The financial cycle in India

(1)

The Financial Cycle in India

Name: Paulien Janse Student Number: 6297056 Program: Master Economics Specialization: International Economics and Globalization Department: Faculty of Economics and Business

Date: 2 August 2015 1st_{supervisor: Dr. K. Mavromatis}

(2)

2 Statement of Originality

This document is written by student Paulien Janse who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in

creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

ABSTRACT

This research examines whether a financial cycle is present in India and whether it influenced the Indian economy during the recent global financial crisis. Monthly data were collected for seventeen financial variables from the Indian economy from 1994 – 2012. Factor analysis was performed to see which variables compose the financial cycle. It was examined how the cycle looks like and whether it had a significant impact on the financial variables during crisis-periods. The outcome indicates that there is a financial cycle present in India and that it does have a significant impact on fourteen financial variables during some periods of the recent financial crisis. Furthermore, the Indian financial cycle has an amplitude that is twenty-three times smaller comparing to the amplitude of the financial cycle of the United States, while it has a much higher frequency. This conclusion corresponds to a lower financial liberalization in India compared to the United States, but the Indian economy has still become more financially liberalized and globalized over time.

(4)

4

1. Introduction

Different economists have argued that financial crises, asset price bubbles and credit bubbles are not financial events that are triggered by some special occurrences (for example how the collapse of Lehman Brothers could have triggered the recent financial crisis) but rather that a pattern exists in the behavior of financial variables (Drehmann et al., 2012; Borio, 2012; Claessens et al., 2011). The pattern in the behavior of financial variables can be called the financial cycle (Claessens et al., 2012). Although the idea of the financial cycle was established already in the 1910s by von Mises (1981), revived attention has been given to the concept in the 1990s. The reason for revived attention in the 1990s were some important financial events: the collapse in the asset market in Japan in the early 1990s, the financial crisis in emerging markets in Asia in the mid-1990s, the boom in advanced economies in equity markets at the end of the 1990s, and especially the recent global financial crisis emerging in 2007 (Claessens et al., 2011). What these events have in common is that strong credit growth corresponds to an upturn in the financial markets and the real economy but at some point in time the upturn changes into a downturn, and causes an unforeseen crisis (Lamont, 2008). The implications of when the financial cycle is present are quite substantial. When policymakers can examine the behavior of the financial cycle and make predictions about its future behavior, policymakers could mitigate the effects or even anticipate future crises.

Substantial research has been done on the financial cycle; not only for OECD countries (Claessens et al. 2011) but also for emerging market economies (Nier et al., 2014). However, the specific case of India has not yet been studied. The Indian economy was exposed very limitedly to the United States housing market, yet the recent financial crisis was felt in India (Sinha, 2010). This raises the question how India was hit. Accordingly, the question arises whether there is an Indian financial cycle present and whether it influenced the recent global financial crisis in India. For example, the Indian central bank has both increased and decreased the interest rate in the past few years (Sinha, 2010). This leads to the research question: “How does the Indian financial cycle look like and did it influence the Indian economy in the recent global financial crisis?”. Therewith, this thesis takes a first step in the empirical process of examining whether the financial cycle is present in India.

To examine the presence of a financial cycle in India, this thesis looks at data for multiple financial indicators such as exchange reserves, money supply, interest rates, stock exchange, net domestic credit and inflation. First, a factor analysis is performed to examine which

(7)

7

variables capture the most common variance in the dataset and can be used to compose the Indian financial cycle. Then, an autoregressive model is used where the financial cycle is regressed on itself to examine how the cycle looks like. Also, the cycle and its lags are regressed on all financial variables from the dataset using ordinary least squares (OLS) to analyze the effect of the cycle on those variables. The same regression is used in order to see whether the cycle and its lags can explain movement in the financial variables during the recent financial crisis. Section 3 discusses the dataset and methodology in more detail.

The financial cycle is expected to be present in India and to have a significant impact on the Indian economy because of increasing financial liberalization and increasing globalization, despite the low involvement of India in the United States housing market (Sinha, 2010).

The main conclusion is that a financial cycle is indeed present in India, and that it significantly affects the movements in fourteen financial variables during some periods of the recent financial crisis. Also, the Indian financial cycle has an amplitude that is twenty-three times smaller than the amplitude of the financial cycle in the United States, but the frequency is much higher. The latter indicates that the Indian economy is less financially integrated and globalized compared to the American economy, but India becomes more financially integrated and globalized over time.

The thesis is structured as follows. Section 2 explains the financial cycle mechanism and what variables are key components in the movement of the cycle. Section 3 discusses the current stance of literature on the financial cycle and on the Indian economy. Section 4 addresses the composition of the dataset, the methodology and the empirical model used in the data analysis. Section 4 also explains how the cycle was constructed using factor analysis. Section 5 presents and analyses the outcomes of the regressions, and discusses the validity of the results. The thesis is concluded by section 6 where a short summary is provided and the research question formulated in the introduction is answered.

(8)

8

2. Financial Cycle Mechanism

The Austrian Business Cycle Theory (ABCT) states that economic fluctuations are caused by the creation of credit by banks, due to a lowering of the interest rate by the central bank (von Mises, 1981). This view has a large resemblance with the financial cycle. The financial cycle is identified by upturns and downturns in financial variables such as credit, house and equity prices, which are financial variables influenced by the interest rate set by the central bank (Claessens et al., 2011). More details on the similarities between the ABCT and the financial cycle follow in section 3. Section 2 lays the focus on explaining the mechanism of the financial cycle.

2.1 Theory of the Trade Cycle

The theory of the financial cycle stems from an idea originated in the 1910s by Ludwig von Mises, and is elaborately described in his essay The “Austrian” Theory of the Trade Cycle (Ebeling, 1996). According to von Mises, the mechanism of the financial cycle is described as follows. When the central bank issues bank notes, it expands credit beyond its fundamentals. That is, credit is not backed up by the assets on the balance sheet of the central bank. The interest rate decreases to a lower level than without intervention, thus lower than its fundamental value. Hence, the information reflected in the value of the interest rate does no longer reflect its true value. Because of the lower rate, economic activity is encouraged. Before the lowering of the interest rate, some firms and investment projects were not profitable and were not initiated. However, when the interest rate is lowered, those projects can be profitable and therefore can be undertaken. Households and firms increase lending, which increases the demand for assets, raising the asset price and in turn increasing the expansion of credit. Furthermore, when more projects are undertaken, the demand for production factors increases, which raises wage and capital rent, leading to a rise in prices of consumption goods. When banks continue expanding credit, wages and prices will rise accordingly. The total availability of production factors has not changed in the short term. Some existing production factors have been taken from existing firms and projects and are used in new firms and projects. The above reasoning is described as the financial boom.

The increase in credit cannot continue indefinitely. When the public starts to realize that the boom and rising prices will not end any time soon, a panic starts: the public does no longer want to hold money and tries to buy goods. As a consequence, commodity prices rise even faster, with foreign currencies rising accordingly, whereas the price of the domestic currency falls massively or even completely collapses. The larger the boom, the larger will be the collapse in prices.

