Introduction

By changing the monetary policy rate a central bank tries to change the entire spectrum of interest rates, such as money market rates, bond yields, bank deposit and lending rates, and asset prices such as stock prices and house prices. As per the monetary theory, this is known as the monetary transmission mechanism. In fact, the effectiveness of the monetary policy transmission is built on the idea of how much and how fast monetary policy can influence its ultimate goals, viz., price stability and growth. In India, the banking system is the pre-dominant sector for financial intermediation. Thus key to achieving the ultimate objective of the monetary policy lies in how effectively the policy rate changes are transmitted to the deposit and lending rates. In other words, in an effective transmission mechanism, it is imperative that monetary policy signals pass through the banking system without any ‘leakage’ and in a quick time. However, effective transmission requires fulfilment of various pre-conditions. A crucial pre-condition is transparency in the process of pricing loans by banks, not only for customer protection but also for better assessment of transmission by the monetary authority. In India, simultaneously with interest rate deregulation in October 1994, the banking regulator has been stipulating the adoption of a specific benchmark for the pricing of loans by banks. Benchmarks can be either internal to the banks like the cost of funds or lending rates charged to the best customers or they can be external to the banking system, e.g., market-determined T-bill rate.Footnote 1Since the deregulation of lending rates was one of the first steps towards financial sector liberalization, which had begun a couple of years ago, no suitable external benchmark was available for the pricing of loans. Accordingly, the Reserve Bank of India (RBI) mandated an internal benchmark—prime lending rate (PLR)—in October 1994.Footnote 2 In April 2003, the RBI supplanted PLR with the benchmark PLR (BPLR), which was followed by the base rate in July 2010 and the marginal cost of funds-based lending rate (MCLR) in April 2016. None of these benchmarks met the expectations. Effective October 2019, RBI mandated an external benchmark system for retail and micro and small-scale enterprise (MSE) sectors, which was expanded to include medium enterprises, effective April 2020.

Against this backdrop, this study examines the nature of pass-through to lending interest rates in India under three different internal benchmark regimes, viz., BPLR, base rate and MCLR during the period April 2004–July 2019. For econometric analysis, however, our study covers the period Q4:2012–13 to Q2:2018–19, where we study the interest rate channel of monetary transmission in India during the base rate and the MCLR regimes using the quarterly bank-level data of all domestic banksFootnote 3. The beginning of the time period of our study is keeping in view the availability of data on the main variable under study. To the best of our knowledge, it is the first study of its kind in that transmission during two different interest rate benchmark regimes has been compared. Although various studies are available in the literature, which have dealt with the monetary transmission through various channels, this paper makes the first attempt in the literature at examining the degree of passthrough under different internal benchmark regimes Accordingly, it is a new contribution to the literature on monetary policy transmission.

After the introduction dealt with in the previous section, the section “Review of literature” deals with the survey of literature on the related field which is relevant to our study. Section “Stylized facts on monetary transmission in India under various interest rate regimes” discusses the stylized facts covering the three internal benchmark regimes. Section “Methodology and data” deals with the data and methodology for econometric analysis adopted in the study. Section “Results” presents the empirical findings. Concluding remarks are provided in the section “Conclusion”.

Review of literature

Literature on monetary policy transmission and its channels is nearly a century old starting, perhaps with Keynes (1936). Some of the key contributors to the literature over the decades include Friedman and Schwartz (1963), Ando and Modigliani (1963), Tobin (1969), Taylor (1995), Obstfeld and Rogoff (1995), Meltzer (1995), Bernanke and Gertler (1995). These studies have mostly dealt with the working of the traditional monetary or interest-rate channel and credit channel of transmission. More recently, Mohanty and Turner (2008), Mukherjee and Bhattacharya (2011), Mishra et al. (2010), and Trichet (2011) examined the efficacy of various channels of the monetary transmission mechanism for both developed and emerging market economies. While there is a convergence on the idea that money does influence aggregate demand and prices in this literature, there is disagreement over the relative importance of alternative transmission channels.

During the last 20 years, there has been a plethora of studies on monetary transmission in India. Some of these studies have examined the efficacies of various channels in transmission mechanisms in India. For example, Singh and Kaliranjan (2007) highlighted the importance of interest rate channels in monetary policy transmission during the post-reform period of the Indian economy. Mallick and Sousa (2012), Bhattacharya et al. (2010), Aleem (2010), Pandit and Vashisht (2011), Sengupta (2014) and Das (2015) have assessed various aspects of monetary transmission by using different alternative approaches. RBI (2002), Patra and Kapur (2012), Goyal (2008) and Anand et al. (2010) employed the New Keynesian model (NKM) to examine the transmission process for India. Further, Patra and Kapur (2000), Dua and Gaur (2009), Paul (2009), Patra and Ray (2010), Mazumder (2011), Singh (2011) have estimated individual equations under the NKM framework concentrating on the Phillips curve. Some studies like Khundrakpam (2011), Khundrakpam and Jain (2012), Mohanty (2012), Kapur and Behera (2012), Bhoi et al. (2017), etc. have tried to examine the degree of pass-through by using the Indian data.

Most of the initial studies in the international literature have employed aggregate time series data to examine monetary policy transmission in their analyses. Over the last three decades or so, studies have often employed individual bank-level data to capture the effect of bank-specific characteristics to examine the effectiveness of various channels of monetary transmission mechanisms. In this respect, Gambacorta (2008), Were and Wambua (2014), Altavilla et al. (2016), Yang and Shao (2016), Holton and D’Acri (2018), Gambacorta and Shin (2018), Abuka et al. (2019), Sapriza and Temesvary (2020), are important to be mentioned. All these studies have taken bank-level data for different counties to examine the impact of bank-specific characteristics on the transmission mechanism.

