Institute for International Integration Studies IIIS Discussion Paper

No.310 / December 2009

IIIS Discussion Paper No. 310

Is Gold a Safe Haven? International Evidence

Dirk G. Baur, Dublin City University Business School Thomas K. McDermott - Trinity College Dublin Institute for International Integration Studies

IIIS Discussion Paper No. 310

Is Gold a Safe Haven? International Evidence

Dirk G. Baur, Dublin City University Business School Thomas K. McDermott, Trinity College Dublin Institute for International Integration Studies

Disclaimer Any opinions expressed here are those of the author(s) and not those of the IIIS. All works posted here are owned and copyrighted by the author(s). Papers may only be downloaded for personal use only.

Is Gold a Safe Haven? International Evidence

Dirk G. Baur* 1 Dublin City University Business School

Thomas K. McDermott Trinity College Dublin Institute for International Integration Studies

This version: September 2009

Abstract The aim of this paper is to examine the role of gold in the global financial system. We test the hypothesis that gold represents a safe haven against stocks of major emerging and developing countries. A descriptive and econometric analysis for a sample spanning a 30 year period from 1979-2009 shows that gold is both a hedge and a safe haven for major European stock markets and the US but not for Australia, Canada, Japan and large emerging markets such as the BRIC countries. We also distinguish between a weak and strong form of the safe haven and argue that gold may act as a stabilizing force for the financial system by reducing losses in the face of extreme negative market shocks. Looking at specific crisis periods, we find that gold was a strong safe haven for most developed markets during the peak of the recent financial crisis.

JEL: G10, G11, G14, G15 Keywords: gold, safe haven, financial markets, uncertainty

Corresponding author, address: Dirk Baur, DCU Business School, Collins Avenue, Dublin 9, Ireland, email: [email protected]. The authors would like to thank seminar participants of DCU Business School, in particular Billy Kelly, Hiroyuki Kawakatsu and Conor McKeating. We also thank Niels Schulze for his comments.

INTRODUCTION
Since the beginning of the financial crisis in July of 2007, the nominal gold price has risen 42%. 2 The performance of gold is all the more impressive given the losses suffered in other asset classes during the present crisis. While the financial media regularly refer to gold as a safe haven asset, the claim has rarely been tested in the literature. This paper aims to investigate the potential role for gold as a safe haven from losses in financial markets.

A haven is defined as a place of safety or refuge. In times of stormy weather, ships seek out the safe haven of a port or harbour to ride out the storm. A safe haven asset must therefore be some asset that holds its value in stormy weather or adverse market conditions. Such an asset offers investors the opportunity to protect wealth in the event of negative market conditions. Whether or not investors will seek out a safe haven depends on a number of factors. Our analysis suggests that investors are most likely to look for a haven in response to severe market shocks suffered over a short period. We also find evidence that the haven effect is generally only present in developed markets and not in emerging markets. Losses suffered in emerging markets, even if severe, do not induce international investor movement towards the safe haven. The role of uncertainty in stimulating interest in the safe haven asset is also tested.

The recent and ongoing financial crisis and the attendant strength of the gold price, presents a strong motivation to test the viability of gold as a haven from losses in financial markets. Evidence of the potential for gold to act as a safe haven asset was presented by Baur and Lucey (2009). Their results show that gold tends to hold its value if stock markets experience extreme negative returns in Germany, the UK and the US. McCown and Zimmerman (2006) examine a number of characteristics of gold as a financial asset. They find evidence of the inflation-hedging ability of gold. Their analysis also suggests that gold shows the characteristics of a zero beta asset, bearing no market risk, while they find the returns on gold over a 33 year period (1970-2003) are just slightly higher than the mean return on Treasury Bills. Capie, Mills and Wood (2005) analyze the role of gold as a hedge against the dollar, finding evidence of the exchange-rate hedging potential of gold. Other examples of studies that examine the financial characteristics of gold include Faugere and Van Erlach (2006), Lucey et al. (2006) and Sherman (1982). In relation to the safe haven property and its role in financial markets, Ranaldo and Sderlind (2007) analyze the safe haven status of various currencies (see also Kaul and Sapp, 2006).
2

Nominal gold price increased 42% from July 2007 to March 2009.

Upper (2000) examines the role of a specific safe haven asset, i.e. German government bonds, during a specific period of market stress. Various strands of investor behaviour literature are also related to our discussion of safe haven assets. The financial contagion literature examines the responses of investors to financial market shocks (e.g. Forbes and Rigobon, 2002). Calvo and Mendoza (2000) find that investors faced with losses in emerging markets will tend to shift their portfolios towards the average portfolio, reflecting investors concerns not only about returns in absolute terms, but also relative to the performance of other investors. This may explain the qualitative difference in our results for developed versus emerging markets. The flight to quality literature emphasizes investors movements from stocks to bonds in response to negative market shocks (see for example Gonzalo and Olmo, 2005, Gulko, 2002 and Hartmann, Straetmans and de Vries, 2004). We aim to examine the role of gold as an alternative quality asset in times of severe market stress. With the exception of Baur and Lucey (2009) none of the above literature explicitly examines the role of gold as a safe haven against losses in financial markets. This paper builds on the work of Baur and Lucey (2009), by extending the analysis in a number of important ways. We look at investor reactions to varying degrees of stormy weather. Our dataset allows us to differentiate between storms of various sizes, in terms of both the severity 3 and the duration 4 of shocks to the financial system. We also pursue a multi-country analysis, using major emerging and developed countries from a sample of 53 international stock markets, thus allowing us to test the safe haven effect across a broad cross-section of world stock markets. Our data cover a 30-year period from March 1979 to March 2009. We also broaden the research approach, by examining two further questions related to the safe haven property and the role of gold in the global financial system: (i) To what extent does gold protect wealth during extreme market conditions, i.e. is it a weak or strong safe haven? 5 And (ii) what role do currency movements play in either driving or disguising the safe haven property of gold?

Using daily data, we find evidence of the strong-form safe haven for seven of the thirteen individual country stock indices tested: Canada, France, Germany, Italy, Switzerland, the UK

The severity of the shock is taken into account by looking at a range of lower quantiles of stock returns. 4 The duration of the shock is represented by different frequencies of data; daily, weekly and monthly returns. 5 A weak safe haven we define as an asset which is not correlated with the reference asset or portfolio. Therefore as the reference asset loses value, the value of the weak safe haven asset on average will remain unchanged. A strong safe haven asset is negatively correlated with the reference asset or portfolio and therefore gains value as the reference asset loses value. See formal definitions in section two.

and the US. The presence of a strong safe haven in these markets suggests the potential for gold to act as a stabilizing force for financial markets by reducing losses when it is most needed, i.e. during crisis periods. An analysis of the role of exchange rate effects also shows that a common currency denomination (US Dollars) of both stock indices and gold generally increases the comovement in all market conditions eliminating or greatly reducing the safe haven property of gold. The paper is structured as follows. Section one examines the performance of gold as a financial asset. Section two sets out the formal definitions of the hedge and safe haven properties. Section three contains the empirical analysis, including reported results, while section four concludes.

I.