(9)

9

When banks are able to stop the expansion of credit at an earlier stage, they can prevent the currency from collapsing and therefore put a ‘brake’ on the boom. In this case, the effect on economic activity will be smaller than when the bubble completely bursts and prices fully collapse.

During the credit expansion, banks raise the interest rate again such that when the bubble bursts, the interest is higher than in the initial situation. Even the higher interest rate is not sufficient to put a stop to the boom. When the economy starts to contract, credit is tightened making it harder for firms and households to obtain credit. Also, it will take some time for the capital that was allocated to the firms that went bankrupt to revert back to producing firms. Increasing the money supply and decreasing the interest rate only provides a temporary stimulus for the economy. In the end, the economy is worse off than before the boom, with stagnated economic activity as a result (Ebeling, 1996).

For clarity, the figure below was created displaying the steps in the mechanism of the financial cycle. Central Bank Increases Money Supply Interest Rate Decreases More Investments Asset Prices and Goods Prices Increase Expansion of Credit Financial Boom

Start of

Panic

Demand

for Goods

Increases

Foreign

Currencies

Rise

Price

Domestic

Currency

Collapses

Financial

Bust

(10)

10

3. Literature Review

This section gives a review of the literature on the Austrian Business Cycle Theory and the financial cycle, and discusses similarities amongst the two. Furthermore, more recent research by the International Monetary Fund (IMF) and the Bank of International Settlement (BIS) is discussed. Also, a summary is given on the Indian economy before and during the recent financial crisis.

3.1 Resemblance Austrian Business Cycle Theory and Financial Cycle

Some authors, including Tempelman (2010), argue that the reasoning by the ABCT mentioned in section 2 has a large explanatory power to clarify not only the dot-com bubble early 2000s but also the recent global financial crisis. A large increase in productivity took place in the United States in the mid-1990s. Subsequently, the Federal Reserve employed a lenient monetary policy with no increase in the interest rate,which led to lower non-borrowed reserves by banks. Loans extended to households and firms increased. However, the amount of loans was larger than the amount that could be involved in GDP transactions, resulting in an increasing demand for shares in financial assets. As a consequence, an investment bubble emerged, better known as the dot-com bubble. It can be argued that when the Fed had employed a more restrictive policy, the investment boom would not have been so large, the end of the boom not so severe, and the misallocation of capital not so great. Research from different scientific analyses, such as Taylor (2007) and Jarocinski and Smets (2008), states that the Fed also pursued too loose monetary policy after the recession in 2001. Again, the Federal Reserve did not increase the interest rate where it could have done so. This lead to excess liquidity on the credit market. This excess liquidity led to an increase in asset prices and subsequently caused the housing boom in the United States. In short, the ABCT resembles very similar to the financial cycle.

According to Tempelman (2010), the key element in the recent global financial crisis is that nonprofessional investors started to copy behavior from professionals. Nonprofessional investors started to join the asset market because of the low cost of funding. However, they did not have the same skills to value an asset according to its fundamental value, and consequently bid up the price. This led to an unsustainable financial boom.

3.2 Duration and Volatility of the Financial Cycle

After the onset of the recent financial crisis, the IMF and the BIS have done considerable research on the financial cycle. According to Claessens et al. (2011) (performing research for the IMF) the financial cycle consists of equity, housing price and credit cycles, with short

(11)

11

booms and long busts. They find that financial cycles are more frequent from 1960 -1986 (pre-globalization period) than from 1986 - 2011. Claessens et al. also found that, on average, a recession has a duration of 4 quarters. Furthermore, the volatility and duration of equity and house prices are larger than those of credit cycles. Over time, especially equity cycles became shorter.

Furthermore, Drehmann et al. (2012) (performing research for the BIS) model the financial cycle on the medium term, and they find that the cycle is best characterized by a joint movement of credit and property prices. Equity prices are excluded from the analysis since their short run volatility is much higher compared to credit and property prices. Drehmann et al. (2012) also distinguish two time periods, where the year 1998 is the benchmark. The reason is that financial cycles that peaked after that year have a much longer duration: nearly 20 years compared to 11 years, as can be seen in figure 2 in appendix A.3. The duration of the financial cycle depends on the degree of financial liberalization in the economy, which can intensify the cycle. Furthermore, financial liberalization has increased the duration and intensity of the financial cycle from the mid-1980s.

Borio (2012) (performing research for the BIS) builds his research further on research by Drehmann et al. (2012) and finds that financial booms and busts are mainly moved by how economic agents value risk and their corresponding actions, combined with financing constraints. Those actions can magnify economic fluctuations and cause financial distress and economic disallocation. Financial cycle peaks are highly likely to be followed by a banking crisis. The most important financial indicators of financial crises are the deviations of credit-to-GDP ratio and asset prices from their historical norms. Borio (2012) extends the research by Drehmann et al. (2012) by arguing that the shape of the financial cycle depends on the policy regime, made up by not only the financial regime but also the monetary and real-economy regime. Financial liberalization enables more free interplay between financial variables. A tight inflation scheme enables monetary policy to be less expansionary during a financial boom. Globalization of the real economy nourishes financial booms. For example, Borio (2012) found that after some former communist countries entered into the global trading system, their financial cycle became larger and more prolonged.

3.3 The Indian Economy

According to the Worldbank report from 1991, India had experienced years of fiscal imbalances and a weakened economy. There was a serious need for the unsustainable fiscal and current account deficits to be reduced. Despite attempts to decrease the fiscal deficit and the introduction of an export-import policy to improve the trade balance, the balance of

(12)

12

payment came under severe pressure because of a slowing export growth to pegged currency areas and issues with commercial borrowing. The country also had to deal with internal conflicts and political instability, which decreased confidence of outside investors and led to very tight external liquidity. Hence, the Indian economy was on the edge of a liquidity crisis. Then, the oil shock in 1990 in the Persian Gulf accelerated the crisis in India. Rising oil prices combined with a depleting stock of foreign exchange reserves forced the country in 1991 to introduce structural reforms, resulting in a bailout loan from the IMF, a devaluation of the rupee of 43% and opening up to international trade (Worldbank, 1991).

Since 1991, India has been taking reforms to stabilize its economy, through expenditure reduction, tax reform, trade reform, industrial policy reform, financial sector reform, and public enterprise reform (World Bank, 1991). In figure 3 it can be seen GDP growth increased after the government started to implement those large reforms in 1991.

As written by Sinha (2010), after the collapse of the Lehman Brothers in September 2008, India experienced reversed market expectations and reversed capital inflows. These reversals had two simultaneous effects. First, it induced an appreciation of the Indian rupee and an increase in the foreign exchange rate market volatility. Combined with reduced access to international financial markets, this led to pressure on the dollar liquidity in the domestic market. Even though the banking system was stable, some financial institutions were pressured. To ensure liquidity, the Reserve Bank of India (RBI) provided more liquidity both for domestic and foreign currencies, and the government implemented fiscal stimuli. Second, these reversals caused asset prices to decrease and equity holdings to be sold, and even more removals of funds from the Indian financial markets. As a result, the RBI increased the intensity of its interventions to keep ensuring liquidity, which decreased the interest rate. In September 2009 however, inflation became increasingly pressuring because of the large increase in the money supply in the year before. To lessen the inflationary pressures, the Reserve Bank pursued countercyclical measures by demanding higher reserve requirements,

-8 -6 -4 -2 0 2 4 6 8 10 12 1961 1963 1965 1967 1969 1970 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013

GDP Growth India

Figure 3

Source Data: Worldbank

(13)

13

which resulted in a higher interest rate. In general, monetary and fiscal policies have been used by the Indian Reserve Bank quite limitedly, since liquidity conditions were relatively comfortable (Sinha, 2010).