There are a few studies available on the Indian economy which used bank-level data to explore the interest rate channel of the transmission process. In this regard, Bhaumik et al. (2010), Das (2013), John et al. (2018), Mishra and Kelly (2017) are worth mentioning. Bhaumik et al. (2010) examined the degree of variation of impact on the reaction of banks to monetary policy due to changes in ownership of banks. The study found that the impact varies due to changes in monetary policy across different types of banks. Das (2013) examined the impact of the financial crisis on some bank-level variables, such as the size, capital and liquidity of banks under the monetary transmission framework. The study found that the net interest margin (NIM) of the banks was impaired for those banks that were capital-constrained and having inferior asset quality. John et al. (2018) find that the deteriorating asset quality of the Indian banks has an adverse effect on monetary transmission in India. Mishra and Kelly (2017) also used the bank-level data to analyse the monetary transmission process in India. The study found that monetary policy shocks have a strong initial and persistent impact on bank lending, while liquidity shocks impact bank lending after a 9-month lag. However, no study has dealt with the role that a benchmark can play in influencing the transmission of monetary policy. This has implications for policy formulation, not only for India, but for other emerging market economies that are similarly placed in this regard in case they intend to change their interest rate setting regime for the financial sector.

Stylized facts on monetary transmission in India under various interest rate regimes

In 1994, the Reserve Bank directed the banks to reveal their prime lending rates (PLRs), which are the lending rates imposed by banks to their prime customers. Subsequently, the Reserve Bank replaced the prime lending rate (PLR) with the benchmark prime lending rate (BPLR) to serve as a reference rate for banks in 2003. Banks were required to compute the BPLR by considering “the cost of funds, operational costs, minimum margin to cover regulatory requirement (provisioning and capital charge), and profit margin” (RBI, 2017). However, the lack of transparency in the process of determination of internal benchmarks under the BPLR system hampered the efficacy of the monetary transmission mechanism. To usher in transparency for making a better assessment of transmission, the Reserve Bank brought forth the base rate system in July 2010 in place of the BPLR system. The lack of uniformity in the manner of calculation of base rate across banks and the inclusion of arbitrary elements in the formula by banks, however, hampered the assessment of transmission during the base rate regime. Besides, the prevalence of price discrimination of old customers vis-à-vis the new borrowers hampered the transmission of outstanding loans (RBI, 2017).

To improve the internal benchmark system, the Reserve Bank instituted the marginal cost of funds-based lending rate (MCLR) system on April 1, 2016. Under this system, banks were expected to determine their benchmark based on the formula prescribed for the calculation of the marginal cost of funds, reducing the scope for discretion from that during the base rate regime. While the MCLR formula was ‘given’ to the banks and, hence, transparent, banks could still play around with the few elements of discretion available to them (RBI, 2017; see Box II.4, p. 16).

From April 2004 to July 2019, the monetary transmission was subject to variable lags across internal benchmark regimes and policy cycles (Table 1). Transmission to lending rates was usually—though not always—higher during the tightening cycle than the easing cycle regardless of the regimes. Unlike the base rate regime, which experienced policy tightening (39 months) and easing (30 months) for relatively similar time periods, the MCLR regime was almost entirely characterized by policy easing with only 8 months of policy tightening (June 2018–January 2019). Transmission during the MCLR regime was muted during April–October 2016 but gathered pace aided by a surfeit of liquidity post-demonetization, which encouraged domestic banks to reduce their saving deposit rates during Q2:2017–18 for the 1st time since the deregulation of saving deposit rates in October 2011 and also their term deposit rates. The reduced cost of funds prompted banks to reduce their MCLRs sharply. Shorn of the one-off demonetization impact on the cost of funds, the performance of the MCLR regime on transmission was not very satisfactory (RBI, 2017).

Table 1 Transmission of Monetary Policy with tightening and easing policy cycles under various regimes of Internal benchmarks (basis points).
Full size table

There is a wide disparity in the manner BPLR as the benchmark was determined opposite the base rate and MCLR. The median BPLR was sharply higher than the median base rate and MCLR (Fig. 1). This is because the BPLR, rather than serving as the benchmark for lending rates to best (prime) customers, typically served as a ceiling for lending rates, with as much as 77 per cent of loans contracted at sub-BPLRs in September 2008 (RBI, 2017). Besides, the correlation coefficient between the BPLR and policy repo rate was found to be very low at 0.2. To the contrary, both the base rate and the MCLR operated as a floor to lending rates, thereby lowering the gap between the benchmark rate and the policy rate.

Fig. 1: Regime-wise transmission of monetary policy.
figure 1

The figure shows the impact of interest rates in response to repo rate (policy rate) changes under different regimes. - BPLR, median base rate, and 1 yr median MCLR. Notes: Data are the median value of the benchmarks.

Full size image

Methodology and data

The focus of the study is to estimate the degree of monetary policy pass-through to domestic banks’ lending rates during the latter half of the Base Rate regime (Q4:2012–13 to Q4:2015–16) and MCLR regime so far (Q1:2016–17 to Q2:2018–19).

Aggregate analysis

Before examining the passthrough of monetary policy to individual banks’ lending rates, we analyse the same at the aggregate level for the entire period under study. Here, we consider four variables, viz. the weighted average lending rate on fresh rupee loans sanctioned by domestic banks (WALRF)Footnote 4, monetary policy variable (WACR)Footnote 5, inflation and real gross domestic product (GDP). Our main objective is to verify the impact of monetary policy changes (WACR) on WALRF. Inflation and real GDP are the control variables of the model. We examined the impact in a vector autoregressive (VAR) framework, for which the impulse response function has been estimated. As per the augmented Dickey–Fuller (ADF) test, all the variables were found to be I (1) as shown in Table 2.

