A Financial Stress Index for a Small Open Economy: The Australian Case

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) A Financial Stress Index for a Small Open Economy: The Australian Case Pedro Gomis-Porqueras, Romina Ruprecht, Xuan Zhou 2023-029 Please cite this paper as: Gomis-Porqueras, Pedro, Romina Ruprecht, and Xuan Zhou (2023). “A Financial Stress Index for a Small Open Economy: The Australian Case,” Finance and Economics Discussion Series 2023-029. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.029. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

A Financial Stress Index for a Small Open Economy: The Australian Case∗† Pedro Gomis-Porqueras1, Romina Ruprecht2, and Xuan Zhou3 1 Queensland University of Technology 2Federal Reserve Board 3Renmin University of China This Version: April 25, 2023 Abstract We construct a Financial Stress Index (FSI) for a small open economy, which aims to provide clear and timely signals of financial market strains. This can be used in developing appropriate responses to address these adverse events. To do so, we use the principal component framework and apply it to Australian monthly data on interest rates, spreads, exchange rates, house price growthandinflationexpectations. Decomposingtheindexintoforeignanddomesticcomponents, we find that the foreign factors can explain more than half (57.4%) of our Australian Financial Stress Index (AFSI). To determine the information content of our index, we run a series of Granger causality tests on several economic and financial observables. We also estimate whether including the AFSI can improve the prediction of the different economic and financial outcomes relative to a specification that uses only its own previous data. We find that including the AFSI improves the forecasts for future retail sales growth and bank credit growth. Finally, we show that financial stress can have non-linear effects on bank credit growth. In particular, an increase in financial stress affects credit growth more adversely if AFSI is high. This result further highlights the importance of an accurate and timely measure of financial stress in an economy for researchers and policy makers. Keywords: financial stress index, financial stability, small open economies JEL Classifications: F30, G01, G15 ∗The views expressed in this paper are solely those of the authors and should not be interpreted as reflecting the views of the Board of Governors, its staff or anyone associated with the Federal Reserve System. All remaining errors are our own. †We would like to thank the seminar participants at the Federal Reserve Board, seminar participants of the Empirical Macro reading group and the Economic Theory reading group at the University of Basel, Ed Lin and Michelle Wright for their comments and suggestions. Contact: Pedro Gomis-Porqueras: peregomis@gmail.com, Xuan Zhou: aadazhou@gmail.com, Romina Ruprecht: romina.d.ruprecht@frb.gov 1

1 Introduction Equity and credit markets are a cornerstone of the financial system and are critical for economic growth. Although it was generally recognized that financial frictions could play an important role in economic fluctuations, the 2007-2009 financial crisis made it clear that the adverse effects of financial disruptions on economic activity could be far worse than anticipated. In particular, Ollivaud and Turner (2015) find that among the 19 OECD countries, that experienced a banking crisis over the period 2007-11, the median loss in potential output in 2014 is estimated to be about five and a half per cent, compared with a loss in aggregate potential output across all OECD countries of about three and a half per cent.1 In order to prevent such loses, it may be useful to have various measures that try to capture disruptions in the normal functioning of financial markets. To identify such abnormal episodes researches gather a variety of financial data capturing increased uncertainty that private investors are facing, increased asymmetric information in financial markets and decreased willingness to hold risky and illiquid assets. This paper aims to develop a clear and timely measure of financial stress for a small open economy that can serve as a warning system for investors and policymakers when monitoring financial markets. A Financial Stress Index (FSI) aims to monitor the current state of financial markets by creating a time series that has the property that increases in the index indicate increased financial stress. Since no single financial market is independent from the activity in other markets within a country and across countries, we consider a variety of financial and macroeconomic measures. In addition, since we are considering a small open economy, we also explicitly consider external factors that directly affect domestic output and prices through trade and price channels as well as indirectly through domestic financial markets. Because of the potential for negative spillovers from financial markets to the real economy, accurately measuring financial stress is important to investors and policymakers alike. Thus, understanding the specific channels by which negative disturbances to financial markets can spill over to the rest of the economy can be helpful in providing clear and timely signals of market strains to develop appropriate responses to address these adverse events. Financialstressinitselfisabstractandhasnouniquedefinition. Inthispaper, wefollowthenotion advocatedbyHakkioetal.(2009). Weassumethatfinancialstressisdefinedasperiodswithincreased uncertainty about the fundamental value of financial assets or the behavior of investors, increased asymmetric information and a decreased willingness to hold risky or illiquid assets.2 To construct an Australia Financial Stress Index (AFSI), we collect monthly data on interest rates, exchange rates, spreads, house price growth and inflation expectations. We aggregate this data by applying principal component analysis (PCA) to establish a single index (the AFSI) that provides a measure of financial 1Laeven and Valencia (2013) estimated the output loss of banking crises in advanced countries to be 32% of trend incomeandthefiscalcoststobe4%ofGDP.Similarly,Cardarellietal.(2009)findsthatrecessionsthatwerepreceded by financial crises tend to be more severe and Afonso et al. (2018) provide evidence for the adverse impact of financial stress on the economy for several countries. 2See Balakrishnan et al. (2011) or Illing and Liu (2006) for alternative definitions and Kliesen and Smith (2010) for an overview. 2

markets stress in Australia. The underlying assumption is that financial stress moves the various observables series jointly. The data selection for the AFSI closely follows the approaches taken in the existing literature, but also includes data that are important and relevant for small open economies, such as Australia.3 Given that we are interested in small open economies, we decompose the AFSI and measure how much of financial stress in Australia are due to financial stress abroad. We do so by regressing the AFSIonfinancialstressindicesforChina, JapanandSouthKoreaasAustralia’smaintradingpartner, and the United States, which is a global financial center. We find that foreign financial stresses can account for more than half of the AFSI. Moreover, to evaluate the information content of our financial stress measure, we compare it to several stress indices for both Australia and other countries. We also consider and compare the AFSI to broader measures of stress such as economic policy uncertainty indices. When doing so, we find that the AFSI is largely in line with foreign financial stress measures. We also show that there are small differences between AFSI and other indices that capture financial stress and economic uncertainty for Australia.4 We also determine whether the AFSI has relevant information content by helping better predict macroeconomic outcomes. To do so, we run a series on Granger causality tests on real GDP per capita growth, retail sales growth, bank credit growth and the unemployment rate. We find that the AFSI can improve forecasts for both retail sales growth and bank credit growth relative to forecasts that only rely on their own past data. Finally, we estimate the effect of a financial stress shock to the economy. To do so, we consider a time-varying VAR that includes central bank balance sheet growth, bankcreditgrowth, retailsalesgrowthandtheunemploymentrate. Ourresultsshowthattheeffectof an increase in financial stress depends on the level of financial stress. Overall, when financial stress is higher we observe a stronger reaction of a financial stress shock to central bank balance sheet growth and bank credit growth. These latter findings indicate a non-linear relationship between financial stress and economic outcomes. The various analysis carried out in this paper suggest that the AFSI contains relevant macroeconomic information, as in Davig et al. (2010), van Roye (2014), Hatzius et al. (2010) and Groen et al. (2020). Thus, the AFSI might be able to provide clear and timely signals of financial stress in the Australian economy, which may be useful for both policymakers and investors alike. The remainder of this paper is structured as follows. Section 2 provides a literature review, while section 3 describes the data and the methodology used to construct the AFSI. Section 4 compares the index to several measures of financial stress and uncertainty in Australia and abroad. Section 5 relates the index to economic different outcomes, while section 6 concludes. 3Kliesen et al. (2012) provide an overview of the variables included in some selected FSI. They furthermore argue that papers constructing Financial Condition Indices (FCI) tend to include more information on quantities, prices and economic indicators, whereas papers measuring FSI tend to focus solely on prices. 4These latter indicators rely on different data or capture economy wide notions of uncertainty. 3

2 Literature Review This paper connects with two strands of literature. The first one studies the effect of financial stress and financial crises on the economy, while the second quantifies and measures financial stress. In order to measure financial stress and disruptions, Financial Stress Indices (FSI) or Financial Condition Indices (FCI) are constructed using a wide array of financial and macroeconomic variables. Whileearlypapersprovidemeasuresoffinancialstressasbinaryvariable, meaningeithertheeconomy is in a crisis or not, other papers aim to construct indices that indicate the severity of financial stress. These include Illing and Liu (2006) for Canada, Hakkio et al. (2009) (Kansas City Fed FSI or KCFSI), Kliesen and Smith (2010) (St. Louis FSI or STLFSI), Oet et al. (2012) (Cleveland FSI or CFSI), Hatzius et al. (2010) and Groen et al. (2020) for the United States , Cardarelli et al. (2009) (IMF FSI) for selected advanced economies, Park and Mercado Jr (2014) for emerging market economies, Balakrishnan et al. (2011) for 18 emerging economies, van Roye (2014) for Germany and Hanschel et al. (2005) for Switzerland, among others. In the existing literature on FSI and FCI, the underlying data used to construct these indices vary greatly. Kliesen et al. (2012) provide an overview of the variables included in FSI and FCI. The authors conclude that papers constructing FCI tend to include more information on quantities, prices and economic indicators, whereas papers measuring FSI tend to focus solely on prices. The methodology to construct FSI also varies to some extent across papers, as there are several ways to combine multiple variables into a single index. Cardarelli et al. (2009), Hanschel et al. (2005) and Balakrishnan et al. (2011) use an equal variance weighting average to construct a single index, whereas Hakkio et al. (2009) and Kliesen and Smith (2010) use principal component analysis.5 Illing and Liu (2006) use factor analysis, credit-aggregate based weighted averages, variance equal weighted averages and transformation of variables using the sample CDF’s to construct indices. Lastly, van Roye (2014) uses a dynamic approximate factor model to construct an index for Germany. In terms of the information content of these indices, Davig et al. (2010) show that financial activity tendstobelowerinhighfinancialstressperiods. Moreover, theauthorsalsofindthatafinancialstress shock has a larger macroeconomic impact during periods of heightened financial stress. Similarly, van Roye (2014) finds that financial stress has a negative impact on the economy once financial stress increases above a specific threshold. Hatzius et al. (2010) test several existing FSI to see if they can improveforecastsofeconomicactivity. TheyfindthatincludingtheFSItogetherwithsinglemeasures (such as the stock market index, real M2, the term spread, the federal funds rate or the short-term credit spread) can improve forecasts. Moreover, the authors show that some FSI can outperform the stock market as an indicator of financial stress. Also Groen et al. (2020) find that including financial stress can improve forecasts to industrial production. ThispaperisalsocloselyrelatedtoHartiganandWright(2023), whoconstructaFCIforAustralia 5To compare the different methods, Park and Mercado Jr (2014) use both a variance equal weighted average and principal component analysis. 4