In short, there was no asset price bubble in India, hence no excessive credit growth in the asset market. As stated in section 2, in the financial cycle, a lowering of the interest rate is viewed as the initiator of the boom. The RBI did lower the interest rate but as a mere consequence of the financial cycle boom, not as the initiator of the boom.

According to Sinha (2010) and Subbarao (2009), the peculiarity about the recent financial crisis in India is that the Indian banking system had no direct exposure to the troubled American subprime mortgages, contrary to many other countries. There were relatively little foreign banks present in the Indian financial markets and the central bank pursued a prudential policy. In addition, economic growth in the period before the crisis stemmed mainly from domestic consumption and investment. However, India had a deep financial integration and trade globalization, where external funding had increased considerably in the years before the crisis. Therefore, from the literature it can be concluded that the Indian economy was hit by the crisis not because of an unhealthy banking system but because of its increasing financial integration and trade globalization.

(14)

14

4. Methodology

This section describes the model and methodology this research uses, how the dataset is constructed and how the financial cycle is constructed using factor analysis.

4.1 Model

4.1.1 Research Method

According to Breitung and Eickmeier (2005), dynamic factor models are used in monetary policy and international business cycles, where the cycles are explained by a common dynamic factor. They argue that dynamic factor models have two main advantages over other methods for modelling cycles. First, measurement error and local shocks reflected by idiosyncratic shocks can be removed. Second, factor models are less restrictive in their assumptions compared to structural models (Breitung and Eickmeier, 2005). Therefore, this thesis uses a dynamic factor model to determine the financial cycle. Factor models are not only used to model cycles but often also used to predict future values (Stock & Watson (2010), Bai and Ng (2008), Breitung and Eickmeier (2005)). However, forecasting is not the focus of this research and therefore will not be described.

Significant research has been done about dynamic factor models, where multiple dynamic factors are found. Bai and Ng (2008) use panel data to find multiple factors. Breitung and Eickmeier (2005) also use panel data and a multiple factor model, applied in monetary policy. Phillips et al. (2007) focus on a single factor model using panel data, where they allow the common factor to have time-varying loadings. Stock and Watson (2010) use time-series data to find multiple factors. Since this thesis uses time-series data, it cannot use the model by Philips et al., Bai and Ng and Breitung and Eickmeier. Therefore, the methodology is built on the model by Stock and Watson (2010).

4.1.2 Methodology of Stock and Watson (2010)

According to Stock and Watson (2010), the methodology for a multiple factor model using time-series data is described as follows.

Some latent dynamic factors 𝑓𝑡 are the driving forces of the comovement of a vector of the

observed variables 𝑋𝑡. These observed variables are also influenced by idiosyncratic

disturbances, which come from measurement error and variable-specific effects. The latent dynamic factors follow a vector autoregressive (VAR) structure. The above is described in equations as follows.

(15)

15

𝑓𝑡 = 𝛹(𝐿) 𝑓𝑡−1+ 𝜂𝑡 (2)

Where 𝑋𝑡 = (𝑥1𝑡, 𝑥2𝑡, … , 𝑥𝑛𝑡), 𝑒𝑡 = (𝑒1𝑡, 𝑒2𝑡 , … , 𝑒𝑛𝑡), 𝑓𝑡 = (𝑓1𝑡 , 𝑓2𝑡, … , 𝑓𝑞𝑡), and

𝜂𝑡= (𝜂1𝑡 , 𝜂2𝑡, … , 𝜂𝑞𝑡), such that there are 𝑛 series and 𝑞 dynamic factors. 𝐿 indicates a lag,

such that 𝜆(𝐿) and 𝛹(𝐿) are respectively 𝑛 ∗ 𝑞 and 𝑞 ∗ 𝑞 series. 𝜆𝑖(𝐿) is the dynamic factor

loading for the ith_series,_𝑥

𝑖𝑡, hence 𝜆𝑖(𝐿)𝑓𝑡 is the common component of the ith series. The

observed variables in equation (1) are assumed to be stationary, which is ensured by taking the first difference of the logarithm of the variables. In this way, unit roots and trends are removed, and changes in the level of the time series are removed.

The idiosyncratic disturbances are assumed to be uncorrelated with the error in the movement of the common factor, so 𝐸(𝑒𝑡, 𝜂𝑡−𝑘) = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑘. Also, the idiosyncratic disturbances in

equation (1) are allowed to follow an autoregressive structure. Furthermore, the dynamic component of this model requires the idiosyncratic disturbances to be mutually uncorrelated also over time, 𝐸(𝑒𝑖𝑡, 𝑒𝑗𝑠) = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑠 𝑖𝑓 𝑖 ≠ 𝑗. Hence, the error terms of the observed

variables are not allowed to cross-correlate.

In this thesis the model is as follows.

𝑥1,𝑡= 𝜆1,1𝑓𝑡+ 𝜆1,2𝑓𝑡−1+ ⋯ + 𝜆1,𝑚+1𝑓𝑡−𝑚+ 𝑒1,𝑡 (3) 𝑥2,𝑡= 𝜆2,1𝑓𝑡+ 𝜆2,2𝑓𝑡−1+ ⋯ + 𝜆2,𝑚+1𝑓𝑡−𝑚+ 𝑒2,𝑡 (4) ⋮ 𝑥22,𝑡= 𝜆22,1𝑓𝑡+ 𝜆22,2𝑓𝑡−1+ ⋯ + 𝜆22,𝑚+1𝑓𝑡−𝑚+ 𝑒22,𝑡 (5) 𝑒1,𝑡= 𝛽1,1𝑒1,𝑡−1+ 𝛽1,2𝑒1,𝑡−2+ … + 𝛽1,𝑟𝑒1,𝑡−𝑟+ 𝜉1,𝑡 (6) 𝑒2,𝑡= 𝛽2,1𝑒2,𝑡−1+ 𝛽2,2𝑒2,𝑡−2+ … + 𝛽2,𝑟𝑒2,𝑡−𝑟+ 𝜉2,𝑡 (7) ⋮ 𝑒22,𝑡= 𝛽22,1𝑒22,𝑡−1+ 𝛽22,2𝑒22,𝑡−2+ … + 𝛽22,𝑟𝑒22,𝑡−𝑟+ 𝜉22,𝑡 (8) 𝑓𝑡 = 𝜓1𝑓𝑡−1+ 𝜓2𝑓𝑡−2+ ⋯ + 𝜓𝑝𝑓𝑡−𝑝+ 𝜂1,𝑡 (9)

(16)

16

4.1.3 First Generation Time-Domain Maximum Likelihood

Stock and Watson (2010) propose three generations of methods to estimate the dynamic factor model: first generation time-domain maximum likelihood via the Kalman filter, second generation nonparametric averaging methods and third generation hybrid principal components and state space methods. The software Stata has the option to estimate the dynamic factor using the Kalman filter. However, the model in Stata would estimate how the dynamic factor is composed of the observed variables (𝑓𝑡 is explained by 𝑋𝑡, so the financial

cycle is explained by the behavior of the financial variables), while the goal of this thesis is to examine how the cycle influences the observed variables (𝑋𝑡 is explained by 𝑓𝑡, so the financial

variables are explained by the behavior of the financial cycle) using the model by Stock and Watson (2010). Therefore, factor analysis is performed to see what variables capture the largest unobserved variance of the observed variables and therefore compose the dynamic factor. Section 4.2 first describes the dataset where after section 4.3 describes in detail the rest of the methodology concerning factor analysis and regressions.