Table 2 Unit root tests for.
Full size table

Then we examined if the variables are cointegrated. Accordingly, we checked the long-run relationship using the Johansen cointegration model. The idea is if they are cointegrated then we go for vector error correction (VECM) model estimation and if they are not cointegrated, then we adopt the VAR model. The Johansen cointegration test results are given in Table 3Footnote 6.

Table 3 Results of cointegration test.
Full size table

Results of the Johansen cointegration test show that there exists a cointegrating relation as per both the trace test and the maximum eigenvalue test, under the 5% level. In other words, there is a stable long-term equilibrium relationship among the variables. We now run the vector error correction model (VECM) to examine both the short-run and long-run dynamics of the series. Conventional ECM for cointegrated series are:

$$\Delta y_t = \beta _0 + \mathop {\sum}\nolimits_{i = 1}^n {\beta _i\Delta y_{t - i}} + \mathop {\sum}\nolimits_{i = 0}^n {\delta _i\Delta x_{t = i}} + \varphi z_{t - i} + \mu _t$$
(1)

where zt−1 is the error correction term and is the OLS residual from the following long-run cointegrating regression:

$$y_t = \beta _0 + \beta _1x_t + \varepsilon _t$$
(2)

and is defined as

$$z_{t - 1} = ECT_{t - 1} = y_{t - 1} - \beta _0 - \beta _1x_{t - 1}$$
(3)

The coefficient of ECT, φ, is the speed of adjustment because it measures the speed at which y returns to equilibrium after a change in x.

It is observed from Table 4 that the relationship between WALRF of domestic banks and WACR is positive and statistically significant at 1 per cent level of significance. From the long-run equation of the cointegrated model, we can infer that a 1 percentage point increase in the WALR leads to a 0.36 percentage point increase in the WALRF (Table 4 and Fig. 2). It is our contention, however, that there is considerable heterogeneity among banks in passing on the monetary impulses to their respective lending rates driven by bank-specific factors. Therefore, it is prudent to estimate the impact of a policy shock in the ΔWACR based on a dynamic panel data regression model (GMM model) controlling for this heterogeneity. The remainder of this paper focuses on the disaggregated level, which is discussed in the following section.

Table 4 Vector error correction model between lending interest rate and policy rateFootnote

Data are seasonally adjusted since they are quarterly data.

.
Full size table
Fig. 2: Impulse response function (response to Cholesky 1-sd (d.f. adjusted) innovations).
figure 2

a Response of walrf to walrf innovations. b Response of walrf to wacr innovations. c Response of walrf to RGDP innovations. d Response of walrf to INF innovations. Note: (1) Data are seasonally adjusted. (2) The shocks in the VAR model are identified by using orthogonal impulse response (OIR). The basic idea is to decompose the variance-covariance matrix so that ∑ = PP′, where P is a lower triangular matrix with positive diagonal elements, which is often obtained by a Choleski decomposition. In our analysis, we have used the Choleski decomposition.

Full size image

To sum up, we find that there exists a co-integrating relationship between the monetary policy rate (repo rate) and lending rate, which indicates the existence of a stable long-term equilibrium relationship between the variables. It is also found that the relationship between WALRF (lending rate) of domestic banks and WACR (policy variable)Footnote 8 is positive and statistically significant at 1 per cent level of significance. From the long-run equation of the cointegrated model, we infer that a 1 percentage point increase in the WALR leads to a 0.36 percentage point increase in the WALRF.

Generalized method of moments (GMM) model

Banks typically determine their interest rates on deposits and loans in an oligopolistic market setting (Santomero, 1984; Gambacorta, 2008). Banks are, therefore, not price-takers but price their loans depending on both macroeconomic and microeconomic (i.e., individual bank-specific) factors—the latter including the cost of funds, operating costs, asset quality, etc.

Since the main objective of this study is to assess the nature of transmission of the change in the monetary policy rate to the lending rates of individual banks under different interest rate benchmarks, we have introduced some control variables in the model. Apart from bank-specific characteristics, inflation and GDP are also included as control variables in the study.

Against this backdrop, we estimate the following equation to examine the monetary transmission process:

$$\begin{array}{l}\Delta {{{{walr}}}}_{i,t} = {{{\mathrm{\mu }}}}_i + \mathop {\sum}\nolimits_{j = 1}^n {\alpha _j\Delta walr_{i,t - j}} + \mathop {\sum}\nolimits_{j = 1}^n {\beta _j\Delta wacr_{t - j}} \\ \qquad\qquad\quad +\,\mathop {\sum}\nolimits_{j = 1}^n {\psi _j\pi _{t - j}} + \mathop {\sum}\nolimits_{j = 1}^n {\xi _j\Delta lnG_{i,t - j}} + \varepsilon _{i,t}\end{array}$$
(4)

and

$$\begin{array}{l}\Delta walr_{i,t} = \mu _i + \mathop {\sum}\nolimits_{j = 1}^n {\alpha _j\Delta walr_{i,t - j}} + \mathop {\sum}\nolimits_{j = 1}^n {\beta _j\Delta wacr_{t - j}}\\ \qquad\qquad\quad + \mathop {\sum}\nolimits_{j = 1}^n {\gamma _jX_{i,t - j}} \ast \Delta wacr_{t - j} + \lambda Z_{it - j}\\ \qquad\qquad\quad + \mathop {\sum}\nolimits_{j = 1}^n {\psi _j} \pi _{t - j} + \mathop {\sum}\nolimits_{j = 1}^n {\xi _j\Delta lnG_{i,t - j} + \varepsilon _{i,t}} \end{array}$$
(5)

where i = 1…. N; j = 1,...n; and t = 1 …. T; where N = number of banks; n = the number of lags in the model. walr = weighted average lending rate on fresh rupee loans sanctioned by banks. μi = fixed effects across banks. ∆walrit = change in walr of bank i between quarter t−1 and t. wacrt = weighted average call rate (proxy for policy rate) in quarter t. Xit = a set of bank-specific characteristics for bank i in quarter t that may impact pass-through. πt = inflation rate in quarter t. Zit = bank-specific characteristic. ∆lnGt = real GDP growth in quarter t over quarter (t−4).

