Risk-On Risk-Off: How does Risk-On Risk-Off affect returns to the Australian stock market?

Transcription

1 Risk-On Risk-Off: How does Risk-On Risk-Off affect returns to the Australian stock market? By Tariq Haque* Discipline of Finance The University of Adelaide Adelaide, SA, 5005 Abstract Risk-On Risk-Off is a state of financial markets in which many market participants are either risk averse and sell off risky assets such as equities to finance investment in safe assets such as bonds ( Risk Off ), or less risk-averse and willing to invest in risky assets such as equities by selling off safe assets such as bonds. In this study, we look at the Risk-On Risk-Off effect in international stock and bond markets for the period April 2002 to April We find that the Risk-On Risk-Off effect in international stock and bond markets significantly affects the Australian stock market. When Risk-On Risk-Off is not present in international stock and bond markets the Australian stock market generates an average return of 14.8%pa compared to an average return of 0.5%pa when Risk-On Risk-Off is present. Moreover the average return in Risk-On periods is 86.9%pa while the average return in Risk-Off periods is -89.8% reflecting the huge volatility that the Risk-On Risk-Off effect causes in the Australian stock market. * tariq.haque@adelaide.edu.au. Phone: I thank Kevin Davis for providing valuable comments on an earlier draft of this paper. Of course any errors in this manuscript are my own responsibility. 1

2 1. Introduction Risk-On Risk-Off is a phenomenon in financial markets that has been observed since July 2007 with the early warning signs that preceded the Global Financial Crisis (HSBC Global Research: 2010, 2012). Risk-On Risk-Off occurs when many investors take on risk by purchasing risky assets such as equities and simultaneously selling safe assets such as bonds ( Risk-On ) or become extremely risk averse and sell risky assets in favour of safe assets ( Risk-Off ). The risky assets purchased in Risk-On might include equities, or speculative currencies such as the Australian dollar, or commodities like gold and oil while the safe assets purchased in Risk-Off might include government bonds or safe-haven currencies such as the US dollar, Swiss Francs or the Japanese yen. The Risk-On Risk-Off orientation in financial markets can be affected by good news and bad news. For example the following events caused a Risk-Off orientation in financial markets: (1) a warning on July by US Federal Reserve Chairman Ben Bernanke on the magnitude of the subprime crisis and its impact on economic growth; (2) the run on Northern Rock in the UK in September 2007; (3) the collapse of Lehman Brothers on 15 th September 2008; (4) the downgrading of US-debt by Standard and Poor s on 5 th of August 2011 (Lee: 2012) and hints by Ben Bernanke that QE may be wound down in 2013 and ended in mid The following events caused a Risk-On orientation in financial markets: (1) investors looking to move into risky assets in 2009 (Lee: 2012); (2) a speech by Ben Bernanke on 26 th August 2010 hinting that a second round of Quantitative Easing would be forthcoming; (3) the announcement on 15 th of September 2011 of the European Central Bank, the Bank of England, the Swiss National Bank and the Bank of Japan pumping US dollars into money markets to help ease fears over the Eurozone crisis; (4) the so-called Draghi Put speech on July 26,2012 which promised to take all necessary actions to support the Euro. Risk-On Risk-Off implies that returns to risky or Risk-On assets become strongly positively correlated during both Risk-Off and Risk-On states. Returns to safe or Risk- Off assets also become strongly positively correlated during Risk-Off and Risk-On states. For example, in Risk-Off many investors are selling off equities which reduces the price of equity stocks but are then using those proceeds to buy bonds which increases bond prices. Similarly in Risk-On the price of many stocks increases while the price of many bonds falls. This illustrates that in Risk-On or Risk-Off the correlations of returns to risky 1 See for example: 2

