Characterizing India’s Financial Cycle

1. Introduction

The outbreak of COVID-19 and the resultant outcomes disrupted the real economy, with India’s GDP declining by 24.4% in 2020:Q2 and 7.3% in 2020:Q3. While heightened volatility in the financial market heavily impacted business and consumer sentiment, the resultant effects on the real economy were largely subsided due to a host of policy measures by the government and the Reserve Bank of India. Prior to the COVID-19 pandemic, India’s economic activity was decelerating alongside a large overhang of balance sheet stress. The gross non-performing assets (GNPA) of banks rose from 2.3% of total advances in 2007–2008 to 11.2% in 2017–2018. Though the GDP growth was somewhat better during 2014–2016, the business sentiment was weak and the investment rate was trending downward. The co-existence of financial sector stress along with depressed economic activity reminds us of the ongoing global debate on financial cycle and its relationship with enduring economic depression. The relevance of the financial cycle analysis has increased in the post global financial crisis (GFC) period to explain macro-financial dynamics (Borio, 2012) and to predict systemic banking crises (Drehman et al., 2012; Schüler et al., 2017). In fact, the ‘lost decade’ of Japan and other financial crises in emerging market economies during the mid-1990s were preceded by either prolonged credit booms or asset price booms. These developments draw the attention of researchers to explore the relationship between financial variables and business cycle fluctuations. Studies have found that a deeper recession takes place when the downturn in the business cycle coexists with a downturn in the financial cycle (Claessens et al., 2012; Drehmann et al., 2012). Consequently, central banks, mainly in the advanced countries, have started using financial cycle as a basis for countercyclical macroprudential policies to tame the amplifying effects of financial disruptions on the overall economy.

Deciphering the financial cycle is critical in policymaking for several reasons. First, economic crises like the GFC have highlighted that there exists a strong connection between the financial sector and the real economy. Economic theory suggests that ‘wealth effect’ and ‘change in expectations’ are the two broad channels through which financial sector turbulence can spillover to the real economy. Second, it has now become evident that price stability does not ensure financial stability per se. The experience of the GFC showed that a financial crisis can be preceded by a long period of low and stable inflation. Therefore, basing policy decisions solely on the business cycle and inflation trajectory may obscure risk build-up on the financial side of the economy. Third, the literature highlights that the financial cycle can predict the recession much in advance and hence argues for monitoring the cycle to dampen its effects through appropriate policy responses (Claessens et al., 2012; Borio, 2014; Borio et al., 2018). Contemporary studies in other countries indicate that a financial cycle is typically longer and has a larger amplitude than the business cycle (Borio, 2014; Drehmann et al., 2012; Ng, 2011; Verona, 2016). If this is true for India, then a financial cycle can be a better metric for guiding counter-cyclical macro-prudential measures aiming at long-term financial stability.

We choose India for our analysis for several reasons. First, there is hardly any study documenting the properties of India’s financial cycle. Second, India being the sixth-largest economy in the world draws the attention of experts and investors to study the macro-financial spillover effects between India and the rest of the world. Third, liquid financial markets with attractive returns encourage financial investors to invest in India. Fourth, with a population of over 1.3 billion people, India provides opportunities for a greater market for world products and services. Hence, to make a prudent investment decision to avoid substantial losses and to frame policies to dampen the effects of financial shocks, it is essential to understand the characteristics of India’s financial cycle.

Given the above background, we attempt to measure and study the characteristics of the financial cycle in India. We have followed a three-step approach to analyze the financial cycle. In the first step, we employ turning point analysis and spectral analysis on individual financial variables to get the prima facie evidence on the underlying financial cycle. Spectral analysis has a peculiar advantage that no a priori assumption is needed to derive the duration and amplitude of the cycle. After confirming the evidence from the first step, we extract the cyclical component of variables by using frequency domain based asymmetric band-pass filter method. In the third step, the extracted cycles of select financial variables are combined by using principal component analysis (PCA) to obtain an aggregate measure of the financial cycle.

The rest of the article is organized into five sections. The following section is devoted to the literature review. Data and methodology are discussed in Section 3 and Section 4, respectively. Section 5 deliberates upon the empirical analysis on financial cycles and the last section provides some concluding observations along with policy implications.