Here Eq. (4) represents the benchmark model without inclusion of any bank-specific variable and Eq. (5) represents the model including bank-specific characteristics.

The selection of instruments is done based on diagnostic tests (viz., AR(1), AR(2) and Sargan). We have added real GDP growth and inflation in the model to control for the demand effects (Ehrmann et al., 2003; Gambacorta, 2004; Holton and D’Acri, 2018) Footnote 9 Since all the variables in the level are integrated into order one, we have taken the difference of the variables in our model (Appendix Table A). All the data were seasonally adjusted to take care of the seasonality aspect.

The above model would have been appropriate if wacr was the proxy for the repo rate for the entire period. However, during Q2 and Q3:2013–14, the marginal standing facility (MSF) rate had become the de facto policy rate supplanting the repo rateFootnote 10. This happened when responding to the hurried flight to safety by FIIs following the Fed announcement on the withdrawal of the stimulus, the RBI raised the MSF rate to defend the exchange rate even as the repo rate—a tool to indicate the monetary policy stance on domestic price stability—was left unchanged resulting in a spread of 300 bps between the MSF rate and the repo rate (from 100 bps earlier). As liquidity tightening measures were also taken, wacr shot up and remained aligned to the MSF (instead of the repo rate) during this period. Once normalcy was restored in the financial markets, the MSF rate was reduced; while simultaneously, the repo rate had to be raised to contain the inflationary pressures resulting in a peculiar situation of the wacr declining even as the repo rate was rising. In Eq. (4), therefore, we have introduced a dummy variable D1 for the two quarters (2013–14: Q3 to 2013–14: Q4) to capture the impact of taper tantrum on walr. We also introduce the dummy D2 for 2016–17: Q3 to 2017–18: Q1 to capture the demonetization effect on walr. After the introduction of the dummy variable, our model becomes:

$$\begin{array}{l}\Delta {\rm {walr}}_{i,t} = \mu _i + \mathop {\sum}\nolimits_{j = 1}^n {\alpha _j\Delta {\rm {walr}}_{i,t - j}} + \mathop {\sum}\nolimits_{j = 1}^n {\beta _j\Delta {\rm {wacr}}_{t - j}}\\ \qquad\qquad\quad+ \mathop {\sum}\nolimits_{j = 1}^n {\gamma _jX_{i,t - j}} \ast \Delta {\rm {wacr}}_{t - j} + \lambda Z_{it - j} + \mathop {\sum}\nolimits_{j = 1}^n {\psi _j\pi _{t - j}}\\ \qquad\qquad\quad + \mathop {\sum}\nolimits_{j = 1}^n {\xi _j\Delta lnG_{i,t - j} + D_1 + D_2 + \varepsilon _{i,t}} \end{array}$$
(6)

where, \(\left. {\begin{array}{*{20}{c}} {D_1 = 1} \\ {D_1 = 0} \end{array}} \right\}\begin{array}{*{20}{c}} {{{{\mathrm{for}}}}\,2013 - 14:\,{{{\mathrm{Q}}}}3\,{{{\mathrm{and}}}}\,2013 - 14:\,{{{\mathrm{Q}}}}4} \\ {{{{\mathrm{otherwise}}}}} \end{array}\)\(\left. {\begin{array}{*{20}{c}} {{{{D}}}_2 = 1} \\ {{{{D}}}_2 = 0} \end{array}} \right\}\begin{array}{*{20}{c}} {{{{\mathrm{for}}}}\,2016 - 17:\,Q3\,{{{\mathrm{to}}}}\,2017 - 18:\,{{{\mathrm{Q}}}}1} \\ {{{{\mathrm{otherwise}}}}} \end{array}\)

Consistent with the literatureFootnote 11, the bank-specific characteristics have been reparametrized in the following way:

$$X_{i,t} = \phi _{i,t} - \left[ {\mathop {\sum}\nolimits_{t = 1}^T {\frac{{\mathop {\sum}\nolimits_{i = 1}^N {\phi _{i,t}} }}{{N_t}}} } \right]/T$$
(7)

where Xi,t is the normalized bank-specific characteristic explained later. Here ϕi,t is the observation of ith bank in period t and Nt is the total number of banks in period t. Further, T is the total number of periods. That means each observation of a particular bank for period t is normalized with respect to the number of banks and the number of periods. In other words, each indicator/ bank-specific characteristic is normalized with respect to the averages across all the banks in the respective sample so that the sum over all observations becomes zero. Since the average of the interaction term between the monetary policy variable (∆wacrt-j) and Xi,t−j in Eq. (4) is zero for the average bank, the parameters βj can be directly interpreted as average monetary policy effect (Ehrmann et al., 2003; Gambacorta, 2004; Holton and D’Acri, 2018). Since there is heterogeneity in interest rate pass-through across the whole banking system, each variable is required to be normalized with respect to the average across all banks in each period of time. Following Gambacorta (2004), we have normalized the size indicator in respect of both—the mean over the whole sample period as well as for each single period—to remove the unwanted trends.Footnote 12

Dependent variable

To measure the impact of pass-through under different regimes, we use the q-o-q change in the weighted average lending rate (walr) on fresh rupee loans sanctioned by banks during the month as the dependent variable. Fresh rupee loans have been preferred as the dependent variable to outstanding rupee loans as the former is priced with reference to the prevailing benchmark, unlike outstanding rupee loans which have a sizeable share of loans priced to the earlier benchmark(s).