4.2 Variables

An often-used approach to measure the financial cycle includes variables such as credit, housing prices and equity prices. Claessens et al. (2011), Drehmann et al. (2012) and Borio (2012), performing research for the International Monetary Fund (IMF), and the latter two for the Bank of International Settlement (BIS), used such parameters. The approach used by the IMF and BIS could also have been used in this paper, where the financial cycle exhibits some similarities to the business cycle (Borio, 2012). However, this research takes a different approach due to the lack of data available for housing prices in India.

Data have been collected on 22 financial variables from 1994-2012 from the Indian economy. When possible, data were collected on a monthly basis. However, data on some important variables were available only on a yearly basis. Therefore, these data are linearly interpolated to monthly data. The first difference of the logarithm is taken for all financial variables, where after the variables are standardized. Standardization is done by subtracting the mean of the variable and after dividing by its standard deviation (Stock & Watson, 2012). More details are provided in section 5.1.

Data for the exchange rate, M1, M2, Bombay stock exchange, 3-month Treasury bill rate, prime lending rate, discount rate, foreign currency reserves and car sales volume were collected on a monthly basis from Datastream. Data for capital account, consumption (private and government), consumer price index, inflation, net domestic credit, net foreign assets, unemployment, real interest rate, claims on the private sector, claims on the government and

(17)

17

foreign direct investment (net inflow) were collected on a yearly basis from the Worldbank World Development Indicators. Yearly data on the real effective exchange rate were collected from the Bank of International Settlement. Yearly data on current account were collected on a yearly basis from the International Monetary Fund World Economic Outlook Database. Besides consumption, the car sales volume is also included in the analysis since it indicates a change in purchases of large durables, which reflects confidence in the economy. Apart from consumption and car sales, some more variables are included in the analysis that are approximately the same: M1 and M2 and consumer price index and inflation. The reason for this is to enable the comparison of which variable has the largest common factor with the other financial variables. During the analysis, the decision will be made which variables are best to compose the Indian financial cycle.

In this thesis, 𝑋𝑡 = (𝑥1𝑡, 𝑥2𝑡, … , 𝑥𝑛𝑡) is a vector of observed financial variables described in this

section that compose the dataset (inflation, unemployment, exchange rate etcetera). 𝑓𝑡 =

(𝑓1𝑡 , 𝑓2𝑡, … , 𝑓𝑞𝑡) is not a vector in this thesis because the goal is to find only one dynamic factor,

the financial cycle. Therefore there is only one equation for the financial cycle, which is given by equation (9). As can be seen from equations (3)-(9), the observed financial variables (𝑥1𝑡, 𝑥2𝑡, … , 𝑥22𝑡) are explained by the financial cycle and its past lags, the error terms of the

observed financial variables (𝑒1𝑡, 𝑒2𝑡 , … , 𝑒22𝑡) are explained by their own past lags, so they are

allowed to be autocorrelated, and the financial cycle (𝑓𝑡) is explained by its own past lags.

4.3 Construction of the Financial Cycle

As mentioned in section 4.2, data have been collected for many financial variables from the Indian economy. However, data for the Indian financial cycle do not exist. Therefore, first the cycle needs to be constructed. Section 4.3 explains how this is done, and further sections describe how this translates into the model by Stock and Watson (2010).

4.3.1 Factor Analysis versus Principal Component Analysis

Engel et al. (2012) and Bernanke et al. (2004) have both built a model where a dynamic factor is used in a similar way as in this model. As they notice, a dynamic factor is autoregressive and unobserved. Engel et al. (2012) estimate the dynamic factor by factor analysis, where the co-movements of the observed variables are captured by the factor. Bernanke et al. (2004) estimate the dynamic factor using principal component analysis. The main difference between factor analysis and principal component analysis is that factor analysis is meant to identify the underlying structure of the observed variables and only the shared variance is analyzed, whereas principal component analysis tries to find geometric abstractions and all of the

(18)

18 observed variance is analyzed, providing as many components as original variables (Fabrigar et al., 1999). The model of this research is more similar to factor analysis because the goal is to use a few variables from the dataset to approximate the common variance and from that estimate the dynamic factor, not to explain all of the observed variance. In appendix A.1, it is explained what is the difference between an orthogonal and an oblique matrix, and the argument is given why the orthogonal matrix is used in this research.

4.3.2 Most Prominent Factors in Factor Analysis

As explained in section 4.3.1, this research uses factor analysis to determine which variables capture the most prominent common movements. According to the Kaiser Criterion, when the eigenvalue is lower than 1.0, the factor should not be used in any further analysis (Kaiser, 1958). There are three factors with an eigenvalue higher or equal to 1.0 as can be seen in table 2 in appendix A.2, so these three factors will be used for further analysis. Table 3 in appendix A.2 displays the rotated matrix. As can be seen in table 4 in appendix A.2, factor 1 is best captured by the standardized differenced value of m1, factor 2 by the real effective exchange rate, and factor 3 by credit.

The model of Engel et al. (2012) estimates the following equation, where the unobserved dynamic factor is a weighted average of three other estimates:

𝐹𝑖𝑡 = 𝛿̂1𝑖 𝑓̂1𝑡+ 𝛿̂2𝑖 𝑓̂2𝑡+ 𝛿̂3𝑖 𝑓̂3𝑡 (10)

In this model:

(19)

19

5. Results and Analysis

This section starts by listing the assumptions imposed by the model by Stock and Watson (2010) and examining whether all assumptions are satisfied. Then, it is specified why there is a need for the construction of two financial cycles. To answer the first part of the research question, the financial cycle is regressed on itself using an autoregressive model to examine how the cycle looks like. To answer the second part of the research question, the financial cycle and its lags are regressed on all financial variables from the dataset. Then, it is determined which lags correspond to crisis-months. In order to analyze what the effect is of the financial cycle on the financial variables during the recent financial crisis, it is checked whether the crisis-month lags of the financial cycle are significant explanatory for the financial variables. The final section compares the financial cycle found in this research to previous literature.

5.1 Assumptions of the Model

The model by Stock and Watson (2010) makes some strict assumptions. This paragraph summarizes precisely which assumptions are imposed, whether all assumptions are satisfied and how it is tested whether they are fulfilled. The four assumptions are as follows.

1. Observed financial variables (inflation, unemployment, exchange rate etcetera) are stationary.

2. Idiosyncratic disturbances in the observed financial variables are uncorrelated with the error in the movement of the common dynamic factor, 𝐸(𝑒𝑡, 𝜂𝑡−𝑘) = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑘.