3 assets becomes more positive while the correlation of returns to safe assets also becomes more positive. The correlation of returns to risky assets relative to safe assets also becomes more negative in Risk-On or Risk-Off states as the former implies that equity prices are going up while bond prices are going down and the latter implies that equity prices are going down while bond prices are going up. If the Risk-On - Risk-Off designation is volatile, then returns to equity portfolios could become quite volatile because in one state they are going up and in the next state they are going down and so on. Alternatively, the Risk-On or Risk-Off designation could last for relatively long periods which implies that investors in equities could face long runs of positive returns followed by long runs of negative returns and may have few diversification options to alleviate the negative returns. Intra-bond and intra-equity correlations seem to rise in financial crises, in recessions, in weak economic conditions and in volatile conditions (HSBC 2010). 2 3 This evidence is consistent with the Risk-Off state described earlier but not necessarily with Risk-On as we would not expect Risk-On to coincide with financial crises or recessions. highlights that Risk-On Risk-Off or RORO is describing a new result, being high intrabond and intra-equity correlations, as the economy moves out of recession and into expansion. This In this study, we look at past international stock and bond return data to identify when Risk-On and Risk-Off periods have occurred and how the Australian stock market and subsets of it have performed at those times. We find that the Australian stock market, on average, returns 14.8%pa in non-roro periods but only 0.5%pa in RORO periods. In RORO periods, the average return in Risk-On periods is 86.9%pa and -89.8%pa in Risk- Off periods. We show that the sectors with the highest average returns in Risk-On are Materials, Energy and Financials. The sectors with the highest average returns in Risk-Off are Telecommunication services, Healthcare and Consumer Staple and hence offer some advice on where investors should invest in Risk-On and Risk-Off periods. The rest of this paper is structured as follows. Section 2 provides further discussion on strategies an investor could pursue in a risk-on risk-off world. Section 3 describes the data in our study and how we identify periods of Risk-On, Risk-Off or neither (which we term Non-RORO periods). Section 4 presents our results on the performance of sectors of the 2 The fact that high intra-bond or intra-equity correlations are seen in volatile markets, however, may reflect the fact that in recessions volatility naturally increases and it is recession that causes increased correlations rather than volatility itself. There is also a large academic literature on correlations increasing during financial crises or recessions. See, for example, Campbell et al (2002).

4 Australian equity market during Risk-On, Risk-Off or Non-RORO periods. Section 5 concludes. 2. Strategies that an investor could pursue in a risk-on risk-off world In a risk-on risk-off world, a key strategy involves factor timing. Factor timing involves identifying factors that are governing stock prices at different points in time and maximizing or minimizing exposure to the relevant factors. The relevant factors could be sectors, industries or value or momentum (Lee: 2012). In Risk-on, for example, investors might increase exposure to the Consumer Discretionary sector and reduce exposure to Consumer Staples while in Risk-off they might increase exposure to Consumer Staples and Utilities and reduce exposure to Consumer Discretionary stocks. The key skill involves timing of the changes to these sector exposures. Another possible strategy involves investing in assets whose returns are relatively uncorrelated with the core assets in an investor s portfolio where those core assets are assumed to be strongly affected by the RORO paradigm and hence highly volatile since they go up in Risk-On periods and down in Risk-Off periods (HSBC: 2012) Finally strategies such as the carry trade and long/short strategies that have exposure to the RORO factor can be minimized through the appropriate use of derivative contracts (HSBC: 2012). 3 More generally the negative returns to stocks associated with Risk-Off can be alleviated through derivative contracts or short-selling (if this is permitted) although mutual funds are unlikely to be allowed to engage in either of these activities. 3. Data We first obtain data on the weekly returns to a set of government bonds (US, UK, Japanese, Australian and European Monetary Union 10-year government bonds) and a set of equity market indices (S&P 500, Russell 2000, FTSE 100, Nikkei 225, Eurostoxx 50, Dax 30 and the ASX200) 4 from January 1999 to April All of these series are all obtained from 3 The carry trade has exposure to the RORO factor if many investors try to implement the trade at the same time. This causes the exchange rate relative to the funding currency to appreciate when initiating the trade and the exchange rate relative to the funding currency to depreciate when unwinding the trade. In a long/short strategy the long sector may be more exposed than the short sector to a RORO factor which means the profitability of the strategy depends on how RORO is affecting asset prices in general. 4 HSBC Global Research (2010, 2012) present an augmented version of this methodology to determine Risk-On and Risk-Off periods that uses data on trade-weighted indices and commodities in their calculations as well as bonds and equities. 4