Independent variables

The objective of this study is to examine the relationship between a monetary policy indicator and lending rates. Instead of the policy repo rate, we have used the weighted average call money rate (wacr), which is the operating target of monetary policy and mimics the policy repo rate as the indicator of monetary policy stance in our model: the correlation coefficient between the two was found to be as high as 0.94 for the period under study (March 2013–September 2018). While the repo rate depicts a step-wise movement, the wacr fluctuates daily also reflecting the liquidity condition in the system and thus, better reflects the overall stance of monetary policy.

We include two important macro variables—CPI inflation and real GDP growth—as controls in our regression model. These variables capture the demand for credit and the risk of lending to certain markets. According to Holton, D’Acri (2018), there is no clear-cut direction of the effect of CPI inflation and real GDP growth on the interest rates. The relationship between each of the macro variables and wacr can be either negative or positive depending on the dominance of demand or risk. When growth is decelerating (and inflation declining) accompanied by declining credit demand, we may expect lending rates to fall. However, the slowdown in real economic activity (accompanied by a decline in inflation) may damage borrowers’ creditworthiness resulting in a rise in risk premia, thereby raising their cost of borrowing.

Bank-level variables

Pricing of loans extended by a bank depends on bank-specific characteristics. In each of the 9 models, we have used a unique bank-specific variable: term deposit rate, total asset size, liquidity, capital to risk-weighted assets ratio (CRAR), return on assets, non-performing assets, non-interest income, operating expenses and investments in securities approved for statutory liquidity ratio (SLR). The rationale behind choosing these variables is detailed below.

Bank deposits are one of the most important components of funding in India. In a cost-plus pricing structure, a direct relationship between the cost of funding and lending rates is expected. However, the deposit rate may not respond 1–1 to the change in monetary policy, impeding transmission to lending rates.Footnote 13

There is a contrasting view on the relationship between the size of bank assets and the lending rate. Maudos and De Guevara (2004), Angbazo (1997) found a negative relation between bank size and net interest margins (interest income minus interest expenditure). Holton and D’Acri (2018) reported that an increase in the size of the bank leads to a decrease in the overall pass-through of the money market rate. Sensharma and Ghosh (2004) found that there is a significant positive relationship between the size of Indian banks and NIM. Further, John et al. (2018) reported contrasting results for bank groups in India depending on the chosen time period.

Liquidity is considered as a barometer to measure the balance sheet condition of a bank; it also influences the degree of pass-through (Holton and D’Acri, 2018). Bluhm et al. (2014) found that banks with illiquid assets were subjected to shocks during crisis and compelled to deleverage. Gambacorta and Mistrulli (2004) showed that banks with liquid assets transmit more in response to monetary policy during an expansionary phase.

Capital to Risk Weighted Asset Ratio (CRAR) is the ratio of a bank’s capital in relation to its risk-weighted assets. Scheduled commercial banks in India are required to maintain a CRAR of 9 per centFootnote 14. Prudent banks may prefer to maintain additional capital over and above the regulatory requirements to meet unanticipated future requirements in an uncertain market environment, which could impede monetary transmission during the expansionary phase (Behera et al.). Since higher CRAR raises costs of intermediation, we expect that banks will pass on the higher costs to their lending rates.

In the case of return on assets (RoA), the relationship between the lending rates and RoA is not linear; it depends on the monetary policy cycle, the health of bank balance sheet and whether the bank is driven by the objective of maximization of profits or sales. For example, a bank with a stronger balance sheet and a higher RoA may be motivated to capture a higher market share irrespective of the policy cycle; hence, it may lower its lending rates faster vis-à-vis its competitors during an easing cycle, but may not increase its lending rates during tightening of monetary policy. Therefore, the relationship between the two is not clear-cut.

Regarding asset quality, a high degree of non-performing assets may prompt reduced pass-through. Exactly this is corroborated by the empirical literature (Holton, D’Acri (2018)). In the Indian case, John et al. (2018) found that deterioration of asset quality impacted monetary transmission.

Non-interest income (NII) is an increasingly important source of income for banks. Banks look at the totality of income—interest and non-interest—from a customer. Banks may aim at expanding their customer base by providing loans at lower rates of interest to those providing fee-based income to banks to maximize their total income (Dumicic and Ridzak, 2013; Maudos and Solis, 2009; Carbo and Rodriguez, 2007; John et al., 2018). We may, therefore, expect a negative relationship between the non-interest income and lending rates of banks.

Operating expenses are expenses in relation to the operations of a business on a daily basis. Various components such as salaries and pensions, administrative expenses, software costs, occupancy costs, etc., come under this head (John et al., 2018; Dumicic and Ridzak, 2013). There is an agreement in the literature that banks normally transmit the burden of operating costs to the customer. This is feasible in an oligopolistic setting. However, where banks are driven by considerations of social banking or where the majority owner itself has considerations other than profit maximization, a negative relationship between operating costs and lending rates could be observed. Further, when banks already earn a very high NIM, they may have the leeway to absorb the rise in operating costs and not pass it on to their customers. Even when a bank prices its loans off an internal benchmark, which has operating costs as one of its components, the bank can reduce the spread it charges over the benchmark to prevent the lending rate charged to the customer from rising to the full extent of the rise in operating costs. In view of this, we may not expect an unambiguous relationship between lending rates and operating expenses.