2. Literature Review

The concept and definition of the business cycle are well-known, but they are still evolving for the financial cycle. Business cycle can be described as the rise and fall in aggregate economic activity over a period and can be measured using real GDP data. Financial cycle, on the other hand, maps out the expansions and contractions in the financial activities. So far, there is no consensus on the definition of financial cycle. As Borio (2012) defines, financial cycle can be best thought of as ‘the self-reinforcing interactions between perceptions of value and risk, attitudes towards risk, and financing constraints, which translate into booms followed by busts’. These interactions can magnify economic fluctuations and possibly lead to serious financial distress. The financial cycle is not observed directly. Instead, it is extracted from appropriate macro-financial variables such as credit, credit to GDP ratio, equity prices, house prices, and so on, using different econometric or statistical techniques.

Until now, only a few papers have tried to measure the financial cycle and investigate its statistical properties. There is a broad agreement in the literature that the financial cycle is longer and has a greater amplitude than the business cycle, and it generally peaks before the financial crisis. However, there is no consensus on the methods of measurement and how to gauge the state of overall financial cycle by using a single yardstick.

The earlier strand of literature describes financial cycle indirectly by relating financial variables, namely, credit or asset prices to real economic activities (Aizenman et al., 2013; Borio et al., 1994, 2016; Detken & Smets, 2004; Goodhart & Hofmann, 2008; Schularick & Taylor, 2012), by developing leading indicators in early warning system (Alessi & Detken, 2011; Borio & Drehmann, 2002, 2009), and by investigating the predictive power of financial indicators in forecasting economic activity (Borio & Lowe, 2004; English et al., 2005; Espinoza et al., 2012; Khundrakpam et al., 2017; Ng, 2011). The direct measurement and characterization of the financial cycle started after the global financial crisis. These papers mainly rely on four approaches: turning point analysis, frequency-based filters, spectral analysis and unobserved component model-based filters.

The first approach, turning point analysis, goes back to a long tradition of identifying business cycles by dating their peaks and troughs, started by Burns and Mitchell (1946). Claessens et al. (2011, 2012) are among the first ones to use turning point analysis for many countries to study the characteristics of financial cycles. Particularly, they identify the peaks, troughs, and slopes in credit, property prices, and equity prices for many countries and conclude that the cycles in these series tend to be longer and highly synchronized. They also provide evidence of strong linkages between financial and business cycles.

The second approach uses statistical frequency-based filters to identify financial cycle. Such filters require the user to pre-specify the range of cycle frequencies, which is why they are also known as non-parametric filters. Aikman et al. (2010, 2015), using frequency-based filter, provide the evidence of a distinct credit cycle, whose length and amplitude are much greater as compared to those of business cycle.

The third approach uses spectral decomposition of respective financial variables (Pontines, 2017; Strohsal et al., 2017). This approach analyzes the complete (frequency) spectrum which provides information of all possible cycles included in the data.

These studies examine the characteristics of the financial cycle taking individual series whereas a composite measure of financial cycle has been analyzed by several researchers recently (e.g., Drehmann et al., 2012; Einarsson et al., 2016; Schüler et al., 2015; Stremmel, 2015). There are three popular approaches to combine multiple variables into a single measure of financial cycle—(a) by taking averages (Drehmann et al., 2012; Schüler et al., 2015); (b) using principal component analysis (PCA) (Hiebert et al., 2014); and (c) employing unobserved components models (Einarsson et al., 2016).

The fourth approach applies the Kalman filter to the observed data series to extract the financial cycle (de Winter et al., 2017; Galati et al., 2016; Menden & Proaño, 2017; Schüler et al., 2015).