using data on asset prices, interest rates and spreads, credit and money, outstanding debt securities, as well as indicators of leverage, banking sector risk, financial system complexity, financial market risk and survey indicators of businesses and consumers.6 The authors find that more restrictive financial conditionsplayanimportantroleinexplainingdownsiderisktogrowthinbothGDPandemployment and upside risk to changes in the unemployment rate. In contrast to Hartigan and Wright (2023), we use primarily monthly price data to construct the AFSI. Using higher frequency data allows us to receive a closer insight to short-term developments in financial markets. 3 Developing the Australian Financial Stress Index Since we are considering a small open economy, domestic output and prices are influenced by two type of factors. The first group characterizes domestic financial markets, while the second reflects the external factors.7 Conditions abroad directly affect domestic output and prices through international trade. In addition, foreign factors can indirectly impact domestic prices and output through domestic financial markets. Moreover, external conditions may influence monetary policy, ultimately affecting domestic financial conditions, output and prices. External factors also influence the exchange rate and domestic asset prices through cross-border capital flows, affecting the terms of trade, wealth and financing conditions. Depending on the extent of domestic financial frictions, financial markets can then also amplify the direct effects of external shocks through a feedback effect that runs from interactions between the real economy and financial markets. Explicitly considering external factors when constructing a FSI for a small open economy is then key. Following the approach used in the literature, we construct such a financial stress index using the principal component analysis framework. To do so, we collect a series of variables that capture different aspects of financial disruptions faced by investors in a variety of financial markets. 3.1 Financial Measures Symptoms of financial stress are informed by both theory and practice. These include: (i) uncertainty about the fundamental value of financial assets or the behavior of investors; (ii) increased asymmetric information; and (iii) decreased willingness to hold risky or illiquid assets. Thus, data used to construct the AFSI needs to capture different financial markets trading patterns and investors’ expectations. In particular, we collect information for the banking sector, the equity market, government securities, international trade, and the foreign exchange market. The data contains monthly information on changes in interest rates, yield spreads, volatility of the effective exchange rate, returns and volatility in equity markets, idiosyncratic volatility of bank stock prices, housing markets, inflation expectations and sovereign real debt spreads. We also consider Australia’s major trading partners. 6The authors use quarterly data in an unbalanced panel from 1976 to 2020 and dynamically demean the series using a 10-year backward-looking estimate of the sample mean. 7These external factors include commodity prices, world demand and global financial conditions. 5

The data considered in this paper spans from January 1990 to August 2020. The corresponding data sources are listed in Table 13, which can be found in Appendix A. Banking Sector It is well known that disruptions in the banking sector contribute significantly to periods of financial stress. To capture this important feature, we include various banking sector measures. Following Hakkio et al. (2009), we construct the banking sector β over a 2-year time horizon. This measure is constructed as follows cov(r,m) β = ; var(m) where r denotes the monthly return on the bank stock index and m denotes the monthly return of the Australian stock market index (ASX200). When β > 1, the volatility in the banking sector is larger than the volatility in the overall equity market, implying that the banking sector is relatively risky.8 Using the estimates of β, we then calculate the residual return of the banking sector aggregate stock market index. Lastly, we estimate the residual return volatility using a GARCH(1,1) process. Inaddition, Hakkioetal.(2009), amongothers, emphasizethatinperiodsoffinancialstress, banks are likely to experience increased uncertainty about the quality of borrowers. As a result, they are less willing to hold risky assets. This increased uncertainty can lead to sharp increases in the inter-bank rate and the rate of bank accepted bills, leading to an increase in the spread of bank refinancing rates and short-term government bond yields. To capture this, we consider the 3-month TED spread and the 3-month BAB treasury spread. The 3-month TED spread measures the spread between the Australia 3-month inter-bank rate (BBSW) and the 3-month government bond yield. The 3-month BAB treasury spread measures the difference between the 3-month BAB (bank accepted bills) rate and the 3-month government bond yield. Finally, to capture changes to banks’ funding costs that are driven by changes in macroeconomic circumstances, we also consider changes in the cash rate, which is the key policy rate of the Reserve Bank of Australia. Equity Market As Hakkio et al. (2009), among others, highlight that financial stress periods are accompanied by increaseduncertaintyaboutassets’fundamentals. Thisleadstosharpchangesinreturnsandincreases in volatility in equity markets. To capture this phenomena, we consider data on S&P/ASX 200 return andS&P/ASX 200 return volatility, whichcaptureactivityintheAustralianequitymarket. Following Park and Mercado Jr (2014), the return volatility is calculated by estimating a GARCH(1,1) process. Housing Market TheGlobalFinancialCrisis(GFC)hasemphasizedhowdisruptionsinthehousingcanleadtofinancial stress. Sharpdeclinesinthehousingmarketcanleadtoincreasedlossesforbanksandinvestors. Thus, it is not too surprising that the housing market is a key factor when designing macro-prudential 8We refer to Park and Mercado Jr (2014) for more on this. 6

policies. In order to capture financial stress arising from the housing market, we include housing price growth data. This series combines the growth of the hedonic home value index in eight major Australian capital cities. Foreign Exchange Market As a small open economy, it is likely that financial stresses and macroeconomic shocks experienced abroad can spill over to the Australian economy. To account for this possibility, we include data on the effective exchange rate volatility. To measure the effective exchange rate volatility, we take first differences of the trade weighted index of the Australian Dollar.9 We then measure the effective exchange rate volatility with a GARCH(1,1) process. Inflation Expectations In addition to changes in behavior and beliefs, agents are likely to adjust their expectations about the future during periods of financial stress as Kliesen et al. (2012), Abdymomunov (2013), Illing and Liu (2006) and others, point out. To capture changes in inflation expectations, we consider data on break-even inflation rate. The break-even inflation rate is calculated by subtracting the 10-year inflation indexed bond yield from the 10-year nominal bond yield. Liquidity and Safety Lastly, during financial stress periods we observe investors shift their portfolio toward safer and more liquid assets.10 To capture this phenomena in the Australian context, we consider data on changes in the 5-year Commonwealth government bond yield to measure changes medium-term expectations and changes in the 10-year Commonwealth government bond yield to measure changes in long-term expectations. In addition, we include data on the term premium to capture changes in the current monetary policy stance relative to investor’s long-term expectations as in Hakkio et al. (2009). The term premium is calculated by subtracting the 3-month Australia treasury bill yield from the 10-year nominal Commonwealth government bond yield. When monetary policy is tight relative to long-run expectations, short-term yields tend to increase relative to long-term yields, leading to an increase in the spread. Given that we are analyzing a small open economy, we also consider the actual and perceived safety of Australian public debt relative to other foreign public debt. To measure changes in the yields of Australian safe assets relative to yields of other foreign safe assets, we include data on the 5-year real sovereign debt spread between Australia and the United States. This is the case as the United States is one of the most important global financial centers. To do so, we compute the real yield of the 5-year government bond yields in Australia and in the United States by subtracting the inflation rate from the nominal yields. The inflation rate measures the monthly year-on-year change in the CPI of both countries. For Australia, the CPI is only available on a quarterly frequency. To 9Thetradeweightedindexisaweightedbasketof18currencies. Theweightsaredeterminedaccordingtothetrade balance of the respective country with Australia. 10We refer to Hakkio et al. (2009), among others, for more on this topic. 7

obtain monthly frequencies, we interpolate the quarterly data based on a cubic interpolation of the values in neighboring grid points. We then compute the spread according to the following equation y = AUD5Y −UnitedStates5Y ; i,t t t where y denotes the real sovereign debt spread in period t, AUD5Y denotes the real 5-year governt t ment bond yield in period t and US5Y denotes the real 5-year government spread for the United t States Having described all the data, Table 1 depicts descriptive statistics for all of the data series. Variable Mean Standard Deviation Change in cash target rate -0.0112 0.0554 Change in 5-year government bond yield -0.0076 0.0589 Change in 10-year government bond yield -0.0060 0.0503 Term premium 0.8439 1.0826 Inflation expectation 2.8331 1.3999 United States real sovereign debt spread 1.3585 1.5661 3-month TED spread 0.2063 0.2266 3-month BAB spread 0.2131 0.1921 Banking sector volatility 0.0315 0.0041 S&P / ASX 200 return 0.4373 3.9057 S&P / ASX 200 volatility 3.8175 0.8765 Effective exchange rate volatility 1.5792 0.3102 Housing price growth 0.0039 0.0063 Table 1: Descriptive statistics. 3.2 Methodology Having identified several financial measures that can capture disruptions in financial markets when negative shocks hit the economy, we can construct the AFSI. Following the approach used in the literature, we consider the principal component methodology. Various authors have considered different components when constructing the financial stress index. In particular, Hakkio et al. (2009) use the first component, while Park and Mercado Jr (2014) consider the first three components. In this paper we consider the first four components. We do so as it allows us to: (i) increase the share of the total variation of the variables explained by the index to 59.9 percent, and (ii) delivers a more realistic stress measure during the Covid-19 pandemic.11 To construct the AFSI, we first transform each of the data previously described so they have the same units. In particular, for each series we subtract the sample mean and divide by the standard 11LookingatthestandardizedtimeseriessuggeststhatseveralofthevariablesincludedinourAFSImeasureshowed a large change in the beginning of the pandemic. The first four components pick up some these changes in the AFSI estimate that would otherwise be subdued when using only the first-two or the first-three components, respectively. 8

deviation. We then apply the principal component analysis to calculate the coefficients corresponding to these variables. The financial stress index is then constructed by adding the weighted first four components. To do so, we first multiply each component with the standardized time series. We then weigh each component by the eigenvalue of the respective component relative to the sum of the eigenvalues of the first four components and add them to a single index. To compute the adjusted coefficients, we scale the coefficients such that the standard deviation of the index is one. We then weigh the adjusted coefficient of each component for each variable by it’s eigenvalue relative to the sum of the eigenvalues of the first four components. Lastly, we add them up to obtain the adjusted coefficient corresponding to the AFSI. The resulting adjusted coefficients are reported in Table 2. Variable Adjusted coefficient Change in cash target rate -0.1735 Change in 5-year government bond yield -0.0248 Change in 10-year government bond yield 0.0168 Term premium 0.0389 Inflation expectation 0.1278 United States real sovereign debt spread 0.1404 3-month TED spread 0.2578 3-month BAB spread 0.2171 Banking sector volatility 0.1959 S&P / ASX 200 return -0.0874 S&P / ASX 200 volatility 0.3015 Effective exchange rate volatility 0.2901 Housing price growth -0.1700 Percent of total variation of variables explained by AFSI 59.9 Table 2: Adjusted coefficients. Since the coefficients are standardized, they represent the effect of a one standard deviation change of each variable on the AFSI.12 For example, the coefficients in Table 2 highlight that a one standard deviationchangeintheexchangeratevolatilityhasapproximately7.25timestheeffectonthefinancial stress index as a one standard deviation change in the term premium. Moreover, a one standard deviation change in the exchange rate volatility has an impact on the index that is roughly twice as large as the one implied by one standard deviation in inflation expectation changes. We also find that one standard deviation change in the exchange rate, equity market, and the banking sector volatility have the highest impact on financial stress. In contrast, changes in medium to long-term government bond yields have a very small and for the 5-year government bond yield even a negative effect. Next, we report the correlations between each of the first four components and the 13 variables included in the index. These are reported in Table 3. 12We refer to Hakkio et al. (2009) for more on this topic. 9

