Does the global stock market influence the price of gold In the long run?

(1)

Does the Global Stock Market Influence

the Price of Gold In the Long Run?

Name Joop Noorlander

Student number 10752110

Programme BSc Economics and Business

Specialization Finance and Organization Coordinator dr. P.J.P.M. Versijp

Supervisor dr. P.J.P.M. Versijp

Date January 2017

A B S T R A C T

This thesis investigates the effect of the global stock market on the gold price in the long term by using the implied convenience yield as dependent variable. The underlying economical mechanism is based on investors’ expectations of a crisis, which may influence their investment behaviour. In 5 different regression equations with data from 1979-2016; the silver price, the dollar/euro exchange rate, the inflation rate, the gold/stock ratio and two dummies indicating high or low storage costs are interchangeably used as control variables. Next to this, five different methods of period selection have been identified, resulting in a total of 25 unique regressions that have been run by using OLS. With the data available, contradicting results have been found: In the regression with daily data, the global stock market has positive and significant coefficients. However, the monthly regression results in coefficients that are not significantly different from zero. Further, the effects of adding the gold/stock ratio and storage costs dummies are investigated. For the gold/stock ratio, no significant effects are found. The storage costs dummies appear to be of some influence.

(2)

1. Introduction

There has been a significant amount of research regarding the short-term influence of the stock market on the price of gold, for example the ‘safe-haven effect’ or gold as a hedge against market- and dollar movements (Baur & McDermott, 2010; Baur & Lucy, 2010; Reboredo, 2013). Furthermore, there is a great amount of literature regarding the long-term influences on gold prices like interest rates, inflation rates and the dollar value. However, literature on long-term effects of the global stock market on the gold price is scarce. This thesis will try to fill this gap by investigating if the global stock market is a long-term indicator variable for the gold price. The following research could contribute to a better understanding of the drivers of the gold price, which is therefore useful in making more precise predictions about the gold price in the future. In the world of investing, the precision of a prediction is of major importance and can lead to superior returns. Prior to investigating this, a valid economical mechanism that supports the causality must be presented. In the following paragraph, this economical mechanism will be explained. Thereafter, the research question will be posed.

Suppose that the stock market is in an ongoing upward trend causing investors to fear an inflating price bubble and a subsequent crash. In anticipation of such an event, some investors could be tempted to replace (a part of) their stock with gold. The reason for this behavior can be found in the characteristics of the asset. Namely, in specific circumstances, gold has no (or negative) correlation with the stock market. This makes gold a proper investment to protect/hedge investors’ portfolio from a possible downturn of the market. This is confirmed by the research of Baur and Lucy (2010). In their article, they define a safe haven as “an asset that is uncorrelated or negatively correlated with another asset or portfolio in times of market stress or turmoil.” They conclude that gold serves the role of a ‘safe haven’ effect for a limited time after an extreme negative shock. Combining their conclusion and definition, this implies that in periods of market stress or turmoil, gold indeed has no- or negative correlation with the stock market. If the investor’ expectations concerning a price bubble surpasses and the stock market crashes, the price of the owned gold will probably remain stable or even increase. To find if this ‘anticipation’ effect is playing a roll in the determination of the gold price, the following research question has been drafted:

Did the MSCI world returns in bull-market periods from January 1979 until December 2016 have a positive effect on the gold price in the long-term via an increase in the convenience yield of gold and therefore can be identified as an indicator variable for the long-term price of gold?

As can be seen in the above research question, this thesis will use the convenience yield as a measurement of the effects of the stock market on the price of gold. According to Lee & Lee (2006),

(4)

the convenience yield is “a measure of the benefits from ownership of an asset that are not obtained by the holder of a long futures contract on the asset.” By using the convenience yield to investigate the research question, an important part of the causality is left unexplained. Namely, why would investors, when expecting a crisis, rather invest in the commodity instead of long futures contracts so that the convenience yield increases?

There are several reasons for this. Remember that the causality of this article is based on the investors’ expectations of a ‘bubble’ and a subsequent crash. Of course, this expectation is part of a bigger set of expectations. Suppose that investors expect the market to crash with a probability of X% and the market to continue rising/remain stable with (1-X)%. This X% chance might give enough incentive to hedge their portfolio by buying (extra) gold. However, there is still a (1-X)% chance that no crash or bubble bust occurs. In this last scenario, the course of the gold price is rather ambiguous. This is due to a high short-term volatility of the gold price (Demidova-Menzel & Heidorn, 2007). For future contracts, this last scenario can cause severe problems. Namely, when investing in gold future contracts, one only has to pay the margin upfront. This margin is usually around 5%-10% of the future face value. Therefore, per invested dollar, the exposure to the gold market is around 10-20 times as great as when investing in the gold asset itself. In the case of a decrease in the price of gold (that is perfectly possible in the above explained situation), an investment in a future contract will decrease in value more than that of a gold commodity, due to greater exposure. Besides, a top-up margin is required in the case of a (large) downturn in the underlying asset. A short-term decrease in the gold price can therefore cause problems with liquidity or even cause bankruptcy, while this risk does not apply for investing in the commodity. Second, gold future contracts have a higher volatility than that of gold and therefore a higher risk (Narayan et al., 2010).

Thus, because of the characteristics of a gold future (big exposure, higher volatility) combined with this specific situation wherein a decrease of the gold price is possible, investing in gold rather than in gold future contract becomes a more sensible decision. Storage- and transaction costs associated with investing in the commodity, but not in the gold future contract, are still present. Since these two costs are likely to remain relatively stable, the overall convenience yield should increase. Therefore, it seems valid to use the convenience yield in this research.