3. Idiosyncratic disturbances in the observed variables (𝑒𝑡) are allowed to follow an

autoregressive structure.

4. Idiosyncratic disturbances in the observed financial variables are mutually uncorrelated also over time, cross-correlation of the error terms of observed financial variables is not allowed for, 𝐸(𝑒𝑖𝑡, 𝑒𝑗𝑠) = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑠 𝑖𝑓 𝑖 ≠ 𝑗.

The solutions to the above assumptions are as follows.

1. For all variables, the first difference of the logarithm was taken to ensure stationarity. The stationarity of all variables is checked using an Augmented Dickey-Fuller test. The Augmented Dickey-Fuller test tests whether the variable follows a unit-root process, with the null hypothesisthat the variable has a unit root against the alternative hypothesis that the variable is stationary (Stock & Watson, 2012). The null hypothesis is not rejected for

(20)

20

the variables consumption, consumer price index, current account, interest rate and claims on the government, which indicates they are nonstationary. They are not included in the analysis, even though there is economic intuition for those variables to be (partly) determining the Indian financial cycle. The p-values, critical values and values of the test statistics are displayed in table 5 in appendix A.2.

2. Assumption number two is guaranteed by ensuring that the variances of the error terms are constant, or homoscedastic, and hence independent of the observed financial variables (Breusch and Pagan, 1979). Homoscedasticity is ensured by using the robust formulas in computing the standard errors in the regression when needed (Stock & Watson, 2012). The need for the robust option is tested using a Breusch-Pagan test. Having homoscedastic variances of the error terms ensures that the error terms of each observed financial variable are constant, and cannot correlate with the error terms of the financial cycle and its lags (Breusch and Pagan, 1979).

3. Since Stata allows a maximum of 100 lags in the regressions, the autoregressive structure of the error terms can only be examined using these 100 lags, which is from June 2002 until December 2012. From figure 4 in appendix A.3 it can be seen that some error terms are partly auto correlated and some are not auto correlated. This corresponds to the assumption that the error terms are allowed to auto correlate.

4. Assumption number four is also guaranteed by ensuring that the variances of the error terms are homoscedastic. In this way, the variances of the error terms of the observed financial variables are not allowed to cross-correlate across observed variables.

5.2 Construction of Two Financial Cycles

The use of filters in Drehmann et al. (2012) and Stock and Watson (2010) differs and hence produce two different financial cycles. This section explains the difference and indicates which filter is used for which purpose in this research.

Drehmann et al. (2012) use the log first differenced series and apply the Christiano-Fitzgerald (CF) filter to the series to remove a potential deterministic trend. The CF filter is a band-pass filter and is optimal for random walk processes, where it estimates and removes the potential drift (Stata Manual Time Series Filter, 2013). Then, they construct the financial cycle from these series.

Stock and Watson (2010) standardize the log first differenced series and apply the Kalman filter in combination with maximum likelihood estimation. The Kalman filter separates the true observation from its expected noise, and therefore gives a more optimal estimate (Owens & Steigerwald, 2006).

(21)

21

5.2.1 Stochastic Trends versus Deterministic Trends

Taking the logarithm of the differenced series eliminates potential stochastic trends. However, there might still be deterministic trends present (Stock & Watson, 2012). Therefore, a filter is applied to remove this deterministic trend. As mentioned above, the key feature of a CF filter is that the series are random walk processes. However, the equation for the financial cycle (equation (9)) in the model by Stock and Watson (2010) does not correspond to a random walk process, unless the equation is autoregressive of order 1 (AR (1)). When the equation is autoregressive of a higher order, there is no longer a random walk and the CF filter is no longer optimal.

Another filter than can be used to remove a potential deterministic trend is the Hodrick-Prescott (HP) filter. The HP filter is a high-pass filter and discerns between the long run trend of a time series and short run fluctuations (De Jong & Sakarya, 2013), where the filter removes the long-run trend (Stata Manual Time Series Filter, 2013).

Since this research follows the model by Stock and Watson (2010), but also wants to compare the financial cycle from this research to the cycle by Drehmann et al. (2012), both the standardized log first differences series including HP filter and the log first differences series including CF filter are used to construct two financial cycles. The first cycle is used to examine how the Indian financial cycle looks like, the second cycle is used to compare the amplitude and frequency of the cycle to the cycle by Drehmann et al. (2012). In this way, the first cycle is adding to the existing literature by applying the theory of Stock and Watson (2010) on a new dataset, the second cycle enables to compare the cycle to existing literature.

5.3 Number of Lags of the Financial Cycle

The key features from the methodology that are inquired in this research are whether the financial cycle is auto regressive, and whether the financial variables are explained by the current and past values of the financial cycle. This section explains how many lags the financial cycle has in an auto regressive model and in regressing on the financial variables.

5.3.1 Autocorrelated residuals

One of the assumptions of the model by Stock and Watson (2010) is that the error terms are allowed to follow an autoregressive structure. That is, it has no implications for the model whether the residuals are autocorrelated. This research performs ordinary least squares (OLS) regressions for the financial cycle and its lags on each financial variable, and an auto regression (AR) of the lags of the financial cycle on itself. In OLS, serial correlation in the error terms does not lead to inconsistent estimates and does not introduce bias, but it does lead to

(22)

22

inconsistent OLS standard errors (Stock & Watson, 2012). Therefore, this research examines the auto correlations of the error terms of each regression using a correlogram. The lags of the error terms that fall outside the confidence interval are significant and therefore will be included as an explanatory lag.

5.3.2 Number Lags of the Financial Cycle by Regressing on Itself

Recall equation (9) from section 4.1.2

𝑓𝑡 = 𝜓1𝑓𝑡−1+ 𝜓2𝑓𝑡−2+ ⋯ + 𝜓𝑝𝑓𝑡−𝑝+ 𝜂1,𝑡 (9)

In regressing the standardized first differenced logarithm and HP filtered financial cycle 𝑓𝑡 on

its own lags, it can be seen in the final correlogram of figure 4 in appendix A.3 that the error term of the 31th_{lag is significantly autocorrelated at a significance level of five per cent. This}

indicates that 31 lags of the financial cycle can explain the current value. There is no need to use the robust option. The reason is that the Breusch-Pagan test indicates that the null hypothesis that the variance of the error terms are homoscedastic could not be rejected at a 10% significance level (Breusch and Pagan, 1979). Hence equation (9) in this model is:

𝑓𝑡= −0.401𝑓𝑡−1− 0.104𝑓𝑡−2+ ⋯ − 0.127 𝑓𝑡−31 (18)

The complete output can be found in appendix B.1.

5.3.3 Number Lags of the Financial Cycle by Regressing on Financial Variables

The outcome of regressing 100 lags of the factor and its current value on the financial variables determines how many lags of the dynamic factor are significantly explaining the financial variables. Again, it is examined which auto correlated lag of the residual is significant at five per cent using a correlogram. If the error terms are not homoscedastic, the robust option is used in performing the regression. The most important features of the output are displayed in table 6 in the appendix A.2, and the complete output for the regressions of all variables is given in appendix B.2. Some correlograms display values that are on the boundary of the confidence interval. In order to ensure that the OLS standard errors are consistent, the boundary values are still included as explanatory lags. All correlograms can be found in figure 4 in appendix A.3.