Variable 1 2 3 4 Change in cash target rate -0.4907 -0.0135 0.2916 -0.5333 Change in 5-year government bond yield -0.6717 0.6403 0.1817 -0.1017 Change in 10-year government bond yield -0.5974 0.6740 0.1417 0.0201 Term premium -0.2339 0.2614 -0.0891 0.3555 Inflation expectation -0.1530 -0.2010 0.7867 0.2491 United States real sovereign debt spread -0.1461 -0.1524 0.7817 0.2553 3-month TED spread 0.7072 0.4626 0.1144 -0.3421 3-month BAB spread 0.7336 0.0978 0.2415 -0.3166 Banking sector volatility 0.2998 0.3590 -0.0127 0.2045 S&P / ASX 200 return -0.1698 -0.0792 -0.1739 0.0758 S&P / ASX 200 volatility 0.4590 0.2571 -0.0311 0.6844 Effective exchange rate volatility 0.4555 0.6130 0.1792 -0.0430 Housing price growth -0.4309 0.1926 -0.4928 0.0656 Table 3: Correlation between the first four components and the 13 variables. As we can see from Table 3, the first component is most heavily correlated with the 3-month TED and the 3-month BAB spread. The second component correlates mostly with exchange rate volatility as well as changes in the 5-year and 10-year government bond yield. The third component is mostly correlated with inflation expectations and the real sovereign debt spread, while the fourth components correlatesmostlywithequityvolatility. Basedontheseresults,weidentifythefirstcomponentasbank funding and the second component as the exchange rate and liquidity components. Furthermore, we identify the third component as the expectations and safety and the fourth component as the equity volatility components. Having identified the first four components, we can generate a time series for the AFSI. By construction, the average value of the index is zero and is associated with normal financial market conditions. Thus, values below zero imply low financial market stress, whereas values above zero suggest higher financial market stress periods. Moreover, larger values of the AFSI indicate greater financial stress and financial disruptions relative to other periods. Figure 1 depicts the resulting AFSI from December 1989 to August 2020. The gray bars indicate Australian recessionary periods. These correspond to two consecutive quarters of negative growth in real GDP per capita, as suggested by Restrepo-Echavarria and Reinbold (2019). As we can see, the AFSI successfully captures the surge in financial stress during the GFC of the 2007-2009 period. Since then the level of financial stress in Australian markets has been higher than in the 1990s. The AFSI also captures the negative impact of the Covid-19 outbreak in Australia, starting in March 2020. The financial stress and disruptions experienced by Australian investors due to Covid-19 in April 2020 has been roughly 2.3 times higher than the financial stress resulting from the Greek debt crisis in June 2010.13 13In Appendix B, we present stress indices where we lag one variable. We present indices for each one of the variables lagged for one period. These indices exhibit a very similar pattern relative to the AFSI presented here. We argue therefore that our index for financial stress would not change materially if we include lags of these variables. 10

6 4 2 0 -2 1990 1995 2000 2005 2010 2015 2020 AFSI, 1990-2020 Figure 1: AFSI (own calculation) for the 1990-2020 period. 4 Evaluating the AFSI We now analyze the properties of AFSI in terms of its external and domestic factors. We also assess the financial stress of the Covid-19 period relative to the global financial crises. Finally, we evaluate the relative performance of our index when compared to other indices. 4.1 Domestic and Foreign Factors GiventhattheUnitedStatesisamajorglobalfinancialcenterandthatChinaisthelargestAustralian trading partner, we consider these two countries when examining the impact of foreign factors in shaping the AFSI. First, we highlight the importance of these countries for Australia by comparing the time series of the AFSI with other financial stress indices for the United States and China. We then further decompose the AFSI. United States To determine the importance of disruptions in one of the most important global financial markets, we compare the AFSI with the Kansas City Fed FSI (KCFSI) and the St. Louis Fed FSI (STLFSI).14 Both US indices apply the principal component analysis and consider just the first component. The 14The STLFSI uses data on a series of interest rates, yields spreads and additional indicators such as different measures of volatility, inflation expectations and data on other financial products. Similarly, the KCFSI includes data on yield spreads and volatility measures. Compared to these indices, we do not include data on corporate bond spreads. InAustralia,thecorporatebondmarketplaysasmallerroleinfirms’financingthanitdoesinanumberofother countries. Since 1990, the share of financial intermediaries in the corporate bond market has increased significantly, making up about 60% of the corporate bond market (Black et al., 2012). 11

main difference between AFSI and the KCFSI/STLFSI is that we include a measure of exchange rate volatility and international public debt spreads. These are arguably important factors when considering small open economies and not global financial centers like the United States. Moreover, the STLFSI includes more information on corporate bond yields. We do not consider corporate bond yields in our measure of financial stress since the number of corporate bonds in Australia is relatively small. Lastly, the KCFSI does not include the term premium and inflation expectations. Figure 2 compares the time series for the AFSI with the two United States indices. The red and grey bars indicate recessionary periods in the United States and in Australia, respectively. 10 8 6 4 2 0 -2 1990 1995 2000 2005 2010 2015 2020 Australia recessions STLFSI US recessions AFSI KCFSI Figure 2: AFSI, KCFSI and STLFSI (Data source: Own calculation, FRED). Broadly speaking, the three indices co-move from December 1989 until March 2020. The STLFSI shows considerably larger financial stress during the financial crises of 2007-2009, when compared to the KCFSI and the AFSI. In addition, from 1998 to 2003 the AFSI shows less financial disruptions than the United States indices. However, the AFSI delivers larger values of financial stress after the the financial crises of 2007-2009. Lastly, the AFSI financial stress measure during the Covid-19 pandemic is somewhat similar to the KCFSI, whereas the STLFSI is somewhat larger. China To capture the potential disruption stemming from the largest Australian trading partner, we now compare the AFSI to a Chinese financial stress developed by the Asia Regional Integration Center of the Asian Development Bank (CHFSI).15 Figure 3 plots the AFSI and the CHFSI where grey bars 15The methodology of the CHFSI is based on the methodology of Park and Mercado Jr (2014). The CHFSI considers the relative volatility of the banking sector relative to equity markets (β, as defined above), changes in equity returns, equity volatility, sovereign debt spreads between the 10-year government bond yield and the 2-year government bond yield and changes in valuation of the domestic currency relative to the United States dollar (Asia Regional Integration Center, 2021). 12

indicate recessions in Australia.16 6 4 2 0 -2 1990 1995 2000 2005 2010 2015 2020 Australia recessions AFSI China FSI Figure 3: AFSI and CHFSI (Data source: Own calculation, Asian Development Bank). As we can see from Figure 3, the AFSI seems to be much less correlated with the CHFSI when compared to the United States indices. This suggests that Australian financial markets are globally connected and less influenced by the financial conditions of its major trading partner. Correlations with Foreign Series To formally asses the co-movement of the US and Chinese indices with the AFSI, we calculate the corresponding correlations. In particular, Table 4 reports the correlation coefficients for our measure of the AFSI, the measure of AFSI obtained from the Asian Development Bank, the KCFSI, the STLFSI and the CHFSI. We also report correlation coefficients with one lag.17 KCFSI STLFSI CHFSI AFSI 0.6590 0.7419 0.3747 AFSI(-1) 0.6303 0.6400 0.3680 Table 4: Correlation of AFSI, KCFSI, STLFSI and CHFSI (Data source: Own calculation, FRED, Asian Development Bank). As we can see from Table 4, the AFSI is more correlated with financial stress in the United States relative to financial stress in China. Among the United States stress indices, the AFSI has the highest correlation with the STLFSI. Domestic and Foreign Decomposition 16Using OECD data on nominal GDP, inflation and population from China, we could not identify recession periods defined as two consecutive quarters of negative real GDP per capita growth for the time period from 1995-2022. 17Correlations reported on Table 4 are at monthly frequency. Thus, the weekly measures for the STLFSI are converted into monthly averages. 13

An alternative way to show the importance of foreign factors in a small open economy is to decompose the AFSI into domestic and foreign components. To do so, we follow the approach taken byMoore(2017). Inparticular, weconsiderfinancialstressindicesfortheUnitedStates, China, Japan and South Korea when decomposing the AFSI.18 First, we focus on the impact of financial stress in the US and China. It is of course plausible that financial stress in Australia can be transmitted from other economies other than the United States and China. Since the United States and China were the major trading partners for the period we are considering, we assume that if financial stress spills over to the Australian economy from abroad, it is most likely that it arises from these economies. The first specification we consider is as follows AFSI = β +β FSI +β FSI +ε ; (1) t 0 1 US,t 2 China,t t where the coefficients β and β measure the contribution of financial stress stemming from the United 1 2 States and China, respectively. The residual ε can be interpreted as the component of financial stress t that does not arise from the United States or China. We refer this as the domestic component. To include additional major trading partners for which financial stress may spillover, we consider the following second specification: AFSI = β +β FSI +β FSI +β FSI +β FSI +ε . (2) t 0 1 US,t 2 China,t 3 Japan,t 4 SouthKorea,t t Our estimates are reported in Table 5. The data covers the period from January 1995 to August 2020. Specifications (i) and (ii) estimate Equation 1, where (i) uses the STLFSI as a measure of financial stress in the US and specification (ii) uses the KCFSI as measure of financial stress in the US. Specification (iii) estimates Equation (2), using the STLFSI for financial stress in the US.19 As we can see from Table 5, the results are similar across the different specifications. Financial stress in Australia exhibits high correlation with financial stress in the US and to a somewhat lower extent with financial stress in China. In particular, a one unit increase in the STLFSI leads to a 0.79 unit increase in the AFSI, whereas a one unit increase in the CHFSI leads to a 0.24 unit increase in the AFSI in specification (iii). Financial stress in Japan and South Korea is not significantly correlated with financial stress in Australia, but including them in the regression yields a higher value of the adjusted R2. Together, financial stress from abroad can explain more than half (57.4%) of the financial stress in Australia. Figure 4 illustrates the results of the decomposition. The top panel shows the AFSI (blue) and the predicted contribution of financial stress arising from the United States, y˜1 = β STLFSI (green). t 1 t The second panel shows the AFSI (blue) and predicted contribution of financial stress emerging from China, y˜2 = β CHFSI (green). The third and fourth panels show the AFSI (blue) and the predicted t 2 t contribution of financial stress in Japan and South Korea, respectively. That is y˜3 = β FSI t 3 Japan,t 18The financial stress indices for Japan and South Korea were obtained from the Asian Development Bank (Asia Regional Integration Center, 2021) and follows the methodology in Park and Mercado Jr (2014). 19Appendix C reports additional regression results. 14

