3.6 Conclusion
4.3.1 Constructing the business and financial cycles
As previously mentioned, there is no clear consensus within the literature regarding
a definitive relationship between banks’ profitability and the cycle. Furthermore,
the potential role of the financial as opposed to the business cycle in predicting
banks’ profitability is largely ignored. Against this background, this current study
constructs estimates for both the business and financial cycles to evaluate their
respective impact on the profitability of banks.
The study takes GDP as the representative variable for the business cycle. I
recognise and acknowledge that a multivariate approach would probably be more
suitable, such as the finance-neutral output gap of Borio et al. (2016) or the finance-
augmented cycle of Montagnoli et al. (2018). However, for simplicity, and conforming
to common practice in this literature, this study will utilise the univariate approach.
Unlike the business cycle, the financial cycle has been far less researched and
defined. Further, there is no agreement on the most reliable econometric method
to analyse its statistical properties, this is usually left to the preference of the re-
searcher. Therefore, this study looks at a number of series to attempt to characterise
it. Studies such as Dell’Ariccia et al. (2012), Jordà et al. (2013) and Taylor (2015)
have used credit to capture the financial cycle, on the grounds that credit captures
the boom-bust of the financial sector. A variable commonly used in policy analysis
is the credit-to-GDP ratio. More often, researchers tend to look at measures of
credit and asset prices, most notably real estate prices (see for example Claessens
et al. (2011) and Claessens et al. (2012)). In Drehmann et al. (2012), credit aggre-
gates (particularly the credit-to-GDP ratio) are used as a proxy for leverage, while
property prices are used as a measure of available collateral. Stremmel (2015) finds
that the key ingredients of the best fitted financial cycle measure for Europe include
credit-to-GDP ratio, credit growth, and house prices to income ratio.
Section 4.3 Chapter 4
This study follows Drehmann et al. (2012), Borio (2014), and Stremmel (2015)
in using three financial variables to approximate the financial cycle.6 These are: (i)
credit to the private, non-financial sector; (ii) credit-to-GDP ratio; and (iii) prop-
erty prices. All the macroeconomic data used to construct the business and financial
cycles span the period 1980–2017. Using such long series allows for capturing var-
ious peaks and troughs, and thus better approximates the cycles. The series used
to approximate the cycles are all in real terms (CPI deflated) and in natural log-
arithmic form. This of course excludes the credit-to-GDP ratio which is expressed
in percentage points. Further, they are all normalised to their respective values in
2010 to maintain comparability.
To construct the cycles, the study uses the recent filter-based method proposed
by Hamilton (2017). The most commonly used technique in literature is the HP-
filter. However, despite its popularity, it comes with several drawbacks that must
be considered when deciding whether to use this approach. For example, Hamilton
(2017) argues that the HP-filter involves several levels of differences, so for a random
walk process, subsequent observed patterns are a mere by-product of having applied
the filter rather than reflecting the underlying data generating process. He further
suggests that the filtered values at the end of the sample vary significantly from
those in the middle, and as such are characterised by spurious dynamics. Hamilton
(2017) finally adds that the HP-filter typically produces values for the smoothing
parameter which are vastly at odds with common practice.
In light of this, Hamilton (2017) suggests an alternative concept of deriving the
cyclical component of a possibly nonstationary series. Specifically, he imposes that
the cyclical component of a possible nonstationary series should address the question
of how different is the value at datet+hfrom the value that would be expected based
6In extending this paper, I will adopt a multivariate approach where all three variables will be
combined in a single measure to construct a proxy of the financial cycle.
Section 4.4 Chapter 4
on its behaviour through date t.7 Therefore, he suggests that if an OLS regression
of yt+h is regressed against some constant and thep= 4 most recent values of y as
of date t,
yt+h =β0+β1yt+β2yt−1+β3yt−2+β4yt−3+vt+h (4.1)
the residuals
b
vt+h =yt+h−βb0 −βb1yt−βb2yt−1−βb3yt−2−βb4yt−3 (4.2)
provide an acceptable approach to gain the transient component for a broad class
of underlying processes.
Hamilton (2017) argues that this proposed procedure holds a few advantages
over the much used HP-filter. First, unlike the cyclical component of the HP-filter,
the value of vbt+h will be difficult to predict from variables that pre-date time t. Second, the value of vbt+h is a model-free and assumption-free summary of the data. Thus, regardless of how the data was generated, as long as (1−L)dy
t ≤ 4, there
exists a population projection of yt+h on (yt, yt−1, yt−2, yt−3,1)’, which can be used
to consistently estimate a cyclical component from the data. Therefore, borrowing
from Hamilton (2017) approach, this study extracts the cyclical component from real
GDP, property prices, private credit, and credit-to-GDP ratio, in order to provide
proxies for the business and financial cycles.
7This concept is related to the definition of the trend component of y
t as gt =
limh→∞limp→∞E(yt+h|yt, yt−1, ..., yt−p+1) in Beveridge and Nelson (1981). This limit exists and
can be calculated provided that (1−L)yt is a mean-zero stationary process.
Section 4.4 Chapter 4