One must note that the empirical findings in the literature on the financial cycle are mostly based on advanced countries’ data. Further, the relevant literature in the Indian context is almost absent, though there are a few studies on credit cycles. For example, Banerjee (2012) studied the linkages between credit and growth cycles of India. She found a significant transformation in the direction of causality between output and credit over time. While output seems to be driven predominantly by credit in the pre-1980s period, it showed nearly no relationship with credit during the 1980s. In contrast, credit is primarily driven by output in the post-1991 period. Further, the study provides an indication of shorter credit cycles possibly because of the chosen methodology to construct the cycles. Anusha. (2015) examined the relationship between bank credit and industrial production growth cycles in India and the US and found a strong coherence between them with credit leading output in the US and the reverse holding true for India. In contrast to the literature, she found evidence of a much shorter credit and growth cycle of about 3 years for both the US and India. She attributed this finding to the choice of reference variable and the use of the multitaper method of the spectrum. A recent study in the Indian context finds that business cycle leading financial cycle through the expectations channel (Banerjee et al., 2021). To be best of our knowledge, there is no study found in the Indian context that examines the properties of financial cycles per se.

While no method is perfect, each one has its unique advantage. All the approaches are based on certain assumptions which are subject to criticisms. The turning point analysis is designed for studying business cycle properties, the frequency based band-pass filter requires a pre-specification of frequency bands and the spectral analysis provides the characteristics of cycles without providing any information about the evolution of the cycle. Therefore, we followed a three-step approach to study the financial cycle in India. First, we use turning point analysis and spectral method to examine empirical properties of cycles of individual financial variables. This approach guides us, in the second step, to specify the frequency range in the band-pass filter to estimate the cycles. In the third step, we combine the extracted cycles of individual variables by using PCA to construct an aggregate measure of the financial cycle and study its linkages with other macroeconomic variables.

3. Data

To provide an aggregate measure of the financial cycle, we focus on cycles in four different market segments which together constitute the core of financial intermediation and spillover effects. Precisely, we examine cycles in credit, housing, equity prices and exchange rates. Credit is a natural choice to analyze the financial cycle as it is the single most important link between savings and investment. Cycles in housing, equity and forex markets are chosen because of their feedback effects which amplify the shocks in the financial system. While movements in equity prices are associated with the leverage cycle, the link between the domestic financial cycle and the global financial cycle is captured through exchange rate cycle.

A combination of volumes and prices data on several financial variables is used in the analysis of financial cycle. The financial variables include real (non-food) bank credit, credit-to-GDP ratio, real equity prices, real effective exchange rate (REER), and real house prices. Credit, equity prices, and house prices are deflated by consumer price index of industrial workers to convert them into real variables. Since long timeseries data of house prices are not available for India, we use construction sector GDP deflator as a proxy of house prices as it is strongly correlated with the property price index available for the period 2009:Q1–2020:Q3. We also use real GDP data to study the business cycle. Since the official data of quarterly real GDP are not available before 1996:Q2, the data from 1980:Q2 through 1996:Q1 are taken from (Bhoi & Behera, 2017). Data of different financial variables are available at various frequencies and time periods. For the study, all the variables are considered at quarterly frequency, seasonally adjusted using Census X-13-ARIMA method. The maximum length of the series is retained to estimate the individual cycles and spectral densities (Appendix I). As the financial cycle usually takes a long time to complete, it calls for a longer data span than is usually required for most other macroeconomic analysis.

4. Methodology

As mentioned in Section 2, several approaches have been discussed in the literature to characterize financial cycles, which are mostly in line with the study of business cycles. Our main aim is to extract cycles from the variables which are generally thought to have the characteristics of the financial cycle. Hence, we use different approaches, namely, spectral methods, turning point analysis, and frequency-based filters together to identify the financial cycle.

Spectral analysis decomposes a time series into underlying sine and cosine functions of different frequencies. In this method, a spectrum—the Fourier transformation of the auto-covariance function—is estimated to determine those frequencies that appear particularly strong or important. The graphical representation of spectral density of the underlying variable provides the relative importance of different frequencies/cycles in explaining the total variation in the data. This method has two major advantages: (a) this procedure does not require any a priori assumption of the cycle’s length/duration; and (b) a very long cycle can be detected, even if the sample is small. We closely follow the approach adopted by Strohsal et al. (2017) in this regard.