THE PERFORMANCE OF GOLD AS A FINANCIAL ASSET

The beauty of gold is, it loves bad news. 6

Gold is often referred to as a store of value and a safe haven asset. In this section we examine some of the characteristics of gold that might explain the perception that it is somehow different from other assets. If gold does indeed love bad news, this would suggest prima facie evidence that it can act as a safe haven. Gold has been used as a store of value and a means of exchange for millennia. The 17th Century British Mercantilist Sir William Petty described gold, silver and jewels as wealth at all times and all places. 7 Golds image as an immutable store of value has become culturally embedded, reinforced by its historic links to money. The gold standard system involved linking the value of currencies directly to gold. Central banks around the world continue to hold gold as one of the forms of reserves used to defend the value of their currencies. In India, the worlds largest consumer of gold, the precious metal holds a unique socio-cultural significance as a symbol of a familys wealth and status. And yet, the perception of gold in the financial media has wavered in recent times from barbarous relic to an attractive each way bet against risks of financial losses or inflation (The Economist 2005, 2009). Investors have traditionally used gold as a hedge against inflation or a falling dollar. Because gold is priced in dollars, if the dollar loses value, the nominal (dollar) price of gold will tend
Harry Rabbit Angstrom Rabbit is Rich, John Updike (as quoted in the Economist, February 26th 2009). 7 Sir William Petty Discourse on Political Arithmetick, 1690.
6

to rise, thus preserving the real value of gold. In this way, gold can act as a hedge against exchange rate risk for investors with dollar holdings. Evidence of this phenomenon is presented in Capie, Mills and Wood (2005). In times of uncertainty, when asset values become ambiguous due to investors unwillingness to trade, the attractiveness of gold may increase due to the relative simplicity of the gold market. If the determinants of its value are easier to understand and evaluate, then gold might become an attractive alternative investment, offering investors a greater sense of certainty during periods of financial market turbulence. As a physical asset, gold has an intrinsic value. Its main use is in jewellery but it also has dental and industrial applications. Its value is not dependent on future earnings or debt and investing in gold carries no risk of default. Other commodities may also share these properties. However, what sets gold apart is its behaviour during periods of falling asset values. Aside from its practical uses, the utility of gold may derive from this very unusual property; recent experience suggests that, unlike other financial assets, the value of gold tends to rise in response to negative market shocks. The price of gold has experienced a secular increase in recent years, in line with the boom across other commodity classes. Figure 1 illustrates this relation. The Goldman Sachs commodities index (GSCI) increased 326% in the decade to its peak in July of last year (2008). Since then, however, the index has lost 70% of its value (July 08 to March 09). The equivalent index for gold rose 222% in the decade to its peak in March of last year (2008). After a brief dip in the latter half of 2008, the gold index has recommenced its upward trajectory in the first half of 2009.

< Insert figure 1 about here>

Figure 2 shows the period of the current global financial crisis. The crisis intensified in September 08 with the collapse of Lehman Brothers. Fears of a global recession sent stock markets plummeting in October 2008. 8 Since October 08 the gold price index has surged, indicating a positive response to the intensification of the financial crisis. This is in sharp contrast to the declining value of commodity index returns over the same period.9

< Insert figure 2 about here>

On October 15th 2008 the Dow Jones index recorded its biggest one-day fall losing 8% of its value in a single days trading since Black Monday in 1987. 9 The correlation coefficient for the gold index and the GSCI went from 0.55 for our total sample period of 30 years (March 1979 to March 2009), to 0.20 for the period since August 2007 (beginning of the credit crunch), to -0.32 for the period since September 2008 (intensification of the financial crisis with the collapse of Lehman Brothers). The correlation between gold prices (in $ per Troy ounce) and the world stock market index went from 0.35 to -0.19 to -0.42 for the same intervals.
8

Unlike other assets, gold prices appear to react positively to negative market shocks. The historic high in real gold prices came about in 1980 when gold surpassed $2000 an ounce (2008 dollars) against a backdrop of rampant inflation and the threat of a global recession following the oil crises of the 1970s. More recently the dramatic rise in the gold price over the past 18 months has come about during the worst financial and economic crisis to hit the global economy since the Great Depression of the 1930s. The composition of gold demand is rapidly changing, in the context of global economic turbulence, while gold supply remains relatively fixed. 10 The total demand for gold is made up of three categories: jewellery, industrial & dental, and investment demand. While jewellery and industrial & dental demand tend to follow the business cycle (being largely determined by consumer spending power), the demand for gold from investors would appear to be counter-cyclical, with demand from this sector rising as the global economy enters recession. 11 Traditionally gold demand has been dominated by the demand for gold jewellery. 12 In recent months, however, the pattern of gold consumption has shifted dramatically with jewellery demand falling in response to a combination of high prices and weak consumer purchasing power. Jewellery demand for gold fell 11% (in volume terms) in 2008, while Q1 2009 saw a 24% drop in the demand for gold jewellery in year-on-year terms. 13 By contrast, the investment demand for gold has soared with ETFs (and similar products) becoming for the first time the single largest category of gold demand. Investor demand for gold rose 64% in 2008. The first quarter of 2009 saw a record level of investment activity in gold, with demand from ETFs increasing 540%, in year-on-year terms, to 465 tonnes at a value of US$13.6bn. 14 The dramatic rises in the gold price that have been witnessed in the past 18 months can therefore be explained by increased investor interest in the precious commodity. According to The Economist the recent surge in gold prices (since the start of 2009) has been driven by

Mining production, which accounts for 60% of total supply, actually fell 3% in 2008. Supply of gold, like that of other commodities, is relatively inelastic, in part due to the difficulty associated with extracting it from the ground and the long lead-in times between the establishment of a new mine and commencement of commercial production. (This process can take as long as five years, according to the World Gold Council). 11 This is evident from recent trends in gold prices, as well as our own analysis, which demonstrates an association between falling stock market values and rising gold prices. 12 Jewellery consumption represented roughly 60% of total identifiable end-use demand, excluding central banks, in 2008, down from approximately 70% in 2007. Source: World Gold Council. 13 Source: World Gold Council. 14 Source: World Gold Council. Increased investor activity in gold has been facilitated by the introduction of Commodity Exchange-Traded Funds (ETFs), which serve to reduce the carrying cost of gold for investors.

10

investors looking to preserve their wealth. 15 This reinforces the perception of gold as representing a secure investment for those with something to lose. The unique characteristics of the gold market; its intrinsic value as a precious metal, the relative inelasticity of supply, and in particular the counter-cyclical elements of the demand for gold, lend weight to the theory that gold can act as a store of value, or a haven, in times of financial market turbulence and/or global uncertainty.

II.

DEFINITIONS

This section defines a hedge and a safe haven to distinguish the two terms from each other. The section builds on the definitions provided in Baur and Lucey (2009) but extends their work in one important respect. We differentiate between a strong hedge and a weak hedge and a strong safe haven and a weak safe haven. This distinction enables an assessment of the benefits of a safe haven asset for the financial system.

Hedge: A strong (weak) hedge is defined as an asset that is negatively correlated (uncorrelated) with another asset or portfolio on average.