5.4 Effect of Indian Financial Cycle during Financial Crisis

The ninth of August 2007 can be viewed as the starting date of the recent global financial crisis (Phillips and Yu, 2010). At this date, BPN Paribas announced that it would halt “withdrawals from three investment funds because it could not ‘fairly’ value their holdings after U.S. subprime mortgage losses roiled credit markets” (Boyd, 2007). The next few days the

(23)

23

European Central Bank, Federal Reserve, Bank of Canada and Bank of Japan started their first of many interventions (Guillén, 2009).

When December 2012 (the final time period of the dataset) is considered time 𝑡, then August 2007 is considered time 𝑡 − 65. Hence, it is examined whether the 65th_{and smaller lags of the}

dynamic factor have a significant effect on the financial variables, measured by the p-value of the lag of the regressions.

As can be seen in the last column in table 6 in appendix A.2, all variables have some significant explanatory lags of the financial cycle from 𝑡 − 65 until 𝑡, except net foreign assets, three-month rate, and the discount rate. There appears to be no one-to-one relation between the number of significant lags and the explanatory power of the model, explained by the adjusted R2_{. When examining the adjusted R}2_{of the regressions, it is noted that most values are larger}

than or equal to 66.4 per cent. A few exceptions are foreign currency reserves, the real exchange rate and inflation, for which it is not obvious why they have lower values for the adjusted R2_{. Overall, the values for the adjusted R}2_{are quite high.}

Interestingly, the adjusted R2_{for M1, the real effective exchange rate and credit are not the}

three highest, since that would be expected because they compose the financial cycle. The value for credit is however very high, 96.8 per cent. Other very high values of the adjusted R2

are found for claims on the private sector and foreign direct investment. As can be seen in table 4 in appendix A.2, factor 3 was chosen to be represented by credit, although the value for claims from the private sector also had a high value. The same goes for foreign direct investment and factor 1, although the values are not as close as for claims on the private sector and factor 3. That could be the reason that claims on the private sector and foreign direct investment have high values for the adjusted R2_.

Comparing behavior of the financial cycles between the cycle using the HP filter (figure 5) and the cycle using the CF filter (figure 6) catches the attention of different volatility patterns. Whereas the financial cycle composed following Stock and Watson (2010) (figure 5) has a high volatility with five peaks in early 2007, early 2008, mid 2009, late 2011 and early 2012, the financial cycle composed following Drehmann et al. (2012) (figure 6) has a much lower volatility with only two peaks at late 2007 and early 2010. However, both cycles reach approximately the same frequency after 2007 as compared to before 2007, only the peaks and troughs become more intense. The effect of the recent financial crisis on the Indian financial cycle is that the movements in the financial cycle, composed by M1, the real effective exchange rate and credit, become more volatile.

(24)

24

5.5 Comparison with Previous Models

The two main models described in the literature review are those by Drehmann et al. (2012) and Claessens et al. (2011). This section compares the model from this analysis to these two analyses, addressing the frequency and amplitude of the financial cycle and the choice of variables making up the financial cycle.

5.5.1 Methodology of Previous Literature

Drehmann et al. (2012) use two approaches in the methodology: turning point analysis and frequency-based filters, as applied to individual series. Using turning point analysis allows the financial cycle to be compared to the business cycle. Furthermore, they argue that the cycles using band-pass filters are not similar to the cycles using turning point analysis and that they cannot be compared easily. When they combine the series from the frequency-based filters into one financial cycle, they aggregate the average of the filtered series, which is also what this research does.

Since the focus of this research is to find how the Indian financial cycle looks like and not to compare the financial cycle to the business cycle, the results of only the frequency-based filters are compared to the results from this research.

Claessens et al. (2011) use only a turning point analysis, derived from the method of finding the peaks and troughs of the business cycle. Since their methodology differs from the methodology from this research, the financial cycles cannot be compared. Therefore, section 5.5.2 and 5.5.3 compare the financial cycle from this research to the financial cycle by Drehmann et al. (2012) only.

5.5.2 Frequency and Amplitude of Previous Literature

Comparing the Indian financial cycle in figure 4 to the United States financial cycle by Drehmann et al. (2012) in figure 5 indicates that the frequency of peaks and troughs is a lot smaller in the United States financial cycle than the Indian financial cycle. Accordingly, a period lasts much shorter in the Indian financial cycle than a period in the cycle by Drehmann et al. (2012). Where the United States financial cycle is varying from -0.15 to 0.15, the Indian financial cycle varies from -0.006 to 0.007, which is about twenty-three times smaller. Also, the amplitude in the cycle by Drehmann et al. (2012) increases over time, just like the amplitude of the financial cycle in this research.

Furthermore, Drehmann et al. (2012) chose to use six main variables: credit, credit/GDP, house prices, equity prices, the aggregate asset price index and GDP. Claessens et al. (2011) chose only three main variables: credit, house prices and equity prices.

(25)

25

5.5.3 Preliminary Conclusion

Overall, when the financial cycle constructed in this research is compared to the financial cycle by Drehmann et al. (2012), it can be concluded that the former has a much smaller amplitude and a larger frequency, so it has a smaller variation and its peaks and troughs are less intense but the cycle varies more often. Also, the former is composed of fewer variables, and the two cycles have only one variable in common which is credit.

5.6 Internal Validity

As mentioned in section 4, the analysis in this thesis is built upon the model by Stock and Watson (2010). The assumptions about the model and their solutions were stated explicitly in section 5.1. Nonetheless, research often brings difficulties with regard to internal and external validity. The two main problems with internal validity in this research are imperfect multicollinearity and heteroscedasticity.

5.6.1 Imperfect Multicollinearity

The problem of imperfect multicollinearity arises due to the ‘double’ measure of some financial indicators, mentioned in section 4.2. The dataset of this research includes four variables that were measured by two financial indicators each. These double measures were included to analyze which one was of most use to this research. For two measures (consumption and car sales; consumer price index and inflation) the problem is solved since consumption and consumer price index are excluded from the analysis because of nonstationarity. However, M1 and M2 and the real exchange rate and the real effective exchange rate are still both included. During factor analysis it was found that M1 was the most determinant component of factor 1, more determining than M2. However, table 4 in appendix A.2 indicates that M2 has a slightly higher value for the adjusted R2_{. Therefore, it is not clear which financial indicator is}

of better use in this research. The real effective exchange rate is the most determining component of factor 2. The real effective exchange rate has a higher adjust R2_{value than the}

real exchange rate, therefore the real effective exchange rate is of better use than the real exchange rate.

However, imperfect multicollinearity is not a big problem in this research. The financial cycle is composed of three variables: M1, credit and the real effective exchange rate. The regressions of the financial cycle and its lags are done on all stationary financial variables. Both M1 and M2 and the real exchange rate and the real effective exchange rate are included in the output, but those four regressions are still four separate linear regressions that do not have any interconnection.