The turning point analysis, a non-parametric method, requires a pre-specified rule which defines a complete cycle in terms of minimum number of periods of increase (expansion phase) and decrease (recession phase) in the relevant variable. The approach was originally introduced by Burns and Mitchell (1946) to date business cycle and it became popular after the development of an algorithm by Harding and Pagan (2002) to locate the turning points in the log-level of a series. The algorithm looks for maxima and minima in a series over a given period and selects the pairs of adjacent, locally absolute minima and maxima that satisfy certain censoring rules. Specifically, the algorithm identifies a peak in a quarterly series x_t at time t, when:

{ [ ( x t − x t − 2 ) > 0 , ( x t − x t − 1 ) > 0 ] a n d [ ( x t + 2 − x t ) < 0 , ( x t + 1 − x t ) < 0 ] } .

Similarly, a cyclical trough occurs at time t, if:

{ [ ( x t − x t − 2 ) < 0 , ( x t − x t − 1 ) < 0 ] a n d [ ( x t + 2 − x t ) > 0 , ( x t + 1 − x t ) > 0 ] } .

On the other hand, frequency-based filters require to pre-specify a frequency range (λ_low to λ_high) to extract cycles from an underlying series. It is essential to know the frequency range a priori and therefore, the cyclical pattern of the series differs with the change in specification of the frequency range. Since the literature on the financial cycle in India is still in its nascent stage, any a priori assumption about the frequency range may lead to biased results. Because of this limitation, we adopted a multi-step approach. We start with spectral method and turning point analysis which provide enough evidence to identify the lower and upper band of the cycles in asymmetric band-pass filter. Based on our analysis and considering the empirical evidence available in the literature, a range of 8–30 years is used to extract the medium-term cycle. Additionally, 1–8 years range is used to estimate short-term cycles following Einarsson et al., 2016.

5. Empirical Analysis

5.1 Characteristics of Financial and Business Cycles

The cyclical phases can be characterized as duration, amplitude and slope. While duration measures the length of a cycle, amplitude assesses the extent of change and slope gauges the speed of a given cyclical phase. The duration of a downturn can be estimated by counting the number of quarters between a peak and the next trough while the duration of an upturn is represented by the number of quarters from trough to peak. The amplitude of a downturn (upturn) measures the change in a variable from a peak to the next trough (from a trough to the level reached in the first four quarters of an expansion phase). Lastly, the slope of a downturn or upturn can be measured as a ratio of the respective amplitude to its duration.

To assess characteristics of India’s financial and business cycles, we have first employed turning point analysis on each variable (Table 1). ¹ While all the variables produced an almost equal number of troughs and peaks, the frequency was higher in the case of equity prices (each 20 expansions and contractions) and exchange rate (15 expansions and 16 contractions) and lower for house prices (each 6 upturns and downturns) ² and credit (8 expansions and 7 contractions). However, we observed only two troughs and one peak in the GDP for the period 1980–2020. As this methodology tracks absolute declines and increases, the likelihood of finding a recession is lower given India’s high trend growth record. ³ Therefore, turning point analysis is employed on year-on-year growth rates of the GDP series. This has produced each 17 upturns and downturns for the sample period, with the latest trough at 2020:Q2 when the GDP declined by 24.4% due to COVID-19.

OPEN IN VIEWER

Table 1.

Characteristics of Financial Cycles

Variables	Amplitude		Duration		Slope
	Expansion	Contraction	Expansion	Contraction	Expansion	Contraction
Credit	73.01	–4.67	29.00	3.14	2.52	–1.49
House prices	28.40	–6.15	13.00	2.83	2.18	–2.17
Equity	43.64	–35.38	5.89	6.05	7.40	–5.85
Exchange rates	8.11	–10.82	6.27	5.53	1.29	–1.95
GDP	6.09	–7.76	4.50	4.76	1.35	–1.63

Source: The authors.

Notes: Amplitude is the average change from trough to peak (for upturn) or peak to trough (for downturn). Duration is the average number of quarters between troughs and peaks (for expansion) or peaks and troughs (for contractions). Slope is the ratio of amplitude to duration.

Result for GDP is based on GDP growth rates.