Safe haven: A strong (weak) safe haven is defined as an asset that is negatively correlated (uncorrelated) with another asset or portfolio in certain periods only, e.g. in times of falling stock markets.

The distinguishing feature of the two types of asset described above is the length of the effect. The important property of the hedge is that it holds on average while the key property of the safe haven is that it is only required to hold in certain periods, e.g. a financial crisis. Assets that work as a hedge against stocks might co-move with stocks in crisis periods since investors sell different types or all assets simultaneously. This can be explained by herd behaviour or contagion (e.g. see Calvo and Mendoza, 2000, Forbes and Rigobon, 2002 and Boyer, Kumagai and Yuan, 2006). In contrast, it is possible that there are assets that are only negatively correlated in crisis periods and co-move with the other asset on average. In this case, investors purchase the asset in these times only. The asset does not lose value in such times and thus works as a safe haven.

15

Haring away, February 26, 2009.

The distinction of a strong and weak hedge and safe haven is not only semantic but also important for investors. If an asset is negatively correlated with another asset or portfolio, investors enjoy positive returns if the other asset or portfolio exhibits (extreme) negative returns. This is not the case if the assets are uncorrelated. Since positive returns of an asset during times of financial stress or turmoil can enhance the stability of the market by reducing overall losses, a distinction of weak and strong properties of the assets is important.

III.

EMPIRICAL ANALYSIS

This section contains the empirical analysis comprising an introduction to the data set, a descriptive and econometric analysis and a summary of the main findings.

A. Data
The data consists of daily, weekly and monthly continuously compounded stock returns of a sub-set of the 53 constituents of a world index provided by Datastream. The constituent international stock indices are denominated in local currency and the weighting is based on the market capitalization. The sub-set comprises the seven largest developed countries (G7), the largest emerging markets (BRIC countries) and Australia and Switzerland. Australia is selected to include a small developed country with a large commodity market and Switzerland is selected to include a small European and non-Euro market with a strong and potentially important currency. We additionally use regional indices which represent groups of stock indices such as North America, Latin America, Europe, EU, EMU and Emerging Markets. The regional indices are denominated in US$. The price of gold is also obtained from Datastream (Gold Bullion in US$ per Troy ounce). The data cover a 30-year period from March 2, 1979 until March 2, 2009 leading to a sample size of 7,826 observations for daily data, 1,566 observations at the weekly frequency and 360 monthly observations. Table 1 reports summary statistics of the data for daily returns. The table contains the name of the market (country), the number of observations, the mean, the standard deviation and the minimum and maximum return for each market. The return on the world index and gold is also presented in the first two rows of the table. The average daily return of the world index is higher than the return of gold but the standard deviation is larger for gold than for the world index. This finding is also represented by the minimum and maximum values of the returns. Gold exhibits more extreme negative values (-0.1787) than the world index (-0.0976) and also more extreme positive values (0.1221 compared to 0.0817 for the world index). This pattern also holds for the returns based on weekly and monthly data.

The data also illustrate that emerging markets generally exhibit higher average returns, a higher risk (standard deviation) and more extreme values than developed markets. The values are generally also increasing with decreasing frequencies, that is, weekly data exhibit larger extreme values than daily data and monthly data show larger absolute values than weekly and daily data.

< Insert table 1 about here>

B. Descriptive Analysis
This section provides a descriptive analysis of the characteristics of the gold-stock market relation. Figure 3 presents the evolution of the world index and the price of gold through time for the 30-year sample period and figure 4 illustrates the time-varying volatility of the respective return series estimated with a GARCH(1,1) model.

< Insert figure 3 about here>

< Insert figure 4 about here>

Figure 3 shows that the world index increased from a value of 100 in 1979 to over 1,600 in 2007 before it fell to values below 800 in 2009. The price of gold increased from around 200 in 1979 to above 1,000 in 2008. The graph also shows that the price of gold and the level of the world index co-move in certain periods (e.g. from 2004 to 2007) and move in opposite directions in others (e.g. from 1998 to 2003). This suggests that the relation of gold and the world equity portfolio is not constant, i.e. the beta changes over time. Obviously the period in which gold and the equity portfolio move in opposite directions is consistent with gold as a hedge against changes in the equity portfolio; gold loses value in the bull market (from 1998 to 2000) and gains value in the bear market (from 2000 to 2003). In contrast, the period from 2003 to 2007 is not consistent with a hedge since there is strong co-movement between the two assets. In this period gold is potentially driven by growth-related factors especially demand from emerging markets. It is also possible that investors anticipated a crash and bought gold as a hedge against such an event. After the stock market crash in 2008, gold regained its negative correlation and thus its hedge or safe haven status. Figure 5 focuses on a period in which gold moves in opposite directions to the world stock market. This pattern is very clear in 2008 when stock markets fell significantly and gold increased by a similar magnitude.
9

< Insert figure 5 about here>

The figure also shows that there is considerable noise in both the world index and the gold price. Figure 6 presents an even closer look and shows how the relationship changes within three months (August 20, 2008 to December 3, 2008) from a negative relation to a positive relation followed by a negative relationship. A least-squares regression of gold returns on world stock returns yields a positive and significant coefficient for the sub-sample shown in the figure. If we divide this period in two equal halves, it yields a negative and significant coefficient for the first sub-sample period and a positive coefficient for the second sub-sample period. This example aims to show that a positive beta on average does not exclude periods in which gold temporarily acts as a hedge or safe haven. The same is true for a negative beta (hedge) on average; it would not exclude periods in which gold co-moves with other assets possibly reducing the hedge property when it is most needed.

< Insert figure 6 about here>

In order to analyze the time-varying behaviour of gold with respect to global stock markets we present coefficient estimates of a rolling window regression of the gold return on the World portfolio index. The window length is set to 250 daily observations which approximately represents one calendar year. Figure 7 illustrates the coefficients (betas) for the World portfolio index. The time-varying betas confirm the hypothesis of a changing relationship. The results show that gold is not a hedge at all times but only in certain times. For example, gold is not a hedge in the beginning of the sample and not in the end of the sample. However, the relation is mostly around zero or even negative for more than ten years (between 1990 and 2003). The negative relationship implies that gold was a hedge for the index in that period. Repeating this analysis revealed a qualitatively similar pattern for the regional indices (EU, EMU, North America, South America, Emerging Markets). However, for ease of interpretation, we only present results for the World index.

< Insert figure 7 about here>

Figure 8 shows that the variation in betas is also evident in individual stock market indices. Again for the sake of clarity we only present the results for a single index the US. However,

10

qualitatively similar results were found for example for the UK and Germany also. The individual stock market indices are denominated in local currency in contrast to the regional indices which are denominated in US Dollar.

< Insert figure 8 about here>

The graphs (figures 7 and 8) show that the beta (i.e. the correlation of stock markets and gold) is not constant over time but changing. Since there are periods in which the beta is positive and periods in which it is negative, gold is not a hedge at all times. This justifies the separation of a hedge at all times into a hedge on average and a safe haven in specific periods. The next section describes different models to analyze the hedge and safe haven property of gold for stock markets.