5 Datastream. 5 These indices are not converted back into Australian dollars as we want to see how domestic investors are responding to the global investment environment in their own countries. We then calculate the average rolling correlation between pairs of bonds where the correlation is calculated on a rolling window of 52 weeks. 6 As we have five bonds, this means calculating a 5x5 correlation matrix based on the first 52 weeks of data and then averaging the 10 entries in the lower triangle of that matrix. 7 We assign that average correlation value to Week 52 in our sample with no value assigned to any of the first 51 weeks in our sample. We then move the window forward one week covering weeks 1 to 53 and repeat the same process to give an average correlation for week 53 and continue this process for the rest of the weeks in our sample. 8 The same procedure is followed for our six equity indices. The results of the rolling correlation calculations are shown in Figure 1 below. Figure 1: Rolling correlations within bonds and stocks The figure shows the average correlation between the bonds in our sample (Ave_Correl_Weekly_Bonds) and the average correlation between the equity indices (Ave_Correl_Weekly_Stocks) in our sample where the correlations are calculated on a rolling window of 52 weeks AVE_CORREL_WEEKLY_BONDS AVE_CORREL_WEEKLY_STOCKS 5 We download the weekly Total Return series for each bond or equity index which we then convert into returns in the normal way. We also use data on the AFMA Australian Fixed Interest Index as a Total Return Index for 10-year Australian Government bonds was not available. 6 We also do the same calculation using a rolling window of 12 and 26 weeks. These results are not reported here but are available from the author on request. 7 A 5x5 matrix has 25 entries: 5 on the main diagonal, 10 in the lower triangle and 10 in the upper triangle. The entries in the upper and lower triangle are identical as a correlation matrix is symmetric. 8 Because some series have missing values at the start, I start this process once there are at least 52 observations for all series. This means some data is lost at the beginning of the sample period. 5

6 Figure 1shows that intra-stock and intra-bond correlations have risen significantly over time and that these correlations vary significantly over time. In the next section we discuss how we incorporate these features to determine when we are in a Risk-On, Risk-Off or non- RORO period. 3.1 How to determine a Risk-On/Risk-Off/Non-RORO period We consider three possible states for financial markets around the world for any given week in our sample. The first is a normal or Non-RORO state where, for that week, either the average intra-bond correlation is below its average in our full sample or the average intra-equity correlation is below its average in our full sample or both are below their longterm average. It should be noted that the correlation value for a given week is actually the average correlation value based on that week and the 51 preceding weeks. A Risk-On Risk-Off or RORO week is defined to be a week where both the intraequity and intra-bond correlations exceed their long term average over the full sample period. All RORO periods or weeks are then classified as either Risk-On or Risk-Off, which are the second and third states of the world that we consider. A Risk-On week is a RORO week where the average return to our equity indices exceeds the average return to our bonds. Similarly a Risk-Off week occurs when a RORO period has been identified and where the average return to bonds exceeds the average return to equities. In our results below, we present results for Non-RORO, RORO, Risk-On and Risk-Off weeks separately. A normal or non-roro period occurs when either the intra-equity or intrabond correlations or both are below their long-term average. Using this methodology, in our sample period, out of a total of 578 observations from April 2002 to April 2013, 9 there are 225 observations corresponding to the RORO paradigm and 353 observations corresponding to the non-roro paradigm. Within the RORO paradigm, there are 115 Risk-On observations and 110 Risk-Off observations. Figure 2 below shows where the Risk-On, Risk-Off and Non-RORO observations occur in our sample. To draw this graph, a cumulative sum is calculated where the cumulative sum starts at zero. If a Risk-Off period occurs, the cumulative sum is reduced by one and if a Risk-On period occurs the cumulative sum is increased by one. If a Non-RORO period is observed, the cumulative sum remains unchanged. 9 We lose some observations at the start of our period because we need 52 observations to start the rolling correlation process and further observations are lost as some series do not have all observations starting from January