Another measure of liquidity is investments in approved securities to maintain the statutory liquidity ratio (SLR). Investments in approved securities over and above SLR can be used for availing liquidity.Footnote 15 An individual bank with more liquid assets may be able to offload the securities to fund credit growth while efficiently transmitting policy rate signals during an easing phase. However, when the overall SLR in the banking system is high and banks are unable to offload the securities to fund credit demand without booking losses, credit may get crowded out and transmission to lending rates impeded.

Estimation

For our estimation, we have used the system GMM dynamic panel data model developed by Arellano and Bover (1995). The rationale for choosing GMM/instrumental variables (IV) method for our empirical exercise is that it addresses various pitfalls associated with the least square-based inference methods when the model is dynamic, i.e., the dependent variable (in our case walr) is regressed on its past values (Bun Maurice and Sarafidis, 2015). In such a case, when lagged dependent variables are taken on the right-hand side of the equation, the problem of endogeneity emerges. Arellano and Bover’s (1995) method resolves this problem by including instrumental variables (IV) in the equation. Further, it also ensures the efficiency and consistency of the estimates as compared with the least squares-based inference methods, provided that the model is not subject to the serial correlation of order two and that the instruments used are valid (Gambacorta, 2008). While the validity of the instruments (over-identification of the model) is tested with the help of the Sargan test, serial correlations of the residuals are tested with the help of A-B serial autocorrelation (AR1 and AR2) test in residuals for the first order and second order.Footnote 16 The use of lagged values of dependent and explanatory variables as IVs is crucial to avoid endogeneity problems. For example, real GDP growth and consumer price index (CPI) inflation not only determine the loan demand, they determine the policy rate as well.

In our econometric exercise, we have chosen appropriate lags and dropped the insignificant lagged variables. The data set has been seasonally adjusted to remove the seasonality bias. Our analysis is based on all the domestic (public and private sector) banks. Study has been conducted by taking bank-specific data from individual banks. All the bank-specific data have been collected from the Reserve Bank. Remaining data have also been collected from RBI publications, such as the Handbook of Statistics on India, Statistical Tables relating to Banks in India and Database on the Indian Economy.

Results

A summary of statistics in respect of the base rate and MCLR regimes is presented in Table 5.Footnote 17Following are the highlights from the table: The size of the banks increased over time (in nominal terms), while the return on assets declined sharply. Non-performing assets of banks increased sharply following the asset quality review (AQR), which coincided with the introduction of the MCLR regime. Operating expenses remained unchanged throughout the period. Non-interest income was 10.4 per cent of total income in the base rate period and increased to 13.6 per cent in the MCLR regime. Surplus liquidity in the banking system was higher during the MCLR regime. CRAR changed marginally—increasing from 12.4 per cent during the base rate period to 12.6 per cent in the MCLR regime.

Table 5 Summary statistics.
Full size table

We have estimated Eq. (4) by using the system GMM as suggested by Arellano and Bover (1995). GMM is efficient when N (number of cross-sections—banks in our case) is large and T (time period) is small. Accordingly, it is appropriate to use GMM. We have used the test statistics in our results as diagnostic tests. The regression results are presented in Tables 6–9 for the whole period and the two sub-periods as discussed earlierFootnote 18.

Table 6 Benchmark regressions.
Full size table

It may be stated that although we included GDP growth in our model at the beginning, subsequently, we dropped the variable, since the inclusion of growth creates multicollinearity problem when dummy variables are introduced.

Following Ehrmann et al. (2003) and Gambacorta (2008), we discuss the estimated long-run coefficients only. Table 6 depicts the benchmark regressions results without the addition of any bank-specific characteristic as shown in Eq. (4). It is observed that a 1 percentage point change in the wacr leads to a 0.12 percentage point change in walrf in the same direction for the whole sample in the long run, while the figures are 0.12 percentage point and 0.21 percentage point, respectively, in case of base rate and MCLR regimes. Thus, the impact of monetary policy on walrf is more during MCLR regime in comparison to the base rate regime in benchmark regressions. The impact of inflation on lending rates is statistically significant and on expected lines.

Whole sample: 2012–13: Q4 to 2018–19: Q2

In this subsection, we introduce the nine bank-specific characteristics and the interaction term of the monetary policy indicator (wacr) with each of these characteristics separately to estimate Eq. (5) in Models 1–9 for the entire sample period. The introduction of the interaction terms is to estimate the influence of each bank-specific characteristic on the lending rate (walrf) for any change in the wacr. These variables have been re-parameterized such that γj in Eq. (5) can be interpreted as an average effect.

The results for the whole sample in Table 7 indicate that the long-run effect of wacr on lending rates is significantly different from zero in all the models. Further, the estimated long-run multipliers of wacr have the expected positive sign and are significantly different from zero in all models. The estimates imply that a 1 percentage point increase/decrease in the wacr leads to an increase/decrease in the walrf by 0.13–0.24 percentage point in the long run (Models 1–9). On the effects of inflation on lending rates, the long-run relationship is positive and statistically significant in all cases except one.