C. Econometric Analysis
In this section we present the econometric models to analyze the safe haven property of gold. The gold price is assumed to be dependent on changes in the stock market. Moreover, we assume that the relationship is not constant but is influenced by specific, extreme, market conditions. Equations 1a-1c present the principal regression model to analyze the safe haven property of gold.

rGold, t = a + bt rstock, t + et bt = c0 + c1 D(rstock q10) + c2 D(rstock q5) + c3 D(rstock q1) ht = + et-1 + ht-1

(1a)

(1b)

(1c)

Equation 1a models the relation of gold and stock returns. The parameters to estimate are a and bt. The error term is given by et. The parameter bt is modelled as a dynamic process given by equation 1b. The parameters to estimate in equation 1b are c0, c1, c2 and c3. The dummy variables denoted as D() capture extreme stock market movements and are equal to one if the stock market exceeds a certain threshold given by the 10%, 5% and 1% quantile of the return distribution. If one of the parameters c1, c2 or c3 is significantly different from zero, there is evidence of a non-linear relationship between gold and the stock market. If the parameters in equation 1b are non-positive (including c0), gold acts as a weak safe haven for the market under study. If

11

the parameters are negative and statistically different from zero, gold functions as a strong safe haven. Gold is a hedge for the market under study if the parameter c0 is zero (weak hedge) or negative (strong hedge) and the sum of the parameters c1 to c3 are not jointly positive exceeding the value of c0. Finally, equation 1c presents a GARCH(1,1) model which is used to account for heteroscedasticity in the data. Equations 1a-1c are jointly estimated with Maximum Likelihood.

Equation 1b focuses on extreme negative returns to model potential non-linearities of the gold-stock index return relationship. If the relationship is, in fact, non-linear, this implies that investors act differently in extreme market conditions compared to normal conditions. An alternative to this specification is to assume that the gold-stock relation changes with uncertainty of the markets. One proxy for uncertainty is the conditional volatility of the world portfolio. If uncertainty is high, the gold-stock relation is different than in periods in which uncertainty is low. If the conditional volatility of the world index is estimated with a GARCH process and different volatility levels (regimes) are chosen, an analogue to equation 1b can be written as follows

bt = c0 + c1 D(hstock q90, t-1) + c2 D(hstock q95, t-1) + c3 D(hstock q99, t-1)

(2b)

where the dummy variable is equal to one if the lagged conditional volatility of the world portfolio lies in the 90% (95% and 99%) quantile and zero otherwise. The problem of generated regressors is acknowledged but arguably small because of the use of volatility levels. Since this approach picks up longer spells than the quantile-based approach (equation 1b) the estimation results are expected to be different. For example, since volatility is highly persistent and clusters around certain periods, the dummy variables are equal to one for longer and consecutive periods than the dummy variables in equation 1b. While the dummy variables in equation 1b can be expected to be equal to one only for one or two consecutive days or weeks, the terms in equation 2b are likely to capture periods of several days or weeks.

Finally, a less statistical and more arbitrary approach would be to identify and define certain periods such as economic or financial crises, bull and bear markets and to use time dummies which are equal to one if the returns overlap with the predefined period and zero otherwise. 16 The model would be specified as in equation 3b below

This approach is related to the work of Ranaldo and Sderlind (2007) who analyze the role of currencies as a safe haven in crisis periods.

16

12

bt = c0 + c1 D(Asian crisis, 1997) + + cn D(subprime crisis, 2008)

(3b)

If the parameters c1, c2, or cn are zero or negative, gold is a safe haven in the respective crisis period. If the parameter is positive, gold co-moves with the stock market and does not meet the criteria for a safe haven. An alternative specification could utilize a dummy variable without the interaction term. Such a dummy would capture the change in the price of gold in a crisis period compared to normal periods. However, such a pure dummy would not indicate any relation of the gold price with stock market changes. Hence, we do not estimate such an alternative model. The next section presents the estimation results and discusses the findings.

Estimation Results
In this section we present the estimation results of the models outlined above. Tables 2a, 2b and 2c show the estimates of a regression model for daily, weekly and monthly data given by equations 1a-1c. The tables contain the estimates of c0 and the total effects for extreme market conditions, that is, the sum of c0 and c1 for the 10% quantile, the sum of c0, c1 and c2 for the 5% quantile and the sum of all coefficient estimates for the 1% quantile. The tables are structured as follows. The top panel contains the estimates for the regional indices (World, North America, Latin America, EU, EMU and Emerging Markets) and its relation with gold in normal and extreme market conditions. The bottom panel contains the estimates for the individual stock markets as listed in table 1.

< Insert table 2a about here>

< Insert table 2b about here>

< Insert table 2c about here>

The results for the regional indices denominated in US Dollar show that gold is not a hedge for most indices except North America. In fact, gold appears to strongly co-move with the regional indices, both on average and in extreme market conditions, across all frequencies of data (daily, weekly and monthly). These findings are hardly surprising. Because the regional indices are denominated in US Dollar terms, the results are influenced by changes in the exchange rate. The common currency denomination of gold and the stock market indices introduces a common feature in the data which yields a greater degree of co-movement compared to a case in which local currency stock index returns are used. This would appear to

13

explain the observed positive correlation between non-US stocks (denominated in US Dollar) and the gold price. 17 The North American index is the exception to the rule, as the currency effect discussed above is obviously redundant in the case of US stocks. Our results show a negative average correlation between gold prices and the North American index across all frequencies of data. However, the relationship is only significant for daily data. In spite of the currency effect, we find highly significant negative correlations between gold and the regional indices for extreme market shocks (at the 1% quantile) for daily data. The currency effect appears to be dominated by other drivers (the safe haven effect) in the case of extreme market disturbances. Our results suggest that gold acts as a hedge for North American stocks, and in the case of daily data, provides investors the compensating property of a strong hedge. Gold is also a (strong) safe haven in the case of extreme negative market shocks at a daily frequency (but generally not at lower frequencies). This result is consistent across all but the Emerging Markets index, indicating that investors react differently to shocks in emerging markets as opposed to developed markets. The results also show that the relation of gold and stock returns is non-linear in many cases. In order to control for the (possibly) distorting effects of the common currency denomination of stock market indices and gold prices, we repeat the analysis for a selection of individual stock markets quoted in local currencies. The results for individual country indices (in local currency) for daily data are as follows. Gold is a hedge for France, Germany, Italy, Switzerland, the UK and the US. Gold is also a safe haven for these markets except for Italy at the 10% quantile. Most of the emerging markets exhibit small positive (but mostly insignificant) coefficient estimates. Australia, Canada and Japan show small but positive and significant coefficient estimates for the average relation excluding the possibility of a hedge and insignificantly positive or small negative coefficients implying a weak safe haven for extreme market conditions. The pattern is qualitatively very similar for weekly and monthly return data. Developed markets with the exception of Australia, Canada and Japan show mostly negative coefficient estimates at all frequencies and all quantiles. The stock markets of Australia, Canada and Japan generally co-move with gold on average and show negative coefficient estimates in some extreme market conditions (e.g. 1% quantile Australia, monthly frequency).