7 Figure 2: Cumulative Risk-Off Risk-On indicator The series below ( Cumulative_Risk_On ) is calculated as follows: the series starts at zero and is either increased by one if a Risk-On period has been identified for the next period, or decreased by one if a Risk- Off period has been identified for the next period, or remains unchanged if the next period is a non-roro period. CUMULATIVE_RISK_ON Figure 2 shows that in our sample period, normal or non-roro conditions are in existence from April 2002 to February From February 2007 to December 2007 there is some volatility in the cumulative risk-on series, as the market tries to interpret information on the subprime crisis. From December 2007 to January 2008 there is a series of risk-off weeks as the subprime crisis begins to affect markets around the world and then relative calm in the markets until October 2008 when another series of risk-off weeks (in response to the Lehman Brothers collapse on 15 th of September 2008) takes the market to a record risk-off state in early June From that point markets switch into a Risk-on phase that lasts until the end of April This period probably coincides with a series of interest rate cuts around the world to stimulate economic growth as well as various quantitative easing strategies being implemented in the United States and the United Kingdom Thereafter, a series of risk-off weeks begins, culminating in a low for cumulative risk in mid-august This phase of risk-off probably pertains to worries about the extent of the Eurozone crisis and worries about the US defaulting on its debt due to protracted debt ceiling negotiations (Lee: 2012). From 10 This depiction of Risk-Off and Risk-On matches fairly closely with the timelines presented in HSBC: The website: has a good summary of quantitative easing implementation in the UK. The website: has a good discussion of the effect of QE in the US.

8 that point, there is some volatility in Risk-On and Risk-Off, possibly in response to ongoing uncertainty in the Eurozone crisis some of which was resolved by a more coherent approach to dealing with the crisis and ongoing QE operations in the US, UK and Japan. 12 Finally, from December 2012 onwards, the markets seemed to have stabilized and we appeared to have entered more normal market conditions. 13 It is also important to note that the long run of Risk-Off and Risk-On phases suggests that factor timing strategies can work (although there will be volatility with the associated returns as the Risk-Off Risk On designations are volatile). 4. The performance of sectors of the Australian equity market during RORO vs non- RORO periods Table 1 below gives average returns and standard deviation of returns, in percent per annum, for the 10 Australian GICS sectors, for Value stocks and Growth stocks, and for large-cap and small-cap stocks as well as the ASX200 and returns to 90 day Bank Accepted Bills and an Australian Government Bond Index. 14 Table 1 shows that in our sample period, the market portfolio has a markedly different performance in non-roro periods compared with RORO periods with average returns of 14.8%pa and 0.5% pa respectively. Most strikingly, within RORO periods, the market portfolio returns 86.9%pa in Risk-On periods and -89.8%pa in Risk-Off periods. The Australian Government Bond Index returns 5.7%pa in non-roro periods compared with 7.3%pa in RORO periods, which consists of Risk-On performance of -7.5%pa and Risk- Off performance of 22.8%pa. The return to 90-day Bank Accepted Bills is 5.4%pa in non- RORO periods and relatively invariant to Risk-On and Risk-Off with means of 4.2%pa and 4.4%pa respectively. 12 The website: has a good summary of major world-wide financial events pertaining to the Eurozone crisis. 13 See for example Saft (2013) and Lefeuvre and Oberg (2013). 14 These annualized figures are obtained by obtaining the weekly average return and weekly standard deviation (both in percent) and multiplying by 52 and 52 respectively. The 10 GICS sectors are: Consumer Discretionary, Consumer Staples, Energy, Financials, Healthcare, Industrials, Information Technology, Materials, Telecommunication Services and Utilities. 8