(i) (ii) (iii) Intercept 0.0088 0.0466 0.0007 (0.1016) (0.1323) (0.1155) STLFSI 0.7436∗∗∗ 0.7885∗∗∗ (0.0736) (0.0684) KCFSI 0.6216∗∗∗ (0.1214) CHFSI 0.2196∗∗∗ 0.3638∗∗∗ 0.2441∗∗∗ (0.0826) (0.1128) (0.0916) Japan FSI −0.0645 (0.0645) South Korea FSI 0.0281 (0.0359) Adjusted R2 0.568 0.5015 0.5737 RMSE 0.676 0.8181 0.0117 N 308 308 308 Newey-West standard errors in parentheses. ∗∗∗ denotes statistical significance at the1% level. Table 5: Regression results of the AFSI decomposition. and y˜4 = β FSI in green. Finally, the bottom panel shows the domestic component, t 4 SouthKorea,t which is the residual, y −y˜1−y˜2−y˜3−y˜4. Figure 4 highlights the results from Table 5. Financial t t t t t stress in Australia shows high correlation with US and some correlation with financial stress in China. Financial stress in South Korea and Japan do not seem to contribute to financial stress in Australia markedly. Foreign Component (US) 6 Australia 4 US predicted 2 0 1990 1995 2000 2005 2010 2015 2020 Foreign Component (China) Australia 4 China predicted 2 0 1990 1995 2000 2005 2010 2015 2020 Foreign Component (Japan) Australia 4 Japan predicted 2 0 1990 1995 2000 2005 2010 2015 2020 Foreign Component (South Korea) Australia 4 South Korea predicted 2 0 1990 1995 2000 2005 2010 2015 2020 Domestic Component 2 1 0 -1 1995 2000 2005 2010 2015 2020 Figure 4: Foreign and domestic decomposition (Data source: Own calculation, FRED, Asian Development Bank). 15

4.2 The Covid-19 Pandemic To capture the financial stress during the Covid-19 pandemic, we report the daily new confirmed Covid-19 cases in Australia as well as four measures of financial stress, namely the KCFSI, STLFSI, the FSI for Australia produced by the Asian Development Bank and our measure of the AFSI.20 These different indices are depicted in Figure 5 together with the number of daily confirmed cases of Covid-19 in Australia. 6 800 4 600 2 400 0 200 -2 0 Oct 2019 Jan 2020 Apr 2020 Jul 2020 Daily new confirmed cases AFSI (own calculation) STLFSI AFSI (Asian Development Bank) KCFSI Figure 5: Financial stress indices and daily confirmed cases of Covid-19 (Data source: Own calculation, FRED, Asian Development Bank, Australian Department of Health). As we can see from Figure 5, with the exception of the FSI produced by the Asian Development Bank, the indices show an increase in financial stress that coincides with the first wave of Covid-19 and a subsequent decrease to average levels of financial stress after that. The FSI obtained from the Asian Development Bank, however shows an sharp increase in financial stress only at the end of the first wave. 4.3 Comparison with Other Indices Australian Economic Policy Uncertainty Index We now consider the Australian economic policy uncertainty index (EPU) by Baker et al. (2016) as a broader alternative measure of financial stress. Figure 6 shows the AFSI relative to the EPU index for a period from January 1998 to August 2020.21 In the following discussion, we also refer to Moore (2017) to identify specific Australian events. In particular, the letters (a) through (f) correspond to the following episodes of large increases of the economic policy uncertainty: (a) Close Australian election fought over goods and services tax introduction and Russian economic crisis; (b) the 9/11 20The Covid-19 case data is obtained through the Australian Department of Health. 21The data is taken from www.policyuncertainty.com 16

attacks; (c) the invasion of Iraq; (d) the great financial crisis; (e) Greek debt crisis, mining tax and carbon policy uncertainty; (f) United States debt ceiling dispute; (g) the Brexit vote, and (h) the beginning of the Covid-19 pandemic. 6 400 d f g h 4 300 b e c a 2 200 0 100 -2 0 2000 2005 2010 2015 2020 AFSI EPU Index Figure 6: AFSI and economic policy uncertainty index (Data source: Own calculation, www.policyuncertainty.com). As we can see from Figure 6, our estimates of heightened financial stress periods overlap with some periods of large increases in the Australian economic policy uncertainty. These instances are the great financial crisis (d) and the Covid-19 pandemic (h). For other periods of high economic policy uncertainty, such as the close Australian election fought over goods and services taxes (a), the Greek debt crisis and mining and carbon policy uncertainty (e) and the United States debt ceiling dispute (f), the AFSI shows somewhat elevated levels of financial stress . Lastly, some periods of higher economic policy uncertainty such as the Brexit vote (g), the 9/11 attacks (b), the invasion of Iraq (c) do not correspond to periods of higher financial stress as captured by the AFSI. Asian Development Bank Index Using Park and Mercado Jr (2014) methodology, the Asia Regional Integration Center of the Asian Development Bank tracks financial stress across several countries. In the case of Australia, the Asian Development Bank uses data on the relative volatility of the banking sector relative to equity markets (β, as defined above), changes in equity returns, equity volatility, sovereign debt spreads between the 10-year government bond yield and the 2-year government bond yield and changes in valuation of the domestic currency relative to the United States dollar (Asia Regional Integration Center, 2021). From now on we refer this index as AFSI-ADB. It is important to highlight that our measure of Australian financial stress index (AFSI) is likely to differ, since we include additional data. Figure 7 depicts the time series of these two financial stress indices. 17

10 8 6 4 2 0 -2 1990 1995 2000 2005 2010 2015 2020 AFSI (own calculation) AFSI (Asian Development Bank) Figure 7: AFSI and AFSI-ADB by Asian Development Bank (Data source: Own calculation, Asian Development Bank). As we can see from Figure 7 and Table 6, our measure of the Australian financial stress is highly correlatedwiththemeasureoffinancialstressinAustraliadevelopedbytheAsianDevelopmentBank. It is worth noting, however, that our measure imply lower financial stress during the 1995-1996, 2002- 2003 and 2016-2017 periods. Moreover, in contrast to our AFSI, the Asian Development Bank index suggests that the financial stress during the Covid-19 pandemic is larger than the one experienced during the Global Financial Crisis.22 AFSI-ADB AFSI 0.64921 AFSI(-1) 0.6604 Table 6: Correlation between AFSI-ADB and AFSI and AFSI-ADB and lagged AFSI (Data source: Own calculation, Asian Development Bank). 5 The AFSI and Macroeconomic Outcomes One of the advantages of measuring financial stress is that can capture disruptions in the normal functioning of financial markets that can lead to adverse real economic outcomes. This can occur in several ways. In times of increased uncertainty, firms and households may delay or reduce hiring, 22In Appendix D, we discuss the differences in our estimate of financial stress and the financial stress index from the Asian Development Bank in more detail. 18

investment, and spending. It is not too surprising then that financial stress, as measured by FSI, can forecast declines in economic activity.23 We now examine the relationship between AFSI and various measures of economic activity. Finally, we also explore how AFSI shocks may affect and propagate in the economy. 5.1 Granger Causality Tests Financial stress episodes are frequently connected with economic downturns. This is the case as they destabilize the financial system and hinder its ability to operate smoothly. Being able to asses whethertheAFSIcanbeanearlywarningsystemand/orhelpimproveforecastsofkeymacroeconomic indicators is of paramount importance. Next, we evaluate these possibilities by considering real GDP per capita growth, bank credit growth and the unemployment rate. Before doing so, Table 7 reports the different correlations between the AFSI and these observables. (1) (2) (3) (4) (5) GDP per capita Bank credit UR UR UR t t+15 t+37 −0.3845 −0.2967 −0.0455 0.1550 0.1932 Table 7: Correlation between the AFSI, GDP per capita growth, bank credit growth and the unemployment rate (Data source: Own calculation, Australian Bureau of Statistics, Reserve Bank of Australia). Unsurprisingly, Table 7 highlights that real GDP per capita growth is negatively correlated with our measure of financial stress. This indicates that periods of higher financial stress tend to be accompanied with lower real GDP per capita growth. Column (2) shows that the AFSI is negatively correlated with bank credit growth, indicating that bank credit growth tends to be lower in periods of financial stress. Finally, columns (3) through (5) in Table 7, we report the correlation between the AFSI and the unemployment rate (UR) for various time periods. The contemporaneous correlation with unemployment at t is negative. In contrast, the 15 and 37 months ahead unemployment rate relativetotheAFSIatperiodt, ispositive. Thiscorrelationincreaseswiththe numberofunemployment lags. GDP We now examine the information content of the AFSI when it comes to aggregate economic performance. Figure 8 plots real GDP per capita growth and our AFSI from January 1990 to June 2020, where the grey bars indicate recessions in Australia.24 As we can see from Figure 8, our measures of higher than average financial stress periods in 1990/91, 2008, 2020 coincide with periods of lower real GDP per capita growth. However, we also 23We refer to Apostolakis and Papadopoulos (2019) for more on this topic. 24Since GDP is only available at a quarterly frequency, we take the mean of the values of the AFSI of each month within a quarter to create a quarterly time series of the AFSI. 19

6 4 5 2 4 e 3 0 g n a h IS F 2 -2 C e g A 1 a tn e c 0 -4 re P -1 -6 -2 -3 -8 1990 1992 1995 1997 2000 2002 2005 2007 2010 2012 2015 2017 2020 AFSI Australia Real GDP Per Capita Growth Figure 8: AFSI and real GDP per capita growth (Data source: Own calculation, Australian Bureau of Statistics). find periods of lower real GDP per capital growth where we do not see a corresponding increase in the financial stress index. Such instances are the recessionary periods in 2000/01, 2006 and 2018/19. Next, we run a Granger causality tests to determine whether including measures of the AFSI can improve the forecasts of real GDP per capita growth relative to a forecast that only includes its past values. The Null hypothesis is that the AFSI does not Granger-cause real GDP per capita growth. We fit a VAR(4) model to real GDP per capita growth and our financial stress index, where the number of lags was chosen based on the AIC criterion and the maximum number of lags was set to 4. The Granger causality tests are reported in Table 8. F-statistic p-value Exclude lagged AFSI in real GDP per capita growth equation 1.1214 0.2896 Exclude lagged real GDP per capita growth in AFSI equation 0.4742 0.4911 Table 8: Granger causality test for AFSI and real GDP per capita growth. Our findings suggest that we cannot reject the null hypothesis that the AFSI does not Grangercause real GDP per capita growth. Thus, our measure of the AFSI cannot improve forecasts of real GDP per capita growth relative to forecasts based on past realizations of real GDP per capita growth. One possible explanation for this finding is that in the case of Australia, an increase in financial stress does not always coincide with negative real GDP per capita growth, as is evident by Figure 1 in the period of the financial crisis in 2008. In addition, given that GDP data is quarterly monthly temporary financial disruptions may not be fully captured. Appendix E provides some robustness Granger tests for the AFSI, the underlying variables used to construct the AFSI and real GDP per capita growth as well as other credit measures. 20