If the dollar loses value relative to other currencies, this will cause the dollar-denominated nominal value of non-US stock markets to rise. A falling dollar is also likely to cause a rise in the nominal dollar price of gold, as discussed in the introduction. Thus movements in the dollar are likely to be driving the co-movement of non-US stocks and the gold price (both denominated in US Dollar).

17

14

Again the results are consistent with gold providing a safe haven for developed markets in times of extreme market conditions. Faced with heavy losses or high levels of uncertainty investors sell stocks and buy gold. However, this pattern of behaviour does not appear to carry over to emerging markets, where the observed positive correlations between stocks and gold prices across all quantiles would seem to confirm the hypothesis that investors react differently to shocks in emerging markets. They may well sell their shares in response to a negative market shock. But rather than seeking the shelter of a haven asset, they instead may be content to shift their portfolios towards the relative safety of developed world markets. The finding that Australia, Canada and Japan prove exceptional might be explained by the nature of their stock markets. Canada and Australia both have significant mining interests in their national stock markets, which would tend to cause stocks and commodity prices (including gold) to co-move. Japans stock market performance differs from the other countries in that stock prices rather stagnated for a long period. This causes gold to play a different role in Japan than in any other market.

The results can be summarized as follows. Gold is a strong hedge for all European markets and the US. Gold is also a safe haven in these markets. The strength of the safe haven effect varies across market conditions (quantiles) and return frequencies. Gold is generally a strong safe haven in very extreme market conditions (1%) for daily and weekly returns in European markets and the US. In contrast, gold is neither a hedge nor a safe haven for the BRIC countries, Australia, Canada and Japan. The coefficients reported in table 2 are relatively small. That is, the coefficients generally lie in the interval [-0.1, 0.1], which means that a 10% change in the stock market leads to a 1% change in the gold price. One reason for the relatively small coefficients is the time-variation in the gold-stock relationship as described and reported above. An explicit analysis of shorter periods, e.g. a strong bull or bear market would lead to larger absolute coefficients. The plots of the time-varying beta estimates presented above show that the betas are larger in absolute terms for sub-periods of 250 trading days. These episodes also include bull and bear markets. We do not analyze such market trends in more detail since it is to some extent arbitrary to identify bull and bear markets and to distinguish such periods from each other and from other periods.

Uncertainty and the Safe haven effect In this section we present the estimation results of the regression model specified in equations 2. We use a measure of extreme market conditions given by the daily conditional volatility of
15

the world index estimated with a GARCH(1,1) model. Different levels of this volatility are used as a proxy for global financial market uncertainty. If investors buy gold in times of high volatility and thus uncertainty, the price of gold should increase in such periods. The results are presented in table 3 and illustrate that gold is a hedge for the Euro countries, Switzerland and the US and works as a safe haven in periods of increased volatility (90% and 95% thresholds) but does not work as a safe haven in spells of extreme volatility (99% threshold) or uncertainty except for the US and China. The marginal effect is positive and statistically significant for all markets but the total effect, despite being smaller than the marginal effect, is only negative for the US and China.

< Insert table 3 about here>

The results imply that in times of increased uncertainty, gold is a safe haven but loses this property for most markets in times of extreme uncertainty proxied by a global volatility estimate. If volatility and uncertainty exhibit extreme levels stocks and gold co-move thereby eliminating the safe haven property. Since volatility is clustered around certain periods, the results also imply that the safe haven is effective for a certain period and not confined to the days of the extreme shocks only. This also explains why the results differ between the two regression models. Since there is volatility clustering but no extreme negative shock clustering the regressions yield different results. The next section analyzes certain periods of high volatility or uncertainty by focussing on three major episodes of financial crisis.

Crisis periods and the Safe haven effect The model specified in equation 3b analyzes crisis periods explicitly and not implicitly as the models in equations 1 and 2 do. The specification of the crisis periods, while more economic and less statistical, is more arbitrary than for the other two models (despite the specification of thresholds) since crisis periods have to be selected and the start date and the length have to be defined. As has been shown in the so called contagion literature, it is not straightforward to define the outbreak of a crisis and its end (see Dungey et al., 2004). We analyze three major crisis periods: (i) the stock market crash in October 1987, (ii) the Asian crisis in October 1997 and (iii) the global financial crisis which originated as a sub-prime crisis in 2007 and peaked in September 2008.

16

We define starting dates and assume that most of the crises and its effects occurred in the first 20 trading days (approximately one month) 18 after the start date. We define October 16, 1987, October 22, 1997 and September 10, 2008 19 as the start date for each crisis period, respectively and set the dummy variables in equation 3b equal to one on that day and the subsequent 20 trading days. On all other days, the dummies are equal to zero. Table 4 presents the results and shows the coefficient estimate and statistical significance for a hedge (second and third columns, respectively) and the total effects and the t-statistic for the marginal effect of the three crisis periods (following column pairs for the coefficient estimates and the t-statistics).

< Insert table 4 about here>

The total effect estimates for the crash in October 1987 indicate a zero correlation of gold and stocks for all markets with the exception of Canada and the US which exhibit a negative and statistically significant coefficient estimate. This implies that gold was a weak safe haven in all markets and a strong safe haven in Canada and the US. The results for the Asian crisis are more heterogeneous and exhibit larger absolute total effect estimates. The total effects estimates are positive and relatively large for several countries including Canada and the US implying that gold was not a safe haven despite the statistical insignificance. The nonsignificance of the coefficient estimates must be treated with more care due to the low number of observations (20 trading days). Finally, the third crisis period studied here shows a clear result for two groups of countries. The European countries and the US exhibit a strong safe haven effect with total effect estimates ranging from -0.72 (Italy) to -1.03 for Germany. Brazil, Canada and India also exhibit a safe haven effect. The other countries show a strong positive co-movement of gold and stocks, most notably Australia with a total effect estimate of 0.77. Since the change in the US Dollar price of gold is the same in each country within a certain period, the differences in the total effects are due to a different performance of the local stock market and changes in the exchange rate.

The safe haven effects in high volatility and financial crisis periods differ since the financial crisis episodes are relatively short (20 trading days) compared to the high volatility spells (at least 78 observations, one percent of the total number of daily observations). Since high
A similar number of trading days for a crisis period is used in the contagion literature (e.g. see Forbes and Rigobon, 2002). 19 Lehman Brothers collapsed on September 15, 2008.
18

17

volatility episodes include but also extend the crisis periods specified in the regression models, the results imply that the safe haven effect holds only for a certain number of days. The safe haven asset tends to exhibit negative returns thereafter due to its hedge property against stock markets. 20

IV.