(26)

26

5.6.2 Heteroscedasticity

The second potential problem is heteroscedasticity in the variance of the error terms, which relates to assumption two described in section 5.1. The financial cycle is estimated by the weighted average of the three observed financial variables M1, the real effective exchange rate and credit. When the financial cycle is regressed on those observed financial variables, the error terms of those three variables are automatically incorporated in the error terms of the financial cycle. However, after each regression of the financial cycle and its lags on the financial variables, homoscedasticity of the error terms was checked using the Breusch-Pagan test. If they were not, the robust option was used. Therefore, homoscedasticity should be ensured.

5.6.3 Other Potential Threats

Besides imperfect multicollinearity and heteroscedasticity, other threats such as omitted variable bias, missing data, simultaneous causality, measurement error and misspecification of the functional form could also cause problems in the internal validity (Stock & Watson, 2012). The way that is dealt with omitted variable bias and missing data is by including as many variables in the dataset as technically feasible. The problem can still be at hand, but can in the current circumstances not be avoided. Simultaneous causality is quite a big problem, but that is the case for many financial variables, whose interconnectedness is very hard to neglect. In other words, there are very limited financial variables that are not interconnected with other financial variables. Interestingly, the regressions with the three variables that compose the financial cycle do not have the highest explanatory power compared to other financial variables, except credit. Measurement error can also be at hand due to human error. This research collected the data and performed the regressions with large discretion. The functional form of the regression function could be polynomial of some degree instead of linear. However, the model by Stock and Watson (2010) indicate a linear functional form, so this research holds on to that.

Some data are only available on a yearly basis, while other on a monthly basis. The yearly data were interpolated to monthly data, as mentioned in section 4.2. This might lead to not as accurate and reliant estimates as when only monthly data had been used.

5.7 Threats to External Validity

The external validity of this research depends on the population and setting to which the generalization is made (Stock & Watson, 2012). In addition, this research did not account for country-specific effects because the data of the financial variables are specifically taken from the Indian economy. Therefore, the results of this study cannot be generalized to other

(27)

27

countries except if a new dataset is selected from the economy from this country and the factor analysis is based on these data.

5.8 Further Research

Due to the lack of data available, this thesis was not able to compare the financial cycle between the pre-1991 and post-1991 periods to examine whether the opening of the Indian economy led to financial liberalization. Further research could incorporate this feature to provide a more complete picture.

(28)

28

6. Conclusion

This thesis examined whether a financial cycle is present in India and whether it influenced the Indian economy during the recent global financial crisis. The research question formulated in the introduction is “How does the Indian financial cycle look like and did it influence the Indian economy in the recent global financial crisis?”.

From the literature, it can be concluded that a financial cycle is present in India and that it has a significant impact on the Indian economy. The reason is that India experiences increasing trade globalization and financial integration, while the country does have a low involvement in the United States housing market and does not have an unhealthy banking system.

The empirical research found that a financial cycle is present in India, and is best represented by the weighted average of the money supply M1, the real effective exchange rate and net domestic credit. The Indian financial cycle and its lags significantly affect the movements in fourteen out of seventeen observed financial variables during some periods of the recent financial crisis. Also, the Indian financial cycle has an amplitude that is twenty-three times smaller than the amplitude of the United States financial cycle, but the frequency is much higher. The fact that the amplitude of the Indian cycle is twenty-three times smaller than the amplitude of the American cycle could be caused by low globalization and integration, hence contradictory to the expectation formed by the literature review. Furthermore, Drehmann et al. (2012) found that the duration and intensity of the financial cycle depends on the degree of financial liberalization in the economy, which can intensify the cycle. The latter translates to more financial liberalization leads to longer periods (smaller frequency) and larger amplitudes. That corresponds to the United States being more financially liberalized than India because the American cycle has both a larger amplitude and a smaller frequency.

Even though it might look like the empirical results contradict to the expectation formed by the literature, that India experiences no globalization and financial integration, this is only true relative to the United States. Since the United States is a very globalized and financially integrated county, many countries comparing to the United States would seem less globalized and financially integrated. Therefore, when analyzed absolute terms, the expectation formed by the literature corresponds by the outcome of the empirical research. The amplitude of the Indian cycle increases over time and therefore becomes more financially integrated and globalized over time.

(29)

29

References

Bai, J., & Ng, S. (2008). Large Dimensional Factor Analysis. Foundations and Trend in Econometrics, 3(2), 89-163.

Bank for International Settlements. (2014, June 29). 84th Annual Report, 65-83

Baum, C.F., Schaffer, M.E., & Stillman, S. (2003). Instrument Variables and GMM: Estimation and Testing. Boston College, Department of Economics, Working Paper, 545

Bernanke, B.S., Boivin, J., & Eliasz, P. (2010). Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach. NBER Working Paper Series, 10220.

Borio, C. (2012). The Financial Cycle and Macroeconomics: What have we learnt? BIS Working Papers, 395.

Boyd, S. (2007, August 9). BNP Paribas Freezes Funds as Loan Losses Roil Markets (Update5). Bloomberg. Retrieved June 15, 2015, from

http://www.bloomberg.com/apps/news?pid=newsarchive&sid=aW1wj5i.vyOg

Breitung, J., & Eickmeier, S. (2005). Dynamic Factor Models. Discussion Paper Series 1: Economic Studies, 38, Deutsche Bundesbank.

Breusch, T.S., & Pagan, A.R. (1979). A Simple Test for Heteroscedasticity and Random Coefficient Variation. Econometrics, 47(5), 1287-1294.

Claessens, S., Kose, M., & Terrones, M. (2011). Financial Cycles: What? How? When? IMF Working Paper, 11/76.

Christiano, L.J., & Fitzgerald, T.J. (2003). The Band Pass Filter. International Economic Review, 44(3), 435-465.

Datastream (2015). 1 April 2015.

de Jong, R. M., & Sakarya, N. (2013). The Econometrics of the Hodrick-Prescott Filter. Review of Economics and Statistics.

Drehmann, M., Borio, C., & Tsatsaronis, K. (2012). Characterizing the Financial Cycle: Don't lose sight of the medium term! BIS Working Papers, 380.

(30)

30

Dua, P., & Ranjan, R. (2010). Exchange Rate Policy and Modelling in India. Development Research Group. Department of Economic Analysis and Policy, Reserve Bank of India, 33. Ebeling, Richard M. "The "Austrian" Theory of the Trade Cycle." The Austrian Theory of the Trade Cycle and Other Essays. Auburn: Ludwig Von Mises Institute, 1996. 25-35.

Engel, C., Mark, N.C., & West, K. (2012). Factor Model Forecasting of Exchange Rates. University of Notre Dame Working Paper #012.

English, W., Tsatsaronis, K., & Zoli, E. (2005). Investigating the Relationship Between the Financial and Real Economy. BIS Papers, 22, 228-252.

European Central Bank. (2014). Capturing the Financial Cycle in the Euro Area Counties. Financial Stability Review.

Fabrigar, L.R., Wegener, D.T., MacCallum, R.C., & Strahan, E.J. (1999). Evaluating the Use of Explanatory Factor Analysis in Psychological Research. Psychological Methods, 4, 272-299.

Guillén, M.F. (2009). The Global Economic & Financial Crisis: a Timeline. The Lauder Institute, Wharton, University of Pennsylvania.