Retail sales growth To have more information on economic activity at a higher frequency, we consider monthly retail sales growth data.25 To capture the information content of the AFSI, we estimate whether including the AFSI can improve the prediction of retail sales growth relative to using only previous values of retail sales growth. To do so we fit a VAR(4) model and conduct a Granger causality test.26 The results are reported in Table 9. F-statistic p-value Exclude lagged AFSI in retail sales growth equation 12.947 0.0115 Exclude lagged retail sales growth in AFSI equation 3.6864 0.4501 Table 9: Granger causality test for AFSI and real GDP per capita growth. Our results indicate that we can reject the null hypothesis that the AFSI does not Granger causes retailsalesgrowthatthe5%significancelevel. Thus, wecanconcludethattheincludingtheAFSIcan improve predictions for retail sales growth. This suggests that the AFSI captures relevant information that is key for retail sales. Loans Another measure of economic activity that is widely used when assessing the performance of an economy is the volume of bank loans. During periods of financial stress and economic downturns, we expect bank credit growth to be lower relative to normal times. Figure 9 depicts the AFSI and monthly bank credit growth data.27 As we can see from Figure 9, the negative correlation is most evident in November 2008 and during the recessions of the 1990s. We, however, also observe periods without a negative correlation. For instance, during the onset of the Covid-19 pandemic, bank credit growth seems to have increased even though this was also a period of higher financial stress. One possible explanation for such anomaly is that the Reserve Bank of Australia implemented several relief programs in order to support bank lending. Next, we perform the Granger causality test to determine if including financial stress, as measured by the AFSI, can improve forecasts of bank credit growth. We do so by fitting a VAR(4) model to the data, which is reported on Table 10.28 As we can see from Table 10, we can reject the null hypothesis that the AFSI does not Granger cause bank credit growth at the 1% significance level. Thus, we conclude that including the AFSI can improve forecasts of bank credit growth. Once again, this finding suggests that the AFSI captures relevant information that is key for the credit growth in an economy. Appendix E provides more Granger causality tests on the AFSI and other measures of credit. 25This data is taken from the Reserve Bank of Australia 26The number of lags was again selected based on the AIC criterion and the maximum number of lags was set to 4. 27The bank data is taken from the Reserve Bank of Australia. 28Again, the number of lags was selected with the AIC criterion and the maximum number of lags was set to 4. 21

6 2% 5 1.5% 4 1% 3 2 0.5% 1 0% 0 -0.5% -1 -2 -1% 1990 1995 2000 2005 2010 2015 2020 AFSI Bank Credit Growth Figure 9: AFSI and bank credit growth (Data source: Own calculation, Reserve Bank of Australia). F-statistic p-value Exclude lagged AFSI in bank credit growth equation 34.861 0.0000 Exclude lagged bank credit growth in AFSI equation 2.1248 0.7128 Table 10: Granger causality test for AFSI and bank credit growth. Unemployment Lastly, we look how our measure of financial stress compares to the unemployment rate.29 Figure 10 depicts the monthly unemployment rate in Australia together with the AFSI. Since GDP growth and unemployment tend to be negatively correlated and we have shown that our AFSI is negative correlated with GDP per capita growth, we expect a positive correlation. As we can see from Figure 10, periods of higher unemployment tend to follow periods of financial stress with a lag. Table 11 reports the Granger causality test for our AFSI measure and the unemployment rate. To do so, we fit a VAR(14) model and test the hypothesis that the AFSI does not Granger cause unemployment.30 F-statistic p-value Exclude lagged AFSI in unemployment rate equation 28.7975 0.0111 Exclude lagged unemployment rate in AFSI equation 25.823 0.0273 Table 11: Granger causality test for AFSI and unemployment rate. Our finding suggest that we can reject the null hypothesis that the AFSI does not Granger cause 29The unemployment data is taken from the Australian Bureau of Statistics. 30The lag was again selected with the AIC criterion. The maximum number of lags was set equal to 40 to account for large lags visible in the data. 22

6 10 5 9 4 8 e 3 g n a h IS 2 7 C e F g A 1 6 a tn e c 0 re 5 P -1 4 -2 -3 3 1990 1995 2000 2005 2010 2015 2020 AFSI Unemployment rate Figure 10: AFSI and unemployment rate (Data source: Own calculation, Australian Bureau of Statistics). unemployment. Moreover, we also find that we can reject the null hypothesis that unemployment Granger causes financial stress. These results indicate a feedback loop and we therefore cannot argue that using the AFSI to predict unemployment does not provide a better forecast compared to a prediction using only passed values of unemployment. 5.2 Responses to an AFSI shock In this section we estimate the effect of a financial stress shock, as measured by an increase in AFSI, to the unemployment rate, growth of bank loans, retail sales growth as a proxy for the real GDP per capita growth and the growth of the central bank balance sheet as a measure of monetary policy. In order to do so, we utilize a time-varying parameter (TVP) VAR model with stochastic volatility developed by Primiceri (2005) and Nakajima (2011). The TVP-VAR model has an advantage over the constant parameter VAR models in that it does not need to arbitrarily divide data into sub-samples to identify the change of structure of the model. Instead, the model allows both time variation in the simultaneous relations among variables. This can occur due to variations in the structural dynamic equations relating the various macroeconomic aggregates. It also allows heteroskedasticity in the innovations, which can be driven by changes in the size of exogenous shocks or their impact on macroeconomic variables. A TVP-VAR model is estimated using monthly data from January 1990 to August 2020. The basic structural model is given by A Y = F Y +...+F Y +u ; (3) t t 1,t 1,t−1 s,t s,t−s t 23

where Y is a vector that includes the AFSI, growth of central bank balance sheet as a measure of t monetary policy, credit growth, change of the unemployment rate and retail sales growth as a proxy for the real GDP per capita growth.31 A ,F ,...F represent 5×5 matrices of coefficients that we t 1,t s,t are going to estimate. Finally, u is a 5 × 1 structural shock where u ∼ N(0,Σ Σ ). We specify t t t t the simultaneous relations of the structural shocks by recursive identification, assuming that A is t lower-triangular. More precisely, we have the following structure     σ 0 0 0 0 1 0 0 0 0 1,t  0 σ 0 0 0  a 1 0 0 0  2,t   21,t      Σ =  0 0 σ 0 0  & A = a a 1 0 0. t 3,t t 31,t 32,t      0 0 0 σ 0  a a a 1 0  4,t   41,t 42,t 43,t  0 0 0 0 σ a a a a 1 5,t 51,t 52,t 53,t 54,t Note that equation (3) can be rewritten as follows (cid:79) Y = X β +A−1Σ ε where X = I (Y ,...,Y ) and ε ∼ N(0,I ). t t t t t t t 5 t−1 t−s t 5 Following Primiceri (2005), let a = (a ,a ,...,a )(cid:48) be a stacked vector of the lower-triangular t 21,t 31,t 54,t elements in A , and h = (logσ2 ,...,logσ2 )(cid:48). We further assume that the time-varying parameters t t 1,t 5,t follow a random walk process as follows β = β +u , a = a +u , h = h +u ; t+1 t βt t+1 t at t+1 t ht where the disturbances have the following structure      ε I O O O t      u O Σ O O  βt   β    ∼ N 0, . u   O O Σ O  at   a  u O O O Σ ht h We estimate the model using the Bayesian methods described as in Nakajima (2011) and choose the number of the VAR lags to be one. Stochastic Volatility of the AFSI First, we estimate the stochastic volatility of a financial stress shock from the period of January 1990 to August 2020. Figure 11 depicts our estimate and the associated standard deviation over time. The stochastic volatility has two episodes of increased volatility. One is during GFC in 2008 and at the beginning of 2020, right at the beginning the Covid-19 pandemic. The first spike in 2008 was followed by a period of higher than average volatility before dropping to pre-GFC period levels. 31We adjust central bank asset growth by retail sales growth in order to measure growth in central bank assets that is in excess of short-term fluctuations in economic activity. 24

2.5 Posterior mean 1SD bands 2 1.5 1 0.5 0 Feb 94 April 98 June 02 Aug 06 Oct 10 Dec 14 Feb 14 Figure 11: Posterior Estimate for Stochastic Volatility of Financial Stress Shock. Impulse Responses Next, we compute impulse response functions and compare the results to a financial stress shock for the periods in November 2005, November 2008 and March 2020.32 We chose November 2005 to capture an environment before the global financial crises, where the Australian economy did not experience large disturbances. Not surprisingly, our measure of the AFSI exhibits low values of financial stress. To highlight the impact of financial disturbances and higher financial stress, we compare it to November 2008 and March 2020. The results are depicted in Figure 12. growth of central bank balance sheet growth of bank loans 1 0.05 11/2005 11/2005 0.8 11/2008 11/2008 3/2020 3/2020 0.6 0 0.4 0.2 0 -0.05 0 5 10 15 0 5 10 15 c1h0a-3nge of undemployment rate growth of retail sales 10 0.1 11/2005 11/2005 8 11/2008 0.05 11/2008 3/2020 3/2020 6 0 4 -0.05 2 -0.1 0 -0.15 0 5 10 15 0 5 10 15 Figure 12: Impulse Responses to a Financial Stress Shock. 32Our results are robust to including two or three lags. 25