CONCLUSIONS

This paper analyzes the role of gold in the global financial system. Specifically, we test the hypothesis that gold acts as a safe haven asset, offering protection to investors against losses in financial markets. The safe haven effect is found for most developed country stock markets. However, the findings are strongest for daily data, especially for extreme shocks occurring with a probability less than one percent. These results suggest that investors react to short-lived and extreme shocks by seeking out the safe haven of gold. In this context, gold can be seen as a panic buy in the immediate aftermath of an extreme negative market shock. More gradual trends in stock markets weekly or monthly losses do not appear to elicit the same impulsive response from investors. We also find evidence of a clear qualitative difference in the way investors react to shocks in developed and emerging markets. Gold is, at best, a weak safe haven for some emerging markets. This result fits with our hypothesis that the safe haven asset plays a relatively minor role in emerging markets. Investors suffering losses in emerging market stocks, rather than seeking an alternative haven asset, may simply readjust their portfolios towards the average by withdrawing from emerging markets in favour of developed market stocks. This reflects the contention of Calvo and Mendoza (2000) that investors worry not only about their absolute performance, but also about their performance relative to other investors. Looking at specific crisis periods, we find that gold was a strong safe haven for most developed markets during the peak of the recent financial crisis. Gold was also a strong safe haven for markets in the US and Canada during the 1987 stock market crash. However, the results for the Asian crisis do not clearly indicate a safe haven effect of gold for any of the markets tested. We also test the hypothesis that increased uncertainty not confined to specific crisis periods leads investors to seek out the safe haven. The results show that gold is a safe haven for increased levels of global uncertainty proxied by the conditional volatility of a world stock market index. However, gold is not a safe haven for extreme levels of global uncertainty. Rising uncertainty causes investors to seek out the safe haven but under extreme uncertainty
20

If gold is a hedge against stocks on average, gold exhibits negative returns if stock prices increase.

18

gold co-moves with stock markets establishing a market of one; that is, all assets move in the same direction. In our analysis, we distinguish between a weak and strong form of the safe haven effect. A weak safe haven will protect investors to the extent that it does not move in tandem with other assets in response to negative market shocks. The strong safe haven, by moving against other assets during periods of market stress, reduces overall losses for investors. We find that gold is a strong safe haven for most major developed world stock markets including the larger Eurozone markets (Germany, France and Italy), Switzerland, the UK and the US. Gold thus has the potential to act as a stabilizing force for the global financial system by reducing losses when it is most needed.

19

REFERENCES
Baur, D.G. and B.M. Lucey (2009). Is gold a hedge or a safe haven? An analysis of stocks, bonds and gold, Financial Review, forthcoming Boyer, B.H., Kumagai, T. and K. Yuan (2006): How Do Crises Spread? Evidence from Accessible and Inaccessible Stock Indices, Journal of Finance, 61, 2, 957-1003 Calvo, G. A. and E. G. Mendoza (2000).Rational Contagion and the Globalization of Securities Markets, Journal of International Economics 51(1), 79-113. Capie, F., T. C. Mills, and G. Wood (2005). "Gold as a Hedge against the Dollar", Journal of International Financial Markets, Institutions and Money 15(4): 343-52. Dungey, M., Fry, R., Martin, V. and B. Gonzalez-Hermosillo (2004): Empirical Modeling of Contagion: A Review of Methodologies," IMF Working Papers, 04/78, International Monetary Fund Economist (2009). Haring Away, February 29, 2009. Available at http://www.economist.com/finance/displaystory.cfm?story_id=13185396. Last accessed April 6, 2009. Economist (2005). The Little Yellow God, December 1, 2005. Available at http://www.economist.com/opinion/displaystory.cfm?story_id=E1_VNQGGPG. Last accessed April 6, 2009. Faugere, C. and J. Van Erlach (2006). The Price of Gold: A Global Required Yield Theory, Journal of Investing 14(1), 99-111. Forbes, K. and R. Rigobon (2002). No Contagion, Only Interdependence, The Journal of Finance 57, 5, 2223-2261 Gonzalo, J. and J. Olmo (2005). Contagion versus flight to quality in financial markets, Economics Working Papers 051810, University Carlos III Madrid. Gulko, L. (2002). "Decoupling: If the U.S. Treasury Repays Its Debt, What Then?" Journal of Portfolio Management 28(3): 59-66. Hartmann, P., S. Straetmans and C.G. de Vries (2004). "Asset Market Linkages in Crisis Periods", Review of Economics and Statistics 86(1): 313-26. Kaul, A. and S. Sapp (2006). "Y2K fears and safe haven trading of the U.S. dollar", Journal of International Money and Finance 25(5), 760-779 Lucey, B., E. Tully and V. Poti (2006). "International portfolio formation, skewness and the role of gold", Frontiers in Finance and Economics 3(1): 1-17. McCown, J.R. and J.R. Zimmerman (2006). Is Gold a Zero-Beta Asset? Analysis of the Investment Potential of Precious Metals, Working Paper Series. Available at SSRN: http://ssrn.com/abstract=920496.

20

Petty, W. (1690). Discourse on Political Arithmetick McMaster University Archive for the History of Economic Thought. Available online at: http://ideas.repec.org/b/hay/hetboo/petty1690.html. Ranaldo, A. and P. Sderlind (2007). Safe Haven Currencies Universitt St. Gallen Discussion Paper no. 2007-22 Sherman, E. (1982). "Gold : A Conservative, Prudent Diversifier", Journal of Portfolio Management (Spring): 21-27. Upper, C. (2000). How Safe was the "Safe Haven"?: Financial Market Liquidity During the 1998 Turbulences, Deutsche Bundesbank Working Paper No. 1/00. Available at SSRN: http://ssrn.com/abstract=219132. World Gold Council statistics available online at: http://www.research.gold.org/supply_demand/.

21

Figure 1: Gold and Commodity spot indices, 1979-2009 The figure shows the evolution of S&P GSCI commodity and gold spot indices over a 30 year sample period from March 1979 until June 2009 (monthly data).
1000

900

800

S&P GSCI Commodity Spot

700

S&P GSCI Gold Spot

600

500

400

300

200

100

0
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

22

Figure 2: Gold and Commodity spot indices June 2007 June 2009 The figure shows the evolution of S&P GSCI commodity and gold spot indices in the global financial crisis in 2007-2009 (monthly data).
1000

900

S&P GSCI Commodity Spot

800

S&P GSCI Gold Spot

700

600

500

400

300

200

100

0
Jul-07 Aug-07 Sep-07 Nov-07 Dec-07 Jan-08 Feb-08 May-08 Jun-08 Apr-08 Jul-08 Aug-08 Sep-08 Nov-08 Dec-08 Jan-09 Feb-09 May-09 Mar-08 Mar-09 Jun-09 Apr-09 Oct-07 Oct-08

23

Figure 3: World index and price of gold, full sample The figure shows the evolution of a world stock index and the price of gold in US$ over a 30-year period from 1979 until 2009 (daily data). World index level is labelled on left vertical axis and gold price is labelled on right vertical axis.