9 Table 1: Descriptive statistics for equity and bond indices, sector returns, value and growth indices and small-cap and large-cap indices for Non-RORO and RORO periods The table shows the annualized returns, in percent and annualized standard deviations, in percent to various series. All series are downloaded from the Morningstar Database. ASX200 is the ASX200 Accumulation Index. RBA_90_DAY_BAB is the return to holding 90-day Bank Accepted Bills. ASX_GOV_BOND is the Morningstar Australian Government Bond Index. All Value and Growth indices are by Russell Australia. ASX100 is the ASX100 Accumulation Index. ASX_SMALL_ORDS is the ASX Small Ordinaries Index. The ASX GICS Sector returns are also shown. Non-RORO RORO RORO: Risk-On RORO:Risk-Off Mean Std Std Std Std Mean Mean Mean Dev Dev Dev Dev PANEL A: EQUITY AND BOND INDICES ASX RBA_90_DAY_BAB ASX_GOV_BOND PANEL B: GICS SECTORS CONS_DISCRETIONARY CONS_STAPLES ENERGY FINANCIALS HEALTHCARE INDUSTRIALS IT MATERIALS TELECOM_SERVICES UTILITIES PANEL C: GROWTH AND VALUE INDICES AU_GROWTH AU_VALUE AU_LARGE_GROWTH AU_LARGE_VALUE AU_SMALL_GROWTH AU_SMAL_VALUE PANEL D: LARGE-CAP AND SMALL-CAP INDICES ASX ASX_SMALL_ORDS Panel B of Table 1 looks at the performance of the GICS sectors comprising the ASX200. In non-roro periods, the sectors with the highest average returns are Energy, Materials (albeit with a relatively high standard deviation), Utilities, Health and Consumer Staples. In RORO periods, the sectors with the highest average returns are Telecommunication Services, Consumer Staples, Utilities, Health and Financials and this list strongly correlates with the sectors that perform best in Risk-Off periods. The sectors that perform best in Risk-Off periods are Telecommunication Services, Health, Consumer Staples, Utilities and Information Technology. Finally the sectors that give the highest average returns in Risk-On periods are Materials, Energy, Financials, Industrials and IT. This list gives some indication of the sectors that investors could overweight in non-roro, Risk-On and Risk-Off periods respectively. A more formal analysis would involve allowance for the volatility associated with these sectors and their correlations with other 9

10 sectors which is important in determining a portfolio s volatility. We again note the extremely large positive returns for all sectors in Risk-On periods, ranging from 31.9%pa for Telecommunication Services to 126.5%pa for Materials. However these large positive returns are offset by extremely negative average returns to all sectors in Risk-Off periods, ranging from %pa for Materials to -16.9%pa for Telecommunication Services. Panel C of Table 1 shows the performance of the Russell Australia Growth and Value Indices in non-roro and RORO periods. It also shows the performance of Growth and Value Indices within the Large-Cap and Small-Cap universes. In general, Value generates higher expected returns than Growth in non-roro periods, and with lower volatility. 15 In RORO periods, Growth generates higher expected returns than Value although with a slightly higher volatility. Growth also generates higher average returns than Value in Risk-On periods albeit with a slightly higher volatility. In Risk-Off periods both Growth and Value lose significant wealth but Growth has a more negative average return. Finally Panel D of Table 1 shows the performance of Large-Cap stocks versus Small- Cap stocks in non-roro and RORO periods. Here we proxy the performance of Large-Cap stocks by the return to the S&P/ASX 100 and the performance of Small-Cap stocks by the return to the S&P/ASX Small Ordinaries Index. 16 Panel D shows that Large-Cap stocks outperform Small-Cap stocks in both non-roro and RORO periods by achieving a higher expected return and at lower volatility. However Small-Cap stocks have a higher average return than Large-Cap stocks in Risk-On periods but a lower average return than their Large- Cap counterparts in Risk-Off periods. Overall the results shown in Table 1 indicate that relative to the Australian bond market, the Australian equity market performs as one would expect in non-roro periods, on a risk-adjusted basis. However, in RORO periods the Australian stock market performs extremely poorly, generating an average return below the risk-free rate and below the average return to an index of Australian government bonds. This underperformance is due to extremely negative returns in Risk-Off periods and this offsets the extremely positive returns achieved in Risk-On periods. The analysis here suggests that investors should consider overweighting their exposure to a combination of defensive and riskier, more speculative sectors (Energy, 15 This is not quite true for Value relative to Growth in the Small-Cap universe because Small-Cap Growth has a slightly higher expected return Small-Cap Value although its standard deviation is higher. 16 Weekly returns to the S&P/ASX 100 were not consistently available until November Thus the calculations shown in Table 1 Panel D are based on weekly returns from November 2005 to April 2013, which have then been annualized. 10