The first thing to note is that a financial stress shock has different effects to the growth of the centralbank’sbalance sheet, creditgrowth, changesin theunemployment rateandretailsales growth. This depends on whether the economy is in a state of low financial stress or in a period of heightened financial stress. An increase in financial stress shock leads to an initial increase in the growth of the central banks balance sheet, followed by a gradual and significant decrease to negative values, and ultimately a return to zero. Similarly, credit growth increases initially as an increase of financial stress and then decreases over time. However, the magnitude of the increase in credit growth is lower during periods of low financial stress, and it eventually converges to zero. In contrast, during periods of high financial stress, credit growth experiences a sharper decrease after the initial increase, eventually leading to negative values, before converging back to zero. Moreover, the increase in the unemployment rate is less pronounced during periods of high financial stress than during periods of low financial stress. Lastly, an increase in financial stress initially results in a decline in retail sales growth, followed by a subsequent recovery over time. Specifically, during a period of low financial stress, retail sales growth surpasses the initial level before declining again. growth of central bank balance sheet growth of bank loans 0.15 0.04 1 month 0.1 1 quarter 0.02 1 year 0.05 0 0 -0.02 1 month -0.05 -0.04 1 quarter 1 year -0.1 -0.06 1990 1995 2000 2005 2010 2015 2020 1990 1995 2000 2005 2010 2015 2020 c1h0a-3nge of undemployment rate growth of retail sales 15 0.1 1 month 10 1 quarter 0.05 1 year 0 5 -0.05 1 month 0 -0.1 1 quarter 1 year -5 -0.15 1990 1995 2000 2005 2010 2015 2020 1990 1995 2000 2005 2010 2015 2020 Figure 13: Time-varying Responses to a Financial Stress Shock for the Whole Sample Next, we examine the iso-interval time-varying impulse responses for the whole sample period. Figure 13 illustrates the effect on central bank balance sheet growth, credit growth, change in the unemployment rate and retail sales growth one month after the shock (green), one quarter after the shock (blue) and one year after the shock (red) varies over time. Generally speaking, the effect of an increase in financial stress is larger in the short-run (one-month/ one quarter) compared to the one-year horizon. These results also highlight the importance of higher frequency data for a financial 26

stress index. The growth of bank loans and unemployment rate are found to be sensitive to the level of financial stress. Specifically, high financial stress periods, such as the global financial crisis in 2008 and the onset of the pandemic in early 2020, are associated with a decrease in the growth of bank loans and a spike in unemployment rate. Conversely, during low financial stress periods, bank loans tend to increase and unemployment rate tends to decrease. The central bank balance sheet growth responds positively to a financial stress shock across all time horizons, while retail sales growth exhibits a consistently negative response. However, the magnitude of these impacts tends to weaken over time, with the effect becoming negligible after a year from the initial shock. 6 Conclusion Theory and empirical findings have emphasized the strong and negative connection between stress episodes on the financial markets and financial and macroeconomic stability as well as their adverse impact on overall economic activity. Thus, it is important to continuously develop and improve tools for the timely capture of financial market disruptions. Within this spirit, in this paper we develop Australian Financial Stress Index (AFSI) that is based on monthly data on interest rate, spreads, volatility measures, exchange rates, housing price growth and inflation expectations. We find that the first four AFSI principal components are private bank’s funding cost (first component), the safe and liquid asset and exchange rate (second component), the inflation expectation and the United States real sovereign debt spread (third component) as well as equity volatility (fourth component). A decomposition of the AFSI into foreign and domestic factors shows that more than half of financial stress in Australia can be attributed to financial stresses arising from external factors. We also show that the AFSI has relevant information content that might be of interest to both investors and policymakers. In particular, we find that the AFSI can improve forecasts for bank credit growth and retail sales growth relative to forecasts that only rely on past data. In our TVP-VAR analysis, we further show that financial stress can have non-linear effects on important macroeconomic aggregates. In particular, an increase in financial stress shock has more adverse effects on bank credit growth if disruption in financial markets, as measured by the AFSI, is high. The various findings regarding the information content of the AFSI further highlights the importance and usefulness of having a measure of financial stress that can be used by policy makers as timely signal of future economic activity. References Azamat Abdymomunov. Regime-switching measure of systemic financial stress. Annals of Finance, 9(3):455–470, 2013. Antonio Afonso, Jaromir Baxa, and Michal Slav´ık. Fiscal developments and financial stress: a threshold VAR analysis. Empirical Economics, 54(2):395–423, 2018. 27

George Apostolakis and Athanasios P Papadopoulos. Financial stability, monetary stability and growth: a PVAR analysis. Open Economies Review, 30(1):157–178, 2019. Asian Development Bank Asia Regional Integration Center. Data and methodology, financial stress index. 2021. URL https://aric.adb.org/database/fsi. Accessed on September 13th 2021. Scott R Baker, Nicholas Bloom, and Steven J Davis. Measuring economic policy uncertainty. The quarterly journal of economics, 131(4):1593–1636, 2016. RaviBalakrishnan, StephanDanninger, SelimElekdag, andIrinaTytell. Thetransmissionoffinancial stress from advanced to emerging economies. Emerging Markets Finance and Trade, 47(sup2):40– 68, 2011. Susan Black, Joshua Kirkwood, Alan Rai, and Thomas Williams. A history of Australian corporate bonds. Reserve Bank of Australia Research Discussion Paper RDP 2012-09, 2012. Roberto Cardarelli, Selim Ali Elekdag, and Subir Lall. Financial stress, downturns, and recoveries. IMF Working Papers No. 09/100, pages 1–58, 2009. Troy Davig, Craig Hakkio, et al. What is the effect of financial stress on economic activity. Federal Reserve Bank of Kansas City, Economic Review, 95(2):35–62, 2010. Jan J Groen, Michael Nattinger, and Adam I Noble. Measuring global financial market stresses. FRB of New York Staff Report, (940), 2020. Craig S Hakkio, William R Keeton, et al. Financial stress: what is it, how can it be measured, and why does it matter? Economic Review, 94(2):5–50, 2009. Elke Hanschel, Pierre Monnin, et al. Measuring and forecasting stress in the banking sector: evidence from Switzerland. BIS papers, 22:431–449, 2005. Luke Hartigan and Michelle Wright. Financial conditions and downside risk to economic activity in Australia. Reserve Bank of Australia Research Discussion Paper 2021-03, 2023. JanHatzius,PeterHooper,FredericSMishkin,KermitLSchoenholtz,andMarkWWatson. Financial conditions indexes: A fresh look after the financial crisis. NBER Working Paper 16150, 2010. Mark Illing and Ying Liu. Measuring financial stress in a developed country: An application to Canada. Journal of Financial Stability, 2(3):243–265, 2006. KevinLKliesenandDouglasCSmith. Measuringfinancialmarketstress. Economic Synopses, Federal Reserve Bank of St. Louis, 2010. Kevin L Kliesen, Michael T Owyang, and E Katarina Vermann. Disentangling diverse measures: A survey of financial stress indexes. Federal Reserve Bank of St. Louis Review, 94(5):369–397, 2012. Luc Laeven and Fabian Valencia. Systemic banking crises database. IMF Economic Review, 61(2): 225–270, 2013. Angus Moore. Measuring economic uncertainty and its effects. Economic record, 93(303):550–575, 2017. J. Nakajima. Time-varying parameter VAR model with stochastic volatility: An overview of methodology and empirical applications. Monetary and Economic Studies, 29:107–142, 2011. 28

Mikhail V Oet, Timothy Bianco, Dieter Gramlich, and Stephen J Ong. Financial stress index: A lens for supervising the financial system. FRB of Cleveland Policy Discussion Paper, (12-37), 2012. Patrice Ollivaud and David Turner. The effect of the global financial crisis on oecd potential output. OECD Journal: Economic Studies, 2014(1):41–60, 2015. doi: 10.1787/eco studies-2014-. Cyn-Young Park and Rogelio V Mercado Jr. Determinants of financial stress in emerging market economies. Journal of Banking & Finance, 45:199–224, 2014. GiorgioE.Primiceri. Timevaryingstructuralvectorautoregressionsandmonetarypolicy. The Review of Economic Studies, 72(3):821–852, 2005. Paulina Restrepo-Echavarria and Brian Reinbold. Has Australia really had a 28-year long expansion? St. Louis Fed On the Economy Blog, 2019. URL https://www.stlouisfed.org/on-the-economy/ 2019/september/australia-28-year-expansion. Accessed on March 2nd 2021. Bjo¨rn van Roye. Financial stress and economic activity in germany. Empirica, 41(1):101–126, 2014. 29

Appendix A Data Series and Sources Data Series Data Source AsianDevelopmentBankFSIAustralia AsiaRegionalIntegrationCenter,AsianDevelopmentBank, https://aric.adb.org/database/fsi,accessedonSeptember132021. ASX200bankingindex GlobalFinancialData: GlobalFinancialData AXBAJD ASX200compositeindex GlobalFinancialData: GlobalFinancialData AXJOD AustraliaCPI ReserveBankofAustralia: ConsumerPriceInflationG1-GCPIAG AustralianGDP ReserveBankofAustralia: GrossDomesticProductandIncomeH1- GGDPCVGDP Australian3-monthinterbankrate(BBSW) GlobalFinancialData: GlobalFinancialData IBAUnitedStates3D Australian3-monthtreasurybillyield GlobalFinancialData: GlobalFinancialData ITAUnitedStates3D BAB3-month ReserveBankofAustralia: InterestRatesandYields MoneyMarket- Monthly-F1.1FIRMMBAB90 Broadmoneygrowth ReserveBankofAustralia: MonetaryAggregates- D3-DMABMN CashRateTarget ReserveBankofAustralia: InterestRatesandYields MoneyMarket- Monthly-F1.1-FIRMMCRI Centralbankassets ReserveBankofAustralia: LiabilitiesandAssets Detailed A1.1-ARBAATAW CommonwealthGovernment5-yearbondyield ReserveBankofAustralia: CapitalMarketYields GovernmentBonds-Monthly-F2.1-FCMYGBAG5 CommonwealthGovernment10-yearbondyield ReserveBankofAustralia: CapitalMarketYields GovernmentBonds Monthly F2.1-FCMYGBAG10 Credit ReserveBankofAustralia: LendingandCreditAggregates-TableD2 Creditgrowth ReserveBankofAustralia: GrowthinSelectedFinancialAggregates-D1-DGFACM DailyconfirmedCovid-19casesinAustralia OurWorldinData https://ourworldindata.org/explorers/coronavirus-data-explorer Economicpolicyuncertaintyindex https://www.policyuncertainty.com/, accessedonSeptember162021. Effectiveexchangerate ReserveBankofAustralia: ExchangeRates-DailyandMonthly-F11- FXRTWI FinancialStressIndexforChina AsiaRegionalIntegrationCenter,AsianDevelopmentBank, https://aric.adb.org/database/fsi,accessedonSeptember132021. Table 12: Data Series and Sources 1/2 30

Data Series Data Source Financial Stress Index for Japan Asia Regional Integration Center, Asian Development Bank, https://aric.adb.org/database/fsi, accessed on October 20 2022. Financial Stress Index for South Korea Asia Regional Integration Center, Asian Development Bank, https://aric.adb.org/database/fsi, accessed on October 20 2022. Housing price growth Sirca: Sirca CoreLogic Hedonic Home Value 8 Combined Capital Cities33 Inflation indexed 10-year bond yield Global Financial Data: Global Financial Data IGAUnited States ID Kansas City Fed Financial Stress idex FRED, https://fred.stlouisfed.org/series/KCFSI, accessed September 13 2021. Population in Australia Australian Bureau of Statistics https://www.abs.gov.au/statistics/people/population Retail sales growth Reserve Bank of Australia: Monthly activity indicators H3 - GISSRTCYP St. Louis Fed Financial Stress Index FRED, https://fred.stlouisfed.org/series/STLFSI2, accessed September 13 2021. United States 5-year government bond yield FRED, Federal Reserve Bank of St. Louis: 5-Year Treasury Constant Maturity Rate [DGS5], https://fred.stlouisfed.org/series/DGS5, accessed September 13, 2021. United States CPI OECD (2021), Inflation (CPI) (indicator). doi: 10.1787/eee82e6e-en, accessed on March 16, 2021. Table 13: Data Series and Sources 2/2 33Sirca data was obtained by Pedro Gomis-Porqueras and Xuan Zhou under the purview of Deakin University licenses. The remaining co-author, Romina Ruprecht, did not have any unauthorized access to this data while working on this paper. 31