2000

1200

1800
WORLD

1000

1600

Gold Bullion (US$)

1400 800 1200

World Index Gold Price

1000

600

800 400 600

400 200 200

0
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

24

Figure 4: Conditional Volatility of daily world index returns and gold returns The figure shows the evolution of daily conditional volatility (GARCH(1,1) estimates) of the world stock index return and the return on gold denominated in US$ over a 30-year period from 1979 until 2009.
0.004

0.0035
h world

0.003

h gold

0.0025

0.002

0.0015

0.001

0.0005

0
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

25

Figure 5: World index and price of gold, 2007 2009 sub-sample The figure shows the evolution of a world stock index and the price of gold in US$ over a sub-sample period from 2007 until 2009 (daily data). World index level is labelled on left vertical axis and gold price is labelled on right vertical axis.
2000
WORLD
1200

1800

Gold Bullion (US$)

1000

1600

1400
800

1200
World Index Gold Price

1000

600

800
400

600

400
200

200

0
Oct-07 Mar-07 Mar-08 Oct-08 May-07 May-08 Aug-07 Sep-07 Nov-07 Dec-07 Aug-08 Sep-08 Nov-08 Dec-08 Mar-09 Jun-07 Jan-08 Feb-08 Jun-08 Jan-09 Feb-09 Apr-07 Jul-07 Apr-08 Jul-08

26

Figure 6: World index and price of gold, August 2008 December 2008 sub-sample The figure shows the evolution of a world stock index and the price of gold in US$ over a short period in 2008. The figure aims to show the time-varying co-movement of gold and stocks. World index level is labelled on left vertical axis and gold price is labelled on right vertical axis.

1600
WORLD

1200

1400

Gold Bullion (US$)

1000

1200
800

1000
World Index Gold Price

800

600

600
400

400
200

200

0
Oct-08 Oct-08 Oct-08 Oct-08 Aug-08 Aug-08 Sep-08 Sep-08 Sep-08 Sep-08 Oct-08 Nov-08 Nov-08 Nov-08 Nov-08 Dec-08

27

Figure 7: Rolling beta estimates of World index returns on gold returns The figure presents the evolution of the correlation of the global stock index with gold based on daily data. The rolling correlation estimates are based on a window length of 250 daily observations and illustrate that the correlation is changing through time. The regional indices (not displayed here for clarity of exposition) show qualitatively similar patterns through time. All indices are denominated in US$.
1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

28

Figure 8: Rolling beta estimate of US stock index returns on gold returns The figure presents the evolution of the correlation of the US stock market index with gold based on daily data. The rolling correlation estimates are based on a window length of 250 daily observations and illustrate that the correlation is changing through time. Other individual country indices (not displayed here for clarity of exposition) also display qualitatively similar patterns through time. All individual country indices are denominated in local currency, gold is denominated in US$.
0.8

0.6

0.4

0.2

-0.2

-0.4
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

29

Table 1: Descriptive Statistics The table presents the descriptive statistics of the world index and gold returns in US dollar and the returns of the stock market indices in local currency. The statistics are based on daily returns.
Obs World Index (in US$) Gold (in US$) AUSTRALIA BRAZIL CANADA CHINA FRANCE GERMANY INDIA ITALY JAPAN RUSSIA SWITZERLAND UK US 7826 7826 7826 3825 7826 4070 7826 7826 5000 7826 7826 3835 7826 7826 7826 Mean 0.0251 0.0170 0.0298 0.0522 0.0263 0.0274 0.0304 0.0179 0.0482 0.0326 0.0090 0.0702 0.0255 0.0278 0.0269 Std. Dev. 0.0083 0.0123 0.0107 0.0173 0.0094 0.0208 0.0118 0.0110 0.0176 0.0135 0.0117 0.0272 0.0096 0.0102 0.0108 Min -0.0976 -0.1787 -0.2944 -0.1055 -0.1165 -0.1429 -0.0989 -0.1214 -0.1825 -0.0984 -0.1574 -0.3062 -0.1231 -0.1301 -0.2071 Max 0.0817 0.1221 0.0793 0.1953 0.0896 0.1571 0.0992 0.1605 0.2572 0.1048 0.1229 0.3017 0.0981 0.0886 0.1090

30

Table 2a: Estimation results daily returns The table presents the estimation results for the role of gold as a hedge and safe haven asset. Negative coefficients in the hedge column indicate that gold is a hedge against stocks. Zero (negative) coefficients in extreme market conditions (quantile columns (0.10, 0.05 or 0.01)) indicate that gold is a weak (strong) safe haven. The results show that gold is a hedge and a strong safe haven for European countries and the US. Gold is not a hedge or a safe haven for emerging countries, Australia, Canada and Japan. Model : rGold, t = a + bt rstock, t + et bt = c0 + c1 D(rstock q10) + c2 D(rstock q5) + c3 D(rstock q1) ht = + et-1 + ht-1

daily frequency indices (in US$) Emerging Markets EMU EU Latin America North America World hedge 0.057 0.081 0.088 0.011 -0.064 0.062 0.10 0.121 0.142 0.138 0.020 0.005 0.126 0.05 0.034 0.122 0.160 0.065 -0.002 0.081 0.01 0.108 -0.087 -0.088 -0.028 -0.070 -0.015

*** *** *** *** ***

**

**

** *

** *** *** *** *** ***

markets

(in local currency) Australia Brazil Canada China France Germany India Italy Japan Russia Switzerland UK US

hedge 0.044 0.005 0.031 0.011 -0.047 -0.058 0.004 -0.078 0.010 0.003 -0.081 -0.067 -0.071

*** ** * *** *** ***

*** *** ***

0.10 0.022 -0.014 0.076 0.040 -0.037 -0.040 0.038 0.015 -0.004 0.016 -0.082 -0.029 -0.013

* ***

0.05 0.058 0.033 0.071 0.031 -0.030 -0.035 -0.002 -0.047 -0.016 0.010 -0.078 -0.050 0.000

**

**

0.01 0.045 0.001 -0.011 0.023 -0.096 -0.122 0.004 -0.147 0.090 0.011 -0.110 -0.094 -0.073

* *** *** *** *** *** * ** ***

***, **, * denote statistical significance at the 1%. 5% and 10% level

31

Table 2b: Estimation results weekly returns The table presents the estimation results for the role of gold as a hedge and safe haven asset. Negative coefficients in the hedge column indicate that gold is a hedge against stocks. Zero (negative) coefficients in extreme market conditions (quantile columns (0.10, 0.05 or 0.01)) indicate that gold is a weak (strong) safe haven. The results show that gold is a hedge and a safe haven for European countries and the US. Gold is not a hedge or a safe haven for emerging countries, Australia, Canada and Japan. Model : rGold, t = a + bt rstock, t + et bt = c0 + c1 D(rstock q10) + c2 D(rstock q5) + c3 D(rstock q1) ht = + et-1 + ht-1
weekly frequency indices (in US$) Emerging Markets EMU EU Latin America North America World hedge 0.097 0.121 0.136 0.046 -0.045 0.115 0.10 -0.082 0.186 0.105 0.058 0.130 0.087 0.05 0.075 0.049 0.062 0.012 -0.084 0.006 0.01 0.105 0.002 0.013 0.000 -0.077 -0.063

*** *** *** ** ***

**

** *

***

***

markets

(in local currency) Australia Brazil Canada China France Germany India Italy Japan Russia Switzerland UK US

hedge 0.063 0.036 0.159 0.035 -0.031 -0.064 0.010 -0.107 0.027 0.016 -0.127 -0.083 -0.025

** *** *** *** ***

*** ***

0.10 0.119 0.015 0.050 0.080 -0.013 -0.045 0.116 -0.068 0.072 0.067 -0.121 -0.245 -0.144

**

** *

0.05 0.133 -0.023 0.124 0.148 -0.062 0.026 -0.019 0.041 -0.166 0.027 -0.080 0.122 -0.153