The coefficient of the interaction term between the term deposit rate (watdr) and wacr (Model 1) is significant and positively related, implying higher the deposit rate, the higher the lending rate, which is in line with our expectations. The coefficient of the interaction term between size and wacr (Model 2) is significant and positive. This implies that a larger-sized bank increases its lending rate when there is an increase in wacr and vice versa. In the case of CRAR, the relationship is significant and negative. This implies that banks with higher CRAR provide credit at a lower rate in response to the easing of monetary policy. In the case of operating expenses, the relationship is significant and positive. This indicates that banks with higher operating expenses provide credit at a higher lending interest rate irrespective of the stance of monetary policy. This is in line with our expectations. That means, higher operating expenses hinder the transmission mechanism of monetary policy during the easing phase. The interaction terms for the remaining five variables are observed to be statistically insignificant.

Table 7 Pass-through results for the whole sample (2012–13:Q4 to 2018–19:Q2).
Full size table

Base rate regime

In our sample, the base rate regime covers the period from 2012–13:Q4 to 2015–16:Q4. The results for this period reported in Table 8 show that 1 percentage point change in wacr leads to a change in the walrf in the range of 0.22–0.35 percentage points after one quarter (lagged coefficient). And the relationship is observed to be statistically significant in all the cases.

Regarding the long-run relationship, the effect of change in wacr on change in walrf is significantly different from zero in all 9 models. Further, the long-run coefficients of wacr have the positive sign as expected. The estimates indicate that a 1 percentage point increase in the monetary policy indicator (wacr) leads to an increase in the walrf of 0.11–0.19 percentage points in the long run. On the effects of macroeconomic variables (i.e. inflation), banks expectedly raise their lending rates in an inflationary situation.

The coefficients of the interaction terms between wacr and CRAR (Model 4), and NII (Model 7) are found to be statistically significant at conventional values. Thus, the results indicate that banks with higher CRAR charge lower lending rates, as observed in the whole sample period. The sign of the CRAR is against our expectations. Higher capital motivates banks to decrease their lending rates, thereby facilitating the transmission process during the easing cycle. In the case of NII, monetary transmission is hindered by the high non-interest income during the easing cycle: this is because when the banks’ non-interest income is high, they appear to be reluctant to reduce their lending rates in response to the reduction of the policy rate. The coefficients of the interaction terms between the remaining seven bank-specific indicators and wacr are statistically insignificant at the conventional level.

Table 8 Pass-through results for base rate regime (2012–13: Q4 to 2015–16: Q4).
Full size table

MCLR regime

The MCLR regime covers the period from 2016–17:Q1 to 2018–19:Q2. The results including their statistical significance vary across models 1-9 (Table 9). The long-run coefficients of wacr have the positive sign as per expectations and are statistically significant in all the models. The estimates show that a 1 percentage point increase/decrease in the monetary policy indicator (wacr) leads to an increase/decrease in the lending rate (walrf) by 0.26–0.47 percentage point in the long run, which is more than that during the base rate regime for each of the models. This is notwithstanding the worsening in the bank balance sheets during the MCLR regime when one would have expected banks to load risk premia onto their lending rates impeding transmission; besides, monetary policy was undergoing an easing cycle (except for 2018–19:Q2) when the speed of transmission to lending rates is usually lower (Singh, 2011). Since the demonetisation dummy was not significant, we could not conclude that transmission during the MCLR regime was facilitated by demonetisation, which was independent of monetary policy.

As in the case of the whole sample period, in some cases, the long-run coefficients are lower than the sum of the lagged coefficients for monetary policy during the MCLR regime.Footnote 19 On the effects of macroeconomic variables, an increase in inflation expectedly leads to an increase in lending rates of banks, similar to our findings in the case of the whole sample period and base rate regime.

The coefficient of the interaction terms with the monetary policy indicator (wacr) are found to be statistically significant in the case of four models, viz., deposit rate (Model 1), size (Model 2), liquidity (Model 3) and NPA (Model 6). Except for the sign of the coefficient for NPAs, the other three are on the expected lines. In other models, the coefficients of the interaction terms are insignificant.

From the results, the following inference can be drawn. The coefficient of the interaction term between the deposit rate (watdr) and wacr (Model 1) is statistically significant and positively related, implying higher the deposit rate, the higher the lending rate, which is in line with our expectations. Further, as in the whole sample period, the higher the size higher is the lending rate. That means, during the easing cycle transmission process is hampered since banks do not reduce their lending rates in response to a reduction in the policy rate. Second, due to the existence of excess liquidity in the banking system, walr declined more in response to the reduction in the policy rate. Lastly, unlike the Base Rate regime, we observe a negative significant relationship between NPA and wacr in the MCLR regime; hence higher the NPA, the lower the walr, which is against our expectations, but consistent with the findings of John et al. (2018). If banks were to price in risk premia to their lending rates, one would have expected a positive correlation; the negative relationship could perhaps be attributed to the risk-averse strategy adopted by many banks following the introduction of asset quality review (AQR) that resulted in a spurt in NPAs; several banks were also barred from risky lending under the prompt corrective action (PCA) framework. As a result, banks altered their lending strategy to focus more on collateralized retail loans like housing and vehicle financing where lending rates are typically lower than in other sectors because of lower default risk and greater competition with many non-bank players in the credit market.

