 /2013.
Harding, D., Pagan, A., 2002. Dissecting the cycle: a methodological investigation. J. Monet. Econ. 49, 365–381.
Harding, D., Pagan, A., 2006. Synchronization of cycles. J. Econometrics 132, 59–79.
ISCO, (1962), Analisi di 122 serie storiche relative al sistema economico italiano secondo la metodologia del National Bureau of Economic Research, ISCO, Rome.
ISTAT, 2001. Annuario Statistico Italiano. Roma: Istat.
ISTAT, 2011. Rapporto Annuale. La Situazione Del Paese Nel 2010. Roma: Istat.
James, H., O'Rourke, K.H., 2013. Italy and the first age of globalization 1861–1940. In: Toniolo, G. (Ed.), The Oxford Handbook of the Italian Economy Since
Unification, editor. OUP.
Jordà, O., Schularick, M., Taylor, A.M., 2013. When credit bites back. J. Money Credit Bank 45 (S2), 3–28.
Juselius, K., 2006. The Cointegrated VAR Model. Oxford University Press.
Koop, G., Pesaran, M., Potter, S.M., 1996. Impulse response analysis in nonlinear multivariate models. J. Economet. 74 (1), 119–147.
Liang, N., 2017. Financial regulations and macroeconomic stability. Keynote Address At the International Finance and Banking Society. Said Business School, Oxford
University July 15-17, 2017.
Lo, M.C., Zivot, E., 2001. Threshold cointegration and nonlinear adjustment to the law of one price. Macroecon. Dynam. 5 (4), 533–576.
Meller, B., Metiu, N., 2017. The synchronization of credit cycles. J. Bank. Finance 82, 98–111.
Mendoza, E.G., Quadrini, V., 2009. Financial integration, financial development, and global imbalances. J. Pol. Econ. 117 (3), 371–416.
Messori, M., 2002. Consolidation, ownership structure and efficiency in the Italian banking system. BNL Quart. Rev. 221 June.
Minsky, H.P., 1964. Longer waves in financial relations: financial factors in the more severe depressions. Am. Econ. Rev. LIV, 3, June.
Mumtaz, H, Pinter, G., Theodoridis, K, 2018. What do VARS tell us about the impact of a credit supply shock. Int. Econ. Rev. 59 (2), 625–646.
Pagan, A., 1997. Policy, theory and the cycle. Oxford Rev. Econ. Policy 13 (3), 19–33.
Perri, F., Quadrini, V., 2002. The great depression in Italy: trade restrictions and real wage rigidities. Rev. Econ. Dynam. 5, 128–151.
Pesaran, H.H., Shin, Y., 1998. Generalized impulse response analysis in linear multivariate models. Econ. Lett. 58, 17–29.
Rajan, R.G., Zingales, L., 2003. The great reversals: the politics of financial development in the twentieth century. J. Financ. Econ. 69 (2003), 5–50.
Rossi, N., Sorgato, A., Toniolo, G., (1993). I conti economici italiani: una ricostruzione statistica, 1890–1990. Rivista di Storia Economica X, 1.
Runstler, G, Vlekke, M, 2017. Business, housing, and credit cycles. J. Appl. Economet. 1–15. https://doi.org/10.1002/jae.2604.
Schularick, M., Taylor, A.M., 2012. Credit booms gone bust: monetary policy, leverage cycles, and financial crises, 1870-2008. Am. Econ. Rev. 102 (2), 1029–1061.
Tamirisa, N.T., Kabundi, A.N., Igan, D.O., De Simone, F., Pinheiro, M., 2009. Three cycles: housing, credit, and real activity. In: IMF Working Papers, October.
Tong, H., 1990. Non-Linear Time Series: A Dynamical System Approach. Oxford University Press.
Watson, M.W., 1994. Business-cycle durations and postwar stabilization of the U.S. Economy, American Economic Review 84 (1), 24–46.
Zarnowitz, V., 1992. Business Cycles: Theory, History, Indicators, and Forecasting. University of Chicago Press.
22

Business Cycles, Credit Cycles, and Asymmetric Effects of Credit Fluctuations Evidence From Italy For The Period of 1861-2013

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Business Cycles, Credit Cycles, and Asymmetric Effects of Credit Fluctuations Evidence From Italy For The Period of 1861-2013

Uploaded by

Copyright:

Available Formats

Journal of Macroeconomics 61 (2019) 103130

Contents lists available at ScienceDirect

Business cycles, credit cycles, and asymmetric effects of credit

ARTICLE INFO ABSTRACT

using quarterly data from 1970 to present day.

and Tsatsaronis, 2014; Bartoletto et al., 2019).

3. Short and medium cycle

3.2. Evidence of medium-term fluctuations

4. Business and credit cycles dating

4.1. The business cycle

Fig. 1. Standardized spectral density (growth rates).

Medium cycle 1 1866 1913 1872 6 41 47 −4.1 106.9 −0.7 1.8

4.2. The credit cycle

short 1 1872 1879 1873 1 6 7 −13.2 79.6 −13.2 10.2

Medium cycle 1 1888 1913 1891 3 22 25 −11.1 124.8 −3.8 3.8

5. Comparing business and credit cycles

5.1. Measures of synchronization

E (CIYC ) = E (SY , t ) E (SC, t ) + (1 E (SY , t ))(1 E (SC , t )) (2)

CI 0.75 0.80 0.71 0.73 0.80 0.88

Short cycle Short cycle Short cycle

CI 0.73 0.76 0.65 0.69 0.81 0.82

5.2. Business and credit cycles: a Granger causality analysis

6. The real impact of financial cycles

6.1. Descriptive evidence from univariate analysis

From credit to GDP (b)

a) Probability of GDP recession

Const 3.9*** 3.5*** 4.2*** 3.7*** 3.9*** 3.7***

6.2. Evidence from multivariate analysis: asymmetric responses to credit shocks

Fig. 3. Impulse response of the linear VAR.

A.1. The Italian business cycle: a comparison

A.2. business cycle: results from band pass filter

A.3. Key events underlying the turning points in medium-term cycles

illustration. Appl. Econ. Lett. 21 (10), 697–701.

You might also like

Const 3.9* 3.5* 4.2* 3.7* 3.9* 3.7*