As evident from Table 1, the amplitudes are much larger in the expansion phase than in the contraction phase for both credit and house prices. Although amplitudes of expansion and contraction phases are broadly similar in case of equity prices, exchange rate, and GDP; it is much smaller for GDP growth and much larger for credit. The amplitude of exchange rate is larger in the downturn or depreciation phase than in the upturn or appreciation phase. The expansion phase of credit and house prices, on average, lasts for the longer than their contraction phase. Among various cyclical phases, credit witnessed the longest duration of the expansionary phase of 16.5 years during 1974Q4 through 1991Q2 while equity prices experienced many short duration expansionary phases of 2 quarters (e.g., 1970Q1 to 1970Q3). The slope of the expansion phase is typically larger for credit and equity prices than that of the contraction phase. In case of business cycle, the slope is marginally higher in the recessionary phase than the recovery phase.

Next, we employ spectral analysis to gauge the duration of cycles in different financial and macroeconomic variables. The estimation is conducted for both pre-reform period, that is, period prior to 1991, and post-reform period. The estimation results are presented only for post-reform period in the case of house prices and credit to GDP ratio because of data unavailability. The estimated periodograms of year-on-year log changes of different variables (except credit to GDP ratio) are plotted Figure 1. ⁴ The periodogram, a plot of spectral density, depicts the contribution of variance in the data series at the respective frequency or period. Note that the spectral densities are shown for the range [0, π/4], that is, for period of ∞ to 2 years. The main cycle ⁵ length, expressed in terms of number of years, is measured at the global peak (λ^max) of the spectrum. To approximate the variance contribution of the main cycle’s amplitude, we report the spectral mass, ⁶ measured in percentage points, located around λ^max. We choose a symmetric frequency band with a length of about π/20. ⁷

OPEN IN VIEWER

Figure 1.

Source: The authors.

OPEN IN VIEWER OPEN IN VIEWER OPEN IN VIEWER OPEN IN VIEWER OPEN IN VIEWER

The average length of the cycles specified in terms of main cycle length for GDP is estimated at 5.15 years. The estimation results also suggest that about 65% of the spectral mass is concentrated in 2 to 8 years range. Both these facts confirm that business cycle is relatively a short-duration phenomenon and lasts between 2 to 8 years. The amplitude of the peak is estimated at around 0.7.

In case of credit to GDP ratio, the location of the peak is at 16.1 years. On the other hand, the periodogram of bank credit has seen a remarkable shift during the pre- and post-reform periods. The average length of main cycles in credit is measured at 18.5 years in the post-1991 period as compared to much smaller cycles of about 5.5 years in the pre-1991 period. The amplitude and the spectral mass have also seen an increase for the cycle range of 8 to 32 years. At peak, the amplitude is much higher for the financial cycle than the business cycle.

The main cycle length of equity prices has also increased from 4.1 years in the pre-1991 period to 6.5 years in the post-1991 period. On the other hand, the main cycle length of exchange rate broadly remained similar and shorter duration though the amplitude of the cycle has reduced significantly in the post-reform period. Further, the amplitude of equity price increased slightly.

Overall, the results from spectral analysis suggest that the average length of the credit cycle is much longer than that of equity prices and exchange rate. Moreover, the length of the cycle estimated from exchange rate is almost identical to the duration of the business cycle. Credit to GDP ratio, house prices, and equity prices experience cycles of much longer duration (particularly, in the post-reform period) than business cycle but shorter relative to the credit cycle. As documented in the literature, we found the evidence suggesting business cycle to be of shorter duration of 2 to 8 years compared to credit cycle of 8 to 32 years.

With the above observations from both turning point analysis and spectral density, we specify two ranges ⁸ to extract short-term cycles (1 to 8 years range) and medium-term cycles (8 to 30 years range) using asymmetric band-pass filter as proposed by Christiano and Fitzgerald (2003). Basically, the identification of short-term cycles in the data is motivated by the works of Comin and Gertler (2006) and Drehmann et al. (2012) in the context of business cycle and financial cycle, respectively. The idea behind the extraction of both short and medium-term cycles is to understand their relative importance in explaining the overall behavior of each variable. Therefore, we calculate the volatility of the cycles using their standard deviations and compare the ratio of volatility of medium-term to that of the corresponding short-term cycles in each variable. We also provide a comparison between the relative volatilities in the pre and post-1991 periods. To know how the volatility has evolved over time, the relative volatility of post-1991 over pre-1991 period is also calculated.