Table 9 Pass-through results for MCLR regime (2016–17: Q1 to 2018–19: Q2).
Full size table

Conclusion

In a bank-dependent economy, efficient monetary policy transmission to banks’ lending rates is a crucial conduit for the successful implementation of monetary policy. However, transmission from policy rate to lending rates has remained partial. To mitigate this problem and ensure customer protection, the Reserve Bank has changed the lending rate benchmark over time since the deregulation of lending rates in October 1994. Against this backdrop, this study has attempted to compare the degree of transmission to the walr charged by individual banks between the two interest rate benchmark regimes—base rate and MCLR—following a change in the monetary policy rate. In our study, we have considered three different time periods, viz., whole sample period (i.e., Q4:2012–13 to Q2:2018–19) and its two sub-periods—base rate regime (i.e., Q4:2012–13 to Q4:2015–16) and MCLR regime (Q1:2016–17 to Q2:2018–19). We have used the GMM model suggested by Arellano and Bover (1995) to estimate the equations for the whole sample and the two sub-periods. For each time period, we have estimated the benchmark regression followed by 9 models - each incorporating a unique bank-specific characteristic.

In order to provide a reference for drawing a comparison with the individual models in a GMM framework, we have run a VEC model by taking the aggregate level time series data, since data were found to be I(1) and also co-integrated. From the VECM results, we find that there is a change in the weighted average lending rate on fresh rupee loans (walrf) of domestic banks of 0.36 percentage points due to a 1 percentage point shock in the wacr—the proxy for the monetary policy rate.

After doing the VECM analysis, we have focused on the bank-level data in a panel data set up. Thus, from the GMM results, we observed that for the whole sample period, the long-run coefficients of wacr on walr are significant in all the models and have the expected positive sign. It is estimated that a 1 percentage point increase in the weighted average wacr leads to an increase in the walrf ranging from 0.13 to 0.24 percentage points, in the long run, depending on which one of the nine different models is chosen. An increase/decrease in inflation expectedly leads to an increase/decrease in lending rate; our findings are similar in this regard for the two sub-periods. Among the bank-specific characteristics, the coefficients of the interaction term with wacr were found to be statistically significant at conventional values for watdr, size, CRAR and operating expenses. Thus, a rise in the watdr (or size) of banks leads to an increase in the lending rate. This means that when the wacr increases, walr increases because of an increase in the watdr (or size). During the easing phase, on the other hand, walr declines in response to a decline in the wacr; however, walr may not decline proportionately because of higher watdr, which hinders effective monetary transmission. Monetary transmission is, therefore, stronger during the tightening period as compared with the easing phase. In the case of the model with size, when the call rate declines, the lending rate declines; but due to the bigger size of the bank, the lending rate does not decline commensurate with the decline in the call rate. Hence, monetary transmission is adversely affected during the easing cycle of monetary policy. CRAR is the 3rd variable for which we have found a statistically significant and negative relationship between wacr and CRAR. This implies that banks with higher CRAR lower their lending rate. Hence, bank recapitalization would facilitate transmission during the easing phase as higher capital enables banks to overcome statutory restrictions and increase lending by lowering their lending rates. Operating cost is the fourth variable for which a positive and significant relationship was found, implying that a rise in operating costs hinders transmission during the easing phase.

For the 1st sub-period, a 1 percentage point increase/decrease in the wacr led to an increase/decrease in the walrf of 0.22–0.35 percentage points for a typical bank in the long run. The coefficients of the interaction term with wacr and two bank-specific variables, viz. CRAR and NII were statistically significant, but not in the same direction. In the case of CRAR, we have found a significant negative relationship as found in the whole sample period. In the case of the model incorporating NII, although the walrf declines in response to a decline in wacr, the extent of the decline gets partly offset when the non-interest income is large. That means the transmission is impeded during the easing cycle due to increased non-interest income.

For the 2nd sub-period, i.e., during the MCLR regime, a 1 percentage point increase in the wacr leads to an increase in walrf of 0.26–0.47 percentage points in the long run. For all models, transmission is more during the MCLR regime than the base rate regime. The interaction term between wacr and four of the 9 bank-specific variables – deposit rate, size, liquidity, and NPA—is significant. The signs of the coefficients are on the expected lines for three of the variables, viz., deposit rate, size, and liquidity. In the case of the deposit rate, it is observed that the higher the deposit rate, the higher the lending rate, as observed in the case of both the whole sample period. Theoretically, it is consistent since a rise in the deposit rate may lead to a concomitant increase in the cost of funds. Therefore, due to the rise in the cost of funds, banks will pass on this cost to the lending rate, which we observe from our findings. In the case of size, it is observed that the higher the size, the higher the lending rate, as observed in the case of both whole sample periods. Liquidity has a negative and significant relationship with the lending rate, implying that the walrf cannot increase commensurately with the rise in wacr in a tightening phase in the presence of higher availability of surplus liquidity in the system. Also, excess liquidity acts as a facilitator of monetary transmission during the easing phase. A major finding of the study corroborating other studies is that during the MCLR regime, banks could not price credit to reflect the sharply rising NPAs in their balance sheet following the AQR. Faced with higher NPAs prompting the tightening of regulatory norms by RBI and risk-averse strategy adopted by banks, banks were unable to increase their walrf on the aggregate lending portfolio even as credit growth decelerated sharply. Since banks did not enjoy manoeuvrability in the pricing of loans when NPAs rose sharply, the transmission did not get obstructed during the easing regime.

To conclude, irrespective of the model chosen, transmission is higher during the MCLR regime than during the Base Rate regime. Furthermore, the syncing of liquidity management with the monetary policy stance, and the introduction of the flexible inflation targeting (FIT) framework coupled with the deceleration in economic activity reducing credit demand could be contributory factors for better transmission during the MCLR regime. Nevertheless, transmission during the MCLR regime was far from satisfactory necessitating the introduction of external benchmark-based pricing of loans for personal loans and micro & small enterprises, effective October 1, 2019, and for medium enterprises since April 1, 2020. The progressive shift from the various internal benchmark-based pricing of loans to the external benchmark augurs well for monetary transmission going forward as the latest available sectoral data on transmission testify.