Hatzius, J., Hooper, P., Mishkin, F., Schoenholtz, K., & Watson, M. (2010). Financial Conditions Indexes: A Fresh Look after the Financial Crises. NBER Working Paper Series, 16150.

International Monetary Fund World Economic Outlook (2015).

(https://www.imf.org/external/pubs/ft/weo/2014/02/weodata/index.aspx), 1 April 2015.

Jarociński, M. & Smets, F.R. (2008). House Prices and the Stance of Monetary Policy. Federal Reserve Bank of St. Louis Review, 90(4), 339–65.

Kaiser, H.F. (1958). The Varimax Criterion for Analytic Rotation in Factor Analysis. Psychometrika, 23(3), 187-200.

Keynes, J. (1936). The General Theory of Employment, Interest, and Money. Zurich: International Relations and Security Network.

Kydland, F., & Prescott, E. (1982). Time to Build and Aggregate Fluctuations. Econometrica, 50(6), 1345-1370.

Lamont, P. (2008, June 30). Further Than Most Expect. Lamont Trading Advisors. Retrieved July 17, 2015, from http://www.ltadvisors.net/Newsletter/Expect.htm

(31)

31

Molenaar, P. (1985). A Dynamic Factor Model for the Analysis of Multivariate Time Series. Psychometrika, 50(2), 181-202.

Ng, T. (2011). The Predictive Content of Financial Cycle Measures for Output Fluctuations. BIS Quarterly Review, 53-65.

Nier, E., Sedik, T., & Mondino, T. (2014). Gross Private Capital Flows to Emerging Markets: Can the Global Financial Cycle be Tamed? IMF Working Paper, 196.

Owens, J. P., & Steigerwald, D. G. (2006). Noise reduced realized volatility: A Kalman filter approach. Advances in Econometrics, 20, 211-227.

Phillips, P., & Sul, D. (2007). Transition Modeling and Econometric Convergence Tests. Econometrica, 75(6), 1771-1855.

Phillips, P.C., & Yun, J. (2010). Dating the Timeline of Financial Bubbles During the Subprime Crisis. Cowles Foundation Discussion Paper, 1770.

Sinha, A. (2010). Impact of the international banking crisis on the Indian financial system. BIS Papers, 54, 225-234.

Subbarao, D. (2009). Impact of the Global Financial Crisis on India Collateral Damage and Response. Speech Delivered at Symposium on "The Global Economic Crisis and Challenges for the Asian Economy in a Changing World", pp. 1-14.

Stata Manual Time Series Augmented Dickey-Fuller Unit-Root Test (2013).

http://www.stata.com/manuals13/tsdfuller.pdf

Stata Manual Time Series Dynamic-Factor Model (2013).

http://www.stata.com/manuals13/tsdfactor.pdf.

Stata Manual Time Series Filter (2013). http://www.stata.com/manuals13/tstsfiler.pdf. Stata Manual Multivariate Statistics Factor Analysis (2013).

http://www.stata.com/manuals13/mvfactor.pdf

Stock, J., & Watson, M. (2005). Implications of Dynamic Factor Models for VAR Analysis. NBER Working Paper Series 11467.

Stock, J., & Watson, M. (2010). Dynamic Factor Models. Oxford Handbook of Economic Forecasting.

(32)

32

Stock, J., & Watson, M. (2012). 14 - Introduction to Time Series Regression and Forecasting. In Introduction to Econometrics (Third International Edition, pp. 558-598). Essex: Pearson Education Limited.

Taylor, J. (2007). Housing and Monetary Policy. NBER Working Paper Series 13682.

Tempelman, J. (2010). Austrian Business Cycle Theory and the Global Financial Crisis: Confessions of a Mainstream Economist. The Quarterly Journal of Austrian Economics,13(1), 3-15.

Von Mises, L. (1981). The Theory of Money and Credit. Indianapolis, IN: Liberty Fund. (Republished and translated, original version written in 1912 in German)

Worldbank. (1991, August 23). Volume I: Policies for Adjustment with Growth. India 1991 Country Economic Memorandum, i-viii.

(33)

33

Appendix A

Explanation of orthogonal versus oblique matrix, tables and figures

A.1 Orthogonal Versus Oblique Matrix

When performing factor analysis in Stata, the direct output provided indicates the smallest number of factors that can explain the common variance of the observed variables (Kaiser, 1958). The output is unrotated, which means that the factors are forced to be orthogonal (independent) but this causes most of the variable loadings to be on the first few factors, and multiple variable loadings on more than one factor (Stata Manual Multivariate, 2013). When the matrices are rotated, it is ensured that the load of the variables is mostly projected on one factor and far less on other factors (Stata Manual Multivariate, 2013). There are two main options for rotation: orthogonal and oblique (Kaiser, 1958). The default for orthogonal rotation is a varimax rotation (Stata Manual Multivariate, 2013). A varimax rotation enables better to match each single factor with a variable (Kaiser, 1958). The assumption that the matrices are orthogonal (independent) is not very realistic. However, if the oblique rotation would have been used, a correlation between the rotated factors is allowed for (Stata Manual Multivariate, 2013). Since the goal of factor analysis in this research is to determine which variable is a good proxy for which factor and then combining those into the factor for the financial cycle, an orthogonal rotation is more suited and hence a varimax rotation is performed.

(34)

34

A.2 Tables

Table 1 – Financial Indicators Financial Indicator

(abbreviation)

Measurement Source (Time Frequency)

Stationary After Log First Difference? Real Exchange Rate (e) Rate (Indian Rupees per

Dollar)

Datastream (M) Yes Money Supply M1 (m1) Volume in Rupees Datastream (M) Yes Money Supply M2 (m2) Volume in Rupees Datastream (M) Yes Bombay Stock Exchange

National 100 Share Price Index (stock)

Index Datastream (M) Yes

3-month Treasury Bill Rate (threemrin)

Rate Datastream (M) Yes

Lending Rate (lending) Rate Datastream (M) Yes Discount Rate (discount) Rate Datastream (M) Yes Foreign Currency Reserves

(fcr)

Volume in Rupees Datastream (M) Yes Car Sales (carsales) Volume in Number of Cars Datastream (M) Yes Capital Account (ka) Volume in Rupees Worldbank (Y) Yes Consumption (consumption) Volume in Rupees Worldbank (Y) No Consumer Price Index (cpi) Index (2010=100) Worldbank (Y) No Inflation (inflation) % Change per Year Worldbank (Y) Yes Net Domestic Credit (credit) Volume in Rupees Worldbank (Y) Yes Net Foreign Assets (nfa) Volume in Rupees Worldbank (Y) Yes Unemployment (unempl) % of Total Labour Force Worldbank (Y) Yes Real Interest Rate (r) Rate Worldbank (Y) No Claims on the Private Sector

(claimpr)

Volume in Rupees Worldbank (Y) Yes Claims on the Government

(claimgov)

Volume in Rupees Worldbank (Y) No Foreign Direct Investment Net

Inflow (fdi)

Volume in Rupees Worldbank (Y) Yes Real Effective Exchange Rate

(reer)

Rate (2010=100) Bank of International Settlement (Y)

Yes

Current Account (ca) Volume in Rupees IMF World Economic Outlook (Y)

The financial cycle in India