Table 2 presents the relative volatilities of individual variables. A ratio greater than unity implies that medium-term cycles are more volatile (and have higher amplitudes) than short-term cycles. A higher value of relative volatility indicates that a larger portion of variability of the series is explained by the medium-term cycles. The relative volatilities at 3.89, 1.52, and 1.77 for credit, house prices and equity prices in the post-1991 period, respectively, are indicating the dominance of cycles at medium-term frequencies in explaining their overall variation. However, relative volatility has fallen significantly in the case of exchange rate suggesting a decline in importance of medium-term cycles in shaping exchange rate behavior. A comparison between the volatility of medium-term cycles for the pre- and post-1991 period indicates that the amplitudes of medium-term cycles have risen in the post-reform period for all the variables (except exchange rate). In sum, the results suggest an increased dominance of medium-term cycles in explaining the overall behavior of financial variables in the post-reform period.

OPEN IN VIEWER

Table 2.

Relative Volatility of Cycles

Period	Credit	House Prices	Equity Prices	Exchange Rate	GDP
Full sample	2.19	1.52	1.65	1.63	0.84
Pre-1991	1.41		1.50	2.45	0.61
Post-1991	3.89	1.52	1.77	0.99	0.88
Post-1991 over Pre-1991*	1.45		1.31	0.43	1.83

Source: The authors.

Note: Relative volatility is calculated taking the ratio of standard deviations of medium-term to short-term cycles; * - Ratio of standard deviations of medium-term cycles in pre and post-1991 periods.

5.2 Measures of Financial Cycle

With a view to construct an overall measure of the financial cycle for India, medium to long term cycles of financial variables are extracted from individual series applying band-pass filter and specifying a band of 8 to 30 years. Figure 2 plots the extracted medium-term cycles of each variable. The duration and amplitudes of the medium-term credit cycle have increased since the mid-1990s. Similarly, the amplitudes of medium-term equity price cycle are found to be much higher. But the duration of the cycle has not changed over the years. On the contrary, medium-term cycles of exchange rate have weakened since the mid-1990s. House price cycles, though have smaller amplitudes, last for several years. Taking the clue from these plots about the emergence of financial cycles in India since the mid-1990s, we have combined the medium-term cycles of credit, equity prices and house prices using PCA to construct an aggregate measure of the financial cycle.

OPEN IN VIEWER

Figure 2.

Source: The authors.

The medium-term cycles of the financial variables for a common period, that is, 1996Q2 through 2020Q4, are presented in Figure 3. All other variables share a common cyclical pattern and characteristics though exchange rate for the initial years was moving in a different direction. The correlation coefficient between the cycles of credit and house prices is the highest at 0.81, followed by 0.76 between house price and equity price cycles and 0.69 between credit and equity price cycles. This strong association is elucidated by the fact that the cycles share some common components. Such a common feature, what the financial cycle aims to capture, is estimated by using principal component analysis.

OPEN IN VIEWER

Figure 3.

Source: The authors.

The above results provide evidence of the existence of a financial cycle in India, particularly, in the post-reform period. To obtain an aggregate/composite measure of the financial cycle, the medium-term cycles of credit, house prices, equity prices and exchange rate are combined by using PCA. ⁹ The factor loadings of the first principal component are used as weights to prepare the composite measure of financial cycle. The estimated first principal component explains about 64% of combined variability in the financial data. ¹⁰ The normalized factor loadings from the PCA show higher weights for both house prices (31.5%), credit (30.5%), and equity prices (27.5%), and lower weight for exchange rate (10.5%). Figure 4 presents an aggregate measure of the financial cycle along with the contributions of individual components. While the overall financial cycle is mainly contributed by the house prices, credit, and equity prices, the role of exchange rate has been negligible. However, the role of house prices has increased significantly since the mid-2000s, with credit and house price cycles are driving the ongoing financial downturn. The figure shows only one clear peak and one clear trough in aggregate financial cycle measure during 1996–2020. The application of turning point analysis on aggregate financial cycle reveals that there were two peaks during 2008:Q2 and 2017:Q3 and two troughs in 2001:Q3 and 2014:Q4. It can also be inferred from the figure that the current downturn in the financial cycle seems to be extended and yet to reach its trough.