*** * ***

***

0.01 0.006 0.076 0.049 0.071 -0.111 -0.164 -0.041 -0.156 0.059 0.031 -0.042 -0.030 -0.014

* **

*** *** ***

** ***

***, **, * denote statistical significance at the 1%. 5% and 10% level

32

Table 2c: Estimation results monthly returns The table presents the estimation results for the role of gold as a hedge and safe haven asset. Negative coefficients in the hedge column indicate that gold is a hedge against stocks. Zero (negative) coefficients in extreme market conditions (quantile columns (0.10, 0.05 or 0.01)) indicate that gold is a weak (strong) safe haven. The results show that gold is a hedge and a safe haven for European countries and the US. Gold is not a hedge or a safe haven for emerging countries, Australia, Canada and Japan. Model : rGold, t = a + bt rstock, t + et bt = c0 + c1 D(rstock q10) + c2 D(rstock q5) + c3 D(rstock q1) ht = + et-1 + ht-1
monthly frequency indices (in US$) Emerging Markets EMU EU Latin America North America World hedge 0.149 0.072 0.108 0.145 -0.097 0.123 0.10 0.154 0.142 0.073 0.026 -0.199 0.026 0.05 0.044 -0.164 -0.163 -0.017 -0.074 -0.014 0.01 0.620 0.330 0.144 0.586 0.093 0.238

** * *** *

*** **

markets

(in local currency) Australia Brazil Canada China France Germany India Italy Japan Russia Switzerland UK US

hedge 0.059 0.099 0.130 0.051 -0.061 -0.064 0.017 -0.105 0.023 -0.009 -0.168 -0.075 -0.144

* * *

**

** **

0.10 0.140 0.223 0.107 0.066 0.056 -0.161 -0.052 0.006 0.089 -0.043 0.128 0.051 -0.215

0.05 0.260 0.007 0.068 0.117 -0.076 -0.101 0.043 -0.046 -0.187 0.019 -0.096 -0.134 -0.010

0.01 -0.057 0.053 -0.103 0.496 -0.073 -0.234 0.060 -0.236 0.347 0.027 -0.148 -0.098 -0.130

**

***, **, * denote statistical significance at the 1%. 5% and 10% level

33

Table 3: Estimation results uncertainty The table presents the estimation results for the role of gold as a hedge and safe haven asset in periods of increased volatility as a proxy for uncertainty. Negative coefficients imply that gold is a hedge on average (second column) or a safe haven in periods of increased or extreme volatility. The results show that gold is a hedge in European markets and the US and a safe haven in periods of increased volatility (columns 4-7) in most markets. In contrast, for the majority of markets, gold is not a safe haven in periods of extreme volatility (last two columns). The total effect is the sum of the hedge coefficient (c0) and the marginal effect. The t-statistics refer to the marginal effect. Model : rGold, t = a + bt rstock, t + et bt = c0 + c1 D(hstock q90, t-1) + c2 D(hstock q95, t-1) + c3 D(hstock q99, t-1) ht = + et-1 + ht-1
hedge coeff. AUSTRALIA BRAZIL CANADA CHINA FRANCE GERMANY INDIA ITALY JAPAN RUSSIA SWITZERLAND UK US 0.14 0.04 0.16 0.04 -0.04 -0.05 0.02 -0.03 0.05 0.03 -0.07 0.02 -0.01 t-stat. 8.71 3.18 8.32 3.97 -2.85 -2.98 2.56 -2.35 3.40 4.07 -3.73 1.19 -0.64 world vola > 90% total effect 0.08 0.07 0.14 0.04 -0.02 -0.05 0.03 -0.06 -0.03 0.02 -0.08 0.00 -0.02 t-stat. -1.03 1.02 -0.48 0.03 0.60 -0.05 0.40 -0.75 -2.10 -0.42 -0.23 -0.54 -0.32 world vola > 95% total effect -0.01 -0.06 -0.02 -0.02 -0.05 -0.06 -0.02 -0.07 -0.01 0.00 -0.07 -0.05 -0.09 t-stat. -1.31 -3.16 -2.53 -1.70 -0.72 -0.15 -1.34 -0.21 0.44 -0.75 0.21 -0.76 -1.25 world vola > 99% total effect 0.11 0.04 0.13 -0.02 0.04 0.06 0.11 0.11 0.14 0.10 0.03 0.11 -0.02 t-stat. 2.24 2.47 2.69 -0.01 1.91 2.28 2.93 3.45 3.08 3.70 1.89 2.95 1.52

number of obs

783

391

78

***, **, * denote statistical significance at the 1%. 5% and 10% level

34

Table 4: Estimation results financial crisis The table presents the estimation results for the role of gold as a hedge and safe haven asset in periods of financial crises (October 1987, Asian Crisis October 1997 and Financial Crisis of 2008 Lehman Brothers bankruptcy September 2008). The results show that gold is a hedge on average and in financial crises in particular in the global financial crisis in September 2008 (last two columns). The safe haven effect of gold in October 1987 is confined to Canada and the US and the results for the Asian crisis do not clearly indicate a safe haven effect of gold. The crisis periods are modelled through a dummy variable which is set equal to one for 20 trading days after the crisis start and equal to zero for all other observations. The total effect is the sum of the hedge coefficient (c0) and the marginal effect. The t-statistics refer to the marginal effect. Model : rGold, t = a + bt rstock, t + et bt = c0 + c1 D(Asian crisis, 1997) + + cn D(subprime crisis, 2008) ht = + et-1 + ht-1
hedge October 1987 Asian Crisis (Oct 1997) t-stat. -0.96 -2.61 -0.75 0.80 0.24 0.68 0.57 -0.10 -2.52 total effect 0.04 0.08 -0.03 -0.07 0.03 -0.01 0.09 -0.01 0.03 -0.15 0.24 -0.03 0.04 0.05 0.16 0.00 0.06 -0.02 0.03 0.14 0.17 t-stat. -0.64 1.32 0.64 -1.07 0.51 0.69 0.71 0.24 0.08 -1.35 0.47 0.51 1.43 Financial Crisis (Lehman Sept 2008) total effect 0.77 -0.31 -0.18 0.24 -0.86 -1.03 -0.08 -0.72 0.42 0.68 -0.90 -0.97 -0.75 t-stat. 2.87 -2.32 -1.22 2.08 -3.69 -3.96 -0.89 -2.93 1.78 2.88 -2.91 -4.20 -3.31

coeff. AUSTRALIA BRAZIL CANADA CHINA FRANCE GERMANY INDIA ITALY JAPAN RUSSIA SWITZERLAND UK US 0.11 0.01 0.14 0.01 -0.03 -0.03 0.01 -0.03 0.04 0.02 -0.04 0.04 -0.02

t-stat. 5.88 0.83 6.49 1.47 -1.64 -1.57 1.30 -2.35 2.67 2.48 -1.93 1.94 -1.21

total effect 0.06

***, **, * denote statistical significance at the 1%. 5% and 10% level

35

Institute for International Integration Studies

The Sutherland Centre, Trinity College Dublin, Dublin 2, Ireland