B Robustness tests with lagged variables To test the robustness of the AFSI, we generate different stress indices, each with one variable lagged by one period. Figure 14 depicts the financial stress indices, each with one variable lagged for one period. Each of these indices exhibit a very similar pattern across time relative to the AFSI. 6 6 6 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 -1 -1 -1 - 1 2 990 1995 2000 2005 2010 2015 2020 - 1 2 990 1995 2000 2005 2010 2015 2020 -2 1990 1995 2000 2005 2010 2015 2020 A A F F S S I I with lagged cash rate A A F F S S I I with lagged 5-year commonwealth bond yield AFSI AFSI with lagged 10 year commonwealth bond yield 7 6 7 6 5 6 5 4 5 4 3 4 3 3 2 2 2 1 1 1 0 0 0 -1 -1 -1 - 1 2 990 1995 2000 2005 2010 2015 2020 - 1 2 990 1995 2000 2005 2010 2015 2020 - 1 2 990 1995 2000 2005 2010 2015 2020 A A F F S S I I with lagged real sovereign debt spread A A F F S S I I with lagged term premium A A F F S S I I with lagged inflation expectations 6 6 6 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 -1 -1 -1 -2 -2 -2 1990 1995 2000 2005 2010 2015 2020 1990 1995 2000 2005 2010 2015 2020 1990 1995 2000 2005 2010 2015 2020 AFSI AFSI AFSI AFSI with lagged TED spread AFSI with lagged BAB spread AFSI with lagged banking sector volatility 6 6 7 5 6 5 4 5 4 3 4 3 2 3 2 2 1 1 1 0 0 0 -1 -1 -1 - 1 2 990 1995 2000 2005 2010 2015 2020 - 1 2 990 1995 2000 2005 2010 2015 2020 - 1 2 990 1995 2000 2005 2010 2015 2020 AFSI AFSI AFSI AFSI with lagged equity return AFSI with lagged equity volatility AFSI with lagged housing growth 7 6 5 4 3 2 1 0 -1 -2 1990 1995 2000 2005 2010 2015 2020 AFSI AFSI with lagged exchange rate volatility Figure 14: AFSI with lagged variables 32

C Decomposition robustness tests Here, we provide some robustness tests for our decomposition results. In addition to financial stress indices from the US and China, we also include indices from Japan and South Korea (as computed by the Asian Development Bank) in our regression. Thus, we estimate the following regression: AFSI = β +β FSI +β FSI +β FSI +β FSI +ε ; (4) t 0 1 US,t 2 China,t 3 Japan,t 4 SouthKorea,t t where the coefficients β and β measure the contribution of financial stress stemming from the United 1 2 States and China, respectively and the coefficients β and β measure the contribution of financial 3 4 stress from Japan and South Korea, respectively. The residual ε can be interpreted as the component t of financial stress that does not arise from the United States, China, Japan and South Korea. We refer this as the domestic component. The data covers a period from January 1995 to August 2020. Our results are reported in Table 14. Regression (i) and (ii) report the results using the stress indices from the US and China, where (i) includes the STLFSI and (ii) includes the KCFSI as in the main text. Regression (iii) includes the stress index for Japan and Regression (iv) includes the stress index for South Korea. Regression (v) includes indices for the US, China, Japan and South Korea. Specifications (i), (ii) and (v) are presented in the main text. We replicate them here for convenience. As Table 14 shows, adding the financial stress index for Japan or South Korea individually does not materially improve the results relative to specification (i), as both the adjusted R2 is lower and the RSME is higher in those specifications. (i) (ii) (iii) (iv) (v) Intercept 0.0088 0.0466 0.0012 0.0089 0.0007 (0.1016) (0.1323) (0.1198) (0.1118) (0.1155) STLFSI 0.7436∗∗∗ 0.7948∗∗∗ 0.7419∗∗∗ 0.7885∗∗∗ (0.0736) (0.0734) (0.0752) (0.0684) KCFSI 0.6216∗∗∗ (0.1214) CHFSI 0.2196∗∗∗ 0.3638∗∗∗ 0.228∗∗∗ 0.2204∗∗∗ 0.2441∗∗∗ (0.0826) (0.1128) (0.0901) (0.0854) (0.0916) Japan FSI −0.0448 −0.0645 (0.051) (0.0645) South Korea FSI 0.0018 0.0281 (0.021) (0.0359) Adjusted R2 0.568 0.5015 0.5724 0.5666 0.5737 RMSE 0.1538 0.8181 0.0212 0.1569 0.0117 Number of observations 308 308 308 308 308 Newey-West standard error are in parantheses. ∗∗∗ denotes statistical significance at the 1% level Table 14: Regression results of the AFSI decomposition. 33

D Robustness tests with the financial stress for Australia from the Asian Development Bank In this section, we test how our estimate of financial stress in Australia compares to the financial stress index for Australia from the Asian Development Bank. Table 15 shows the correlations of the Australian financial stress index from the Asian Development Bank (ADB AFSI) with real GDP per capita, bank credit and unemployment including lags of 15 months and 37 months. (1) (2) (3) (4) (5) real GDP per capita Bank credit UR UR UR t t+15 t+37 0.0882 −0.3054 −0.1019 −0.1184 −0.1727 Table 15: Correlation between the ADB AFSI, GDP per capita growth, bank credit growth and the unemployment rate Compared to our measure of financial stress in Australia, the ADB FSI exhibits a lower correlation with real GDP per capita growth GDP and bank credit growth, whereas the correlation with contemporaneous unemployment is higher. Moreover, the ADB FSI seems to exhibit a positive correlation with GDP and a negative correlation with unemployment across all lags, whereas our estimate of the AFSI exhibits negative correlation with real GDP per capita growth and a positive correlation with unemployment after 15 months. Intuitively, we would expect that financial stress that affects the economy negatively would be negatively correlated with real GDP per capita growth and positive correlated with unemployment. Table 16-19 show Granger causality tests for the ADB with real GDP per capita growth, bank credit, unemployment and retail sales growth. Lags were chosen based on the AIC criterion, where the maximum number of lags was set to 4 for real GDP per capita growth, credit growth and retail sales growth and to 40 for the unemployment rate. F-statistic p-value Exclude lagged ADB FSI in real GDP per capita growth equation 0.7683 0.6810 Exclude lagged real GDP per capita growth in AFSI equation 5.6738 0.0581 Lags: 2. Table 16: Granger causality test for the ADB FSI and real GDP per capita growth F-statistic p-value Exclude lagged ADB FSI in bank credit growth equation 19.218 0.0007 Exclude lagged bank credit growth in AFSI equation 7.0386 0.1339 Lags: 4. Table 17: Granger causality test for the ADB FSI and bank credit growth Basedontheseresults, wecanrejecttheNullhypothesisthattheADBFSIdoesnotGrangercause credit growth and the unemployment rate. Therefore including the ADB FSI can improve forecasts of credit growth and the unemployment rate relative to forecasts based solely on previous values of credit growth and the unemployment rate, respectively. Compared to our estimate of financial stress in Australia (AFSI), we however cannot reject the Null hypothesis that the ADB FSI Granger causes 34

F-statistic p-value Exclude lagged ADB FSI in unemployment rate equation 30.221 0.0071 Exclude lagged unemployment rate in AFSI equation 8.8335 0.8416 Lags: 14. Table 18: Granger causality test for the ADB FSI and the unemployment rate F-statistic p-value Exclude lagged ADB FSI in retail sales growth equation 21.162 0.0003 Exclude lagged retail sales growth in AFSI equation 15.029 0.0046 Lags: 4. Table 19: Granger causality test for the ADB FSI and retail sales growth retail sales growth. Thus, the AFSI has similar results compared to the ADB FSI with respect to credit growth and real GDP per capita growth. 35

E Granger causality tests for variables used to construct AFSI We furthermore test how well our estimate of financial stress in Australia does compared to using single variables that are used to construct the AFSI to predict economic outcomes. We first test each of the 13 variables against real GDP per capita growth, using quarterly averages of the monthly values of the 13 variables used to construct the AFSI. The results are reported in Table 20. Lags are selected based on the AIC criterion with the maximum number of lags set to 4. All significant results that we discuss refer to significance at the 5% significance level. Based on these results, we find that we can reject the Null hypothesis that a given variable does not Granger cause real GDP per capita growth for changes in the cash rate, changes in the 5-yer commonwealth bond yield, housing growth and the S&P / ASX return. To study how the AFSI compares relative to those four variables, we also test if any of these variables can Granger cause a series of financial variables, including credit growth, loans and advances by banks, loans and advances by non-banking financial institutions (NBFI), loans and advances by authorized financial institutions (AFI), narrow credit, personal credit and lending to the government by AFI.34 Table 21 reports the results of the Granger causality tests for the AFSI and credit measures. Tables 22 - 25 report the Granger causality tests for changes in the cash rate, changes in the 5-year commonwealth bond yield, housing price growth and ASX return; and credit measures. Table 22 shows that for six of the seven credit measures used in the Granger causality test, we can reject the Null hypothesis that changes in the cash rate does not Granger cause the given credit measure. However, we can also reject the Null hypothesis that the given credit measure does not Granger cause the cash rate. From Table 23, we find that for changes in the 5-year commonwealth bond yield, we can reject the Null hypothesis that changes the 5-year commonwealth bond yield does not Granger cause credit growth and lending to the government by AFI. Next, Table 24 shows the Granger causality test results for housing price growth. We find that we can reject the Null hypothesis thathousingpricegrowthdoesnotGrangercausecreditgrowth, loansandadvancesfrombanks, loans and advances from AFI and narrow credit. Finally, from Table 25, we find that we can reject the Null hypothesis that the ASX return does not Granger cause credit growth, loans and advances from banks and personal credit. In comparison, from Table 21, we find that we can reject the Null hypothesis that the AFSI does not Granger cause credit growth, loans and advances from NBFI, personal credit and lending to the government by AFI. Thus, in comparison we find that changes in the cash rate and changes in the 5-year commonwealth bond yield can explain less in credit measure relative to our index. Housing price and growth and ASX returns can explain more variables that capture credit measures, the main difference being that both housing price growth and ASX returns seem to Granger cause loans and advances by banks, whereas the AFSI seems to Granger cause loans and advances by NBFI. 34DataondifferentmeasuresofcreditaretakenfromthetableLendingandCreditAggregates(D2)fromtheReserve Bank of Australia. Data on credit growth is taken from the table Growth in Selected Financial Aggrages (D1) from the Reserve Bank of Australia. AFI refers to authorized deposit-taking institutions and non-authorized deposit-taking institutions. Banks refer to authorized deposit-taking institutions, which includes banks, credit unions and building societies. NBFI refers non-authorized deposit-taking institutions, which includes money market corporations, finance companies and securitisers as well as issuers and funds managers (See https://www.rba.gov.au/fin-stability/ fin-inst/main-types-of-financial-institutions.html). Narrow credit includes both loans and advances from AFI plus bill acceptances. Personal credit refers to credit to households that are not mortgages. 36