OPEN IN VIEWER

Figure 4.

Source: The authors.

We construct alternative measures of financial cycle combining (a) credit and equity price cycles; (b) credit, equity price and exchange rate cycles as long time-series data are available for these three variables. The alternative measures show a strong co-movement with the aggregate measure of the financial cycle (Appendix IV, Figure A1). Employing turning point analysis on these alternative measures, we found that the average duration of contraction and expansion of financial cycle without exchange rate is about 24 quarters each leading to the average length of financial cycles to about 12 years. The inclusion of exchange rate enhances the duration of contraction to 29.5 quarters where the amplitude of the contraction is lower as compared to the previous measure. However, all the measures show an increase in both amplitude and duration of financial cycles in the recent period reflecting the proliferation of financial liberalization since the mid-1990s.

5.3 Application of Financial Cycle

The aggregate measure of the financial cycle can be useful for several policy purposes, particularly as an early warning indicator for detecting exuberance or distress in the financial system. It is often found that business cycle recessions are much deeper when they coincide with the contraction phases of the financial cycle (Claessens et al., 2012; Drehmann et al., 2012) and therefore, the peak of the financial cycle may be seen as a warning sign for financial crisis or distress in the economy. On the other hand, GDP growth tends to be more stable in expansion phases of financial cycles with fading out of recession risks due to a rise in asset prices and a decline in leverage. In view of this, we examine the relationship between business and financial cycles. Further, the connection between the financial cycle and non-performing assets is explored to gauge the predictable power of the latter in providing early warning signals about banking sector stress.

Figure 5 plots both financial and business cycles. The figure implies that the average duration of the financial cycle is much longer than that of the business cycle. The longer duration cycle being more persistent and having greater amplitudes could disrupt the structure of financial system immensely. In case it propagates to the business cycle frequency, it can increase the cost of output stabilization. Additionally, the financial cycle seems to be more volatile (having greater amplitudes) than the business cycle (except during the period of COVID-19 when the amplitude of output decline was much larger). On an average, the volatility of the financial cycle is almost 1.5 times larger than that of the business cycle. The figure does not show a strong link between the business cycle and the financial cycle. Nevertheless, as can be clearly seen from the figure that peak of the financial cycle in 2008:Q2 is associated with much weaker economic activity in the subsequent period. Also, there are periodic co-movement between business and financial cycles during 2011–2015. As long timeseries data are not available, it is difficult to find more such evidence in the short sample period of 25 years. However, as the financial cycle is longer and its association with the business cycle during a downturn is severe, dampening of financial cycle through counter-cyclical policy measures is important to enhance macroeconomic and financial stability.

OPEN IN VIEWER

Figure 5.

Source: The authors.

The relationship between the financial cycle and business cycle is further examined by using Granger causality tests. The causality tests are conducted by taking the conventional business cycle of 1 to 8 years duration and the longer duration business cycle of 8 to 30 years along with the aggregate measure of the financial cycle. The results in Table 3 show that business cycle causes financial cycle while the financial cycle can impact the medium-term business cycles only. Financial cycles being characterized by longer duration and greater amplitude can impact the economy severely. This, in turn, can generate business cycles that last longer. Moreover, the correlation between these two is 0.4, implying a strong association between the long-term cycles in economic and financial activities.

OPEN IN VIEWER

Table 3.

Granger-Causality Test Results

(Sample: 1996:Q2 to 2020:Q4)	Using Business Cycle of 1–8 Years Duration		Using Business Cycle of 8–30 Years Duration
Null Hypothesis:	F-Statistic	P-Value	F-Statistic	P-Value
Business cycle does not Granger Cause financial cycle	3.87	0.01	9,126.15	0.00
Financial cycle does not Granger Cause business cycle	0.01	0.99	12,043.10	0.00

Source: The authors.

Note: Lag length of 4 is selected based on Schwarz Information Criteria.

According to Borio et al. (2018),

During expansions, the self-reinforcing interaction between financing constraints, asset prices and risk-taking can overstretch balance sheets, making them more fragile and sowing the seeds of the subsequent financial contraction. This, in turn, can drag down the economy and put further stress on the financial system.