11 Materials, Utilities, Health and Consumer Staples) in non-roro periods. In RORO periods, defensive sectors should be overweighted (Telecommunication Services, Consumer Staples, Utilities, Health and Financials). In Risk-Off periods, defensive sectors should again be overweighted (Telecommunication Services, Health, Consumer Staples, Utilities and Information Technology) while in Risk-On periods more speculative sectors can be overweighted (Materials, Energy, Financials, Industrials and IT). In general Value stocks would be preferred over Growth stocks in non-roro and Risk-Off periods but Growth would be preferred in Risk-On while large-cap stocks would be preferred to small-cap stocks in non-roro and Risk-Off periods while small-cap stocks may be preferred in Risk-On periods. We again emphasize that risk has not been allowed for in making these statements and this is clearly important as most investors are risk-averse to some degree. 5. Conclusion In this study, we highlighted the significance of the Risk-On Risk-Off or RORO paradigm in financial markets. The RORO paradigm is a relatively new phenomenon in financial markets whereby investors are either very risk averse and buy safe assets such as bonds funded by the sale of risky assets such as shares, or less risk averse and buy riskier assets such as shares funded by the sale of bonds. While previous academic studies have documented that correlations increase significantly during crisis periods which come back to more normal levels in non-crisis periods (Campbell et al: 2002), the RORO effect of more recent times is different because even in a nominal recovery phase, correlations remain high and volatility relatively high also. This could suggest that investors are not convinced about the strength of the recovery and are looking for more signs that the recovery is real or it could reflect that conditions in financial markets have permanently changed with excessive debt levels in the US and Japan and several European countries and no real genuine economic growth to go with that. The effect of the RORO paradigm is that diversification benefits are significantly diluted and equity-only or bond-only portfolios have significantly higher volatility. Moreover the higher volatility in RORO periods comes with a lower expected return as the extremely positive returns in Risk-On periods are almost completely offset by extremely negative returns in Risk-Off periods. In a RORO world one technique that may help investors in equities is factor timing which involves increasing exposure to sectors in industries that are expected to perform well in Risk-On periods, and then to overweight those sectors expected to perform well in Risk-Off periods and those sectors that are expected to 11

12 perform well in non-roro periods. As a general rule, sectors that perform relatively well in recessions should be over-weighted in Risk-Off periods such as Consumer Staples and Utilities and sectors that perform well in bull markets such as Materials and Financials should be overweighted in Risk-Off periods while conventional allocations can be used in non- RORO periods. There are many limitations in this research. Among them is the use of weekly returns and the use of a 52-week window to calculate rolling correlations to determine Risk-On and Risk-Off periods. The use of daily returns and a shorter window could potentially give more accurate results. Further the set of safe and risky assets could be expanded to include government bonds and equity indices from more European nations and more countries from Asia and the emerging markets. The set of risky assets in the analysis could also be expanded to include commodities and foreign exchange rates. There are many areas for future research arising from this study including the performance of bonds and property in RORO periods and a study of optimal asset allocations for different asset classes in RORO versus non-roro periods. This is particularly relevant for superannuation funds that make up such a significant proportion of the managed funds industry. Finally studies that forecast the Risk-On Risk Off designation would be invaluable to fund managers and assist them significantly in their sector allocation decisions. 6. References Campbell, R., Koedijk,K, and P.Kofman, 2002, Increased Correlation in Bear Markets, Financial Analysts Journal, 58(1), HSBC Global Research, 2010, Risk on risk off: the full story* HSBC Global Research, 2012, Risk on Risk off, Fixing a broken investment process* Lee, W., 2012, Risk On, Risk Off, Journal of Portfolio Management, 38(3), Lefeuvre,E. and S.Oberg, 2013, The end of Risk on/risk off?, Special Report Economic Research, NATIXIS. Downloaded from: Saft, J., 2013, Risk-on, risk-off may be ending. Downloaded from: *No longer available online but available from the author on request 12