Variable Test F-statistic p-value Changes in cash rate Exclude lagged cash rate in real GDP per capita growth equation 8.3501 0.0039 Exclude lagged real GDP per capita growth in cash rate equation 0.6653 0.4147 Lags: 1 Changes in 5-year Exclude lagged 5-year bond yield in real GDP per capita growth equation 8.979 0.0027 commonwealth bond Exclude lagged real GDP per capita growth in 5-year bond yield equation 1.3761 0.2408 yield Lags: 1 Changes in 10-year Exclude lagged 10-year bond yield in real GDP per capita growth equation 3.3197 0.0685 commonwealth bond Exclude lagged real GDP per capita growth in 10-year bond yield equation 0.4313 0.5113 yield Lags: 1 Real sovereign debt Exclude lagged debt spread in real GDP per capita growth equation 4.0884 0.3942 spread Exclude lagged real GDP per capita growth in debt spread equation 15.702 0.0034 Lags: 4 Inflation Exclude lagged inflation exp. in real GDP per capita growth equation 1.6659 0.4348 expectation Exclude lagged real GDP per capita growth in inflation exp. equation 3.6437 0.1617 Lags: 2 Term premium Exclude lagged term premium in real GDP per capita growth equation 2.1975 0.3333 Exclude lagged real GDP per capita growth in term premium equation 1.0506 0.5914 Lags: 2 TED spread Exclude lagged TED spread in real GDP per capita growth equation 1.2575 0.5333 Exclude lagged real GDP per capita growth in TED spread equation 0.4701 0.7906 Lags: 2 BAB spread Exclude lagged BAB spread in real GDP per capita growth equation 2.4306 0.1190 Exclude lagged real GDP per capita growth in BAB spread equation 1.0463 0.3064 Lags: 1 Housing Exclude lagged housing growth in real GDP per capita growth equation 10.969 0.0119 growth Exclude lagged real GDP per capita growth in housing growth equation 3.6157 0.3061 Lags: 3 S&P / ASX return Exclude lagged ASX return in real GDP per capita growth equation 7.2761 0.0070 Exclude lagged real GDP per capita growth in ASX return equation 0.4926 0.4828 Lags: 1 S&P / ASX Exclude lagged ASX volatility in real GDP per capita growth equation 1.4192 0.2335 volatility Exclude lagged real GDP per capita growth in ASX volatility equation 0.8781 0.3487 Lags: 1 Banking Exclude lagged banking volatility in real GDP per capita growth equation 4.2863 0.1173 volatility Exclude lagged real GDP per capita growth in banking volatility equation 2.7976 0.2469 Lags: 2 Exchange rate Exclude lagged exchange rate vol. in real GDP per capita growth equation 0.0158 0.9000 volatility Exclude lagged real GDP per capita growth in exchange rate vol. equation 0.0214 0.8837 Lags: 1 Table 20: Granger causality test for variables used to construct the AFSI and real GDP per capita growth 37

Variable Test F-statistic p-value Credit growth Exclude lagged AFSI in credit growth equation 34.861 0.0000 Exclude lagged credit growth in AFSI equation 2.1248 0.7128 Lags: 4 Loans and advances - banks Exclude lagged AFSI in bank loans equation 26.21 0.0000 Exclude lagged bank loans in AFSI equation 23.83 0.0001 Lags: 4 Loans and advances - NBFI Exclude lagged AFSI in NBFI loans equation 8.6144 0.0033 Exclude lagged NBFI loans in AFSI equation 2.6829 0.1014 Lags: 1 Loans and advances - AFI Exclude lagged AFSI in AFI loans equation 18.504 0.0010 Exclude lagged AFI loans in AFSI equation 14.385 0.0062 Lags: 4 Narrow credit Exclude lagged AFSI in narrow credit equation 18.54 0.0010 Exclude lagged narrow credit in AFSI equation 12.355 0.0149 Lags: 4 Credit, other - personal Exclude lagged AFSI in personal credit equation 31.321 0.0000 Exclude lagged personal credit in AFSI equation 1.5569 0.2121 Lags: 1 Lending to government Exclude lagged AFSI in government lending equation 12.296 0.0005 by AFI Exclude lagged government lending in AFSI equation 0.0663 0.7967 Lags: 1 Table 21: Granger causality test for the AFSI and credit measures 38

Variable Test F-statistic p-value Credit growth Exclude lagged cash rate in credit growth equation 27.95 0.0000 Exclude lagged credit growth in cash rate equation 23.985 0.0001 Lags: 4 Loans and advances - banks Exclude lagged cash rate in bank loans equation 44.838 0.0000 Exclude lagged bank loans in cash rate equation 20.718 0.0004 Lags: 4 Loans and advances - NBFI Exclude lagged cash rate in NBFI loans equation 3.8383 0.4283 Exclude lagged NBFI loans in cash rate equation 18.59 0.0009 Lags: 4 Loans and advances - AFI Exclude lagged cash rate in AFI loans equation 39.643 0.0000 Exclude lagged AFI loans in cash rate equation 12.767 0.0125 Lags: 4 Narrow credit Exclude lagged cash rate in narrow credit equation 36.828 0.0000 Exclude lagged narrow credit in cash rate equation 12.156 0.0162 Lags: 4 Credit, other - personal Exclude lagged cash rate in personal credit equation 11.805 0.0081 Exclude lagged personal credit in cash rate equation 8.429 0.0379 Lags: 3 Lending to government Exclude lagged cash rate in government lending equation 21.457 0.0000 by AFI Exclude lagged government lending in cash rate equation 6.8871 0.0087 Lags: 1 Table 22: Granger causality test for changes in the cash rate and credit measures 39

Variable Test F-statistic p-value Credit growth Exclude lagged 5-year bond yield in credit growth equation 11.987 0.0175 Exclude lagged credit growth in 5-year bond yield equation 8.8068 0.0661 Lags: 4 Loans and advances - banks Exclude lagged 5-year bond yield in bank loans equation 25.813 0.0000 Exclude lagged bank loans in 5-year bond yield equation 9.8454 0.0431 Lags: 4 Loans and advances - NBFI Exclude lagged 5-year bond yield in NBFI loans equation 4.9932 0.1723 Exclude lagged NBFI loans in 5-year bond yield equation 49.908 0.0000 Lags: 3 Loans and advances - AFI Exclude lagged 5-year bond yield in AFI loans equation 15.267 0.0042 Exclude lagged AFI loans in 5-year bond yield equation 28.844 0.0000 Lags: 4 Narrow credit Exclude lagged 5-year bond yield in narrow credit equation 13.441 0.0093 Exclude lagged narrow credit in 5-year bond yield equation 28.572 0.0000 Lags: 4 Credit, other - personal Exclude lagged 5-year bond yield in personal credit equation 2.2005 0.6989 Exclude lagged personal credit in 5-year bond yield equation 26.14 0.0000 Lags: 4 Lending to government Exclude lagged 5-year bond yield in government lending equation 5.0394 0.0248 by AFI Exclude lagged government lending in 5-year bond yield equation 3.4743 0.0623 Lags: 1 Table 23: Granger causality test for changes in the 5-year commonwealth bond yield and credit measures 40

Variable Test F-statistic p-value Credit growth Exclude lagged housing price growth in credit growth equation 12.705 0.0128 Exclude lagged credit growth in housing price growth equation 2.5818 0.6301 Lags: 4 Loans and advances - banks Exclude lagged housing price growth in bank loans equation 23.775 0.0001 Exclude lagged bank loans in housing price growth equation 5.3437 0.2538 Lags: 4 Loans and advances - NBFI Exclude lagged housing price growth in NBFI loans equation 1.4773 0.6875 Exclude lagged NBFI loans in housing price growth equation 6.8595 0.0765 Lags: 3 Loans and advances - AFI Exclude lagged housing price growth in AFI loans equation 14.563 0.0022 Exclude lagged AFI loans in housing price growth equation 1.8113 0.6125 Lags: 3 Narrow credit Exclude lagged housing price growth in narrow credit equation 15.062 0.0046 Exclude lagged narrow credit in housing price growth equation 3.3367 0.5032 Lags: 4 Credit, other - personal Exclude lagged housing price growth in personal credit equation 11.059 0.0259 Exclude lagged personal credit in housing price growth equation 11.587 0.0207 Lags: 4 Lending to government Exclude lagged housing price growth in government lending equation 2.0419 0.5638 by AFI Exclude lagged government lending in housing price growth equation 1.5009 0.6821 Lags: 3 Table 24: Granger causality test for housing price growth and credit measures 41

Variable Test F-statistic p-value Credit growth Exclude lagged ASX return in credit growth equation 21.404 0.0003 Exclude lagged credit growth in ASX return equation 0.7333 0.9472 Lags: 4 Loans and advances - banks Exclude lagged ASX return in bank loans equation 12.991 0.0113 Exclude lagged bank loans in ASX return equation 8.7163 0.0686 Lags: 4 Loans and advances - NBFI Exclude lagged ASX return in NBFI loans equation 1.1027 0.2937 Exclude lagged NBFI loans in ASX return equation 3.1189 0.0774 Lags: 1 Loans and advances - AFI Exclude lagged ASX return in AFI loans equation 8.6704 0.0699 Exclude lagged AFI loans in ASX return equation 7.5614 0.1090 Lags: 4 Narrow credit Exclude lagged ASX return in narrow credit equation 8.5014 0.0748 Exclude lagged narrow credit in ASX return equation 7.2179 0.1248 Lags: 4 Credit, other - personal Exclude lagged ASX return in personal credit equation 7.2015 0.0273 Exclude lagged personal credit in ASX return equation 1.7555 0.4157 Lags: 2 Lending to government Exclude lagged ASX return in government lending equation 0.0513 0.8208 by AFI Exclude lagged government lending in ASX return equation 0.0869 0.7681 Lags: 1 Table 25: Granger causality test for ASX return and credit measures 42