Therefore, excess leverage ensues from the building up of debt stocks with lenders underestimating the potential risks of the borrowers and collaterals during the tranquil period. When asset prices fall, it reduces the value of collaterals and results in a significant rise in leverages. Excess leverages finally burst and lead to financial distress. Financial cycle, which is an outcome of interactions between leverage and asset prices, is worth monitoring to understand the risks and also conduct suitable macro-prudential policies. Therefore, the upturn of financial cycle needs to be monitored and dampened in case of a sharp upward movement. Similarly, buffers need to be maintained to manage the downturn phase.

Figure 6 presents gross non-performing assets to total advances (GNPA) of the scheduled commercial banks along with the aggregate measure of the financial cycle. It can be viewed from the figure that the upturn phase in the financial cycle is followed by a reduction in stressed assets of banks during 2001–2008; the financial cycle reached its peak in 2008:Q2. The cycle entered the contractionary phase thereafter. Though it is difficult to draw precise inference from the limited number of turning points in the financial cycle in India, it seems that lenders tend to have ignored the risks in the benign period (i.e., upturn phase of the financial cycle) that resulted in rise in GNPA in the downturn. Correct identification of turning points in the financial cycle may, thus, provide guidance for appropriate macro-prudential policies.

OPEN IN VIEWER

Figure 6.

Source: The authors.

The design and effectiveness of macro-prudential policies depend on how best we measure and understand the characteristics of financial cycle. In this study, we have attempted to determine the existence of financial cycle in India by examining the properties of financial cycle in different variables, namely, credit, credit-GDP ratio, equity prices, house prices and exchange rate. The empirical findings clearly indicate the existence of a financial cycle in India. We also find that the amplitudes of financial variables are much larger in the expansion phase than in the contraction phase. The average duration of the business cycle in India is about 5 years as compared to more than 16 years for the credit cycle in the post-reform period. The length of cycles in exchange rate is nearly identical to the duration of the business cycle whereas credit to GDP ratio and house prices experience cycles of much longer duration. On the other hand, equity prices have exhibited both short and medium-term cycles. We also find the rising dominance of medium-term cycles in explaining the overall variation of credit and equity prices in the post-reform period. However, a weakening of the importance of medium-term cycles in shaping exchange rate behavior is observed since the mid-1990s. Overall, the core empirical features of the cycles from individual variables suggest that there exists a financial cycle in India, which has gradually become more prominent with the proliferation of financial liberalization since mid-1990s. The composite measure of India’s financial cycle can be best captured by the joint behavior of credit, house prices, equity prices, and exchange rate wherein the credit, house prices and equity prices contribute the most. Overall, we find a longer duration financial cycle with an average length of about 12 years (6 years of expansion and 6 years of contraction) vis-à-vis a shorter duration business cycle with the overall average length of about 5 years.

We find that the peak of the financial cycle in 2008:Q2 provided the lead information about much weaker economic activity in the subsequent period. Further, we also show that the peak of the financial cycle in India had provided a correct signal about the building up of financial vulnerability much before the increase in stress in the banking sector.

The findings of our study carry critical policy implications. Monetary and fiscal policies are generally designed to stabilize the medium-term business cycle without taking into account financial cycle. This strategy may help to contain recessions in the short run but at the expense of bigger recessions in the future. Given that the length of the financial cycle is longer than that of the business cycle, financial vulnerabilities take some time to grow and the damages they create in the economy require a long time to heal. In an environment where the horizons of market participants are short, and where policymakers attach greater weight to high-frequency movements, the economy can fall into the ‘unfinished recession’ phenomenon. To avoid such a situation, prudential, monetary, and fiscal policies need to focus on the medium to long term as well. While a prudential policy can provide guidance on building up buffers during the boom phase to stabilize the financial system at the time of the bust, fiscal policy can pre-empt a serious balance sheet recession driven by financial bust through bank capitalization. In addition to just focusing on output and inflation, the monetary policy requires to respond systematically to avoid serious building up of financial vulnerabilities and monetary authority to remain watchful of their actions leading to financial sector bubble.