The remainder of this thesis will exist of four sections. In the first section, the literature that is essential for this thesis will be reviewed. Then — in the methodology section — the research method, regression equation, statistical hypotheses, data and data mutations are explained. In the next section, the tabulated results will be displayed and analyzed. The last section will exist of a summary, some concluding remarks and suggestions for improvements or further research.

(5)

2. Literature Review

The first article essential for this thesis is written by Fama & French (1987) and looks into the pricing of commodity futures. According to Fama & French, two common models for the pricing of commodity futures exist: the theory of storage and the expectations/premium theory. In this article they conclude that in the case of precious metals, the ‘theory of storage’ is a precise estimate of the future price. This theory makes use of the no-arbitrage principle. Simply said, this theory states that the difference between commodity future prices, F(t, T) and commodity spot prices S (t) is equal to the forgone interest, R(t)*R(t, T), plus the costs of storage W(t, T), minus the convenience yield of owning an additional unit of the commodity C(t, T) [1]. This formula can be rewritten to obtain a formula for the implied convenience yield [2]. Since no raw data on the convenience yield is available, this formula will be used for finding the daily convenience yield.

𝐹 𝑡, 𝑇 − 𝑆 𝑡 = 𝑆 𝑡 ∗ 𝑅 𝑡, 𝑇 + 𝑊 𝑡, 𝑇 − 𝐶 𝑡, 𝑇 . [1] 𝑦!= 𝐶 𝑡, 𝑇 = 𝑆 𝑡 ∗ 𝑅 𝑡, 𝑇 + 𝑊 𝑡, 𝑇 − [𝐹 𝑡, 𝑇 − 𝑆 𝑡 ] [2]

Equation [2] is fully in line with the findings of Casassus & Collin-Dufresne (2005). In their research, they conclude that in the case of gold, the convenience yield has a positive (significant) dependence on the interest rate. This means that in the case of rising interest rates, the convenience yield becomes higher as well. Further, they state that convenience yields have a positive relation with spot prices. This can be seen in the formula as well. This means that, if the research finds that the stock market has a positive and significant influence on the convenience yield, spot prices are also increasing.

However, due to a lack of data, the storage costs cannot be found. Therefore, the storage costs will be left out of the formula. This will cause a negative bias in the convenience yield. To correct somewhat for leaving out the storage costs, two dummies indicating months with very high or very low storage costs can be added to the regression equation. To identify these months with very high or very low storage costs, the producer price index of General Warehousing and Storage Services is used. First, the average monthly increase in the producer price index data set is calculated (0.043%). Then, this average monthly rate is used to simulate an index that shows how the development of the storage costs would have looked liked in the case of linear growth. The standard deviation in the real index is approximately 6.35. The simulated index will now be compared with the real index. It is remarkable that the real index exceeds the simulated index by at least one standard deviation 56 times (of 283 periods). However, this does not happen once when comparing the indexes in reverse order, causing the creation of two separate decision rules for determining high and low storage costs.

(6)

In order to keep things as simple as possible, it seems appropriate to use the one standard deviation rule for the high storage costs group. This means the following: periods wherein the real index exceeds the simulated index by at least one standard deviation (6.35) are put in the ‘high storage costs’ category. As stated before, this rule generates 56 periods (of 283). See the decision rule below:

𝑅𝑒𝑎𝑙 𝑖𝑛𝑑𝑒𝑥 − 𝑆𝑖𝑚𝑢𝑙𝑎𝑡𝑒𝑑 𝑖𝑛𝑑𝑒𝑥 > 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛_!"#$ (6.35)

However, this does not work for the low storage costs, since the use of one standard deviation decision rule does not generate any fitted period. Therefore, the decision rule for the low storage costs period will be somewhat eased. In order to generate approximately the same amount of periods as in the high storage costs situation, in addition to satisfying the fact that the periods must have low storage costs, it seems appropriate to use the decision rule below [4]. This rule generates 24 periods of 283. Notice that periods that fulfill one of both decision rules are labeled 1 and the others with a 0.

𝑆𝑖𝑚𝑢𝑙𝑎𝑡𝑒𝑑 𝑖𝑛𝑑𝑒𝑥 − 𝑅𝑒𝑎𝑙 𝑖𝑛𝑑𝑒𝑥 > 0.25 ∗ 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛_!"#$(1.5875)

Calculating the convenience yield by using the above-described formula [2], however without storage costs, results in a dataset that does not meet the expectations. Firstly, because the daily convenience yield has a standard deviation of approximately 4.472. For the monthly convenience yield, this is 12.454 (See table I). Remember that the convenience yield measures the benefits of holding the commodity instead of holding a long futures contract. This measure, corrected for interest rates, should not vary much since there seems to be no eligible explanation. The absence of storage costs is not likely to be the missing link in this puzzle. Storage costs are usually not that volatile and therefore are not an explanation of why over-day differences, greater than 10 units, occur frequently. However, this thesis is not the first to encounter these very high levels of convenience yield volatility. Gibson and Schwartz (1990) find the same result while investigating convenience yields and the pricing of oil. In their research, they state that the very high levels of the standard deviation seem to “be a function of the number of sharp oil price declines or increases observed.” Taking into account the high short-term volatility of the gold price (Demidova-Menzel & Heidorn, 2007), this sounds like a plausible explanation. Secondly because of the negative mean of the daily convenience yield. However, this negativity may be explained by the absence of storage costs in the calculation of the convenience yield, since there is a negative bias present in formula [2].

Table I: Descriptive Statistics Convenience Yield

Mean Sample Standard Deviation N

Daily -0.5383192 4.471729317 9908

(7)

In the following section, the remaining part of the regression equation is presented. Since we already accounted for changes in the interest rate and/or storage costs in the implied convenience yield, all remaining effects on the convenience yield come from the difference between the gold future- and gold spot prices. Therefore, influences need to be found that make this difference increase/decrease.

The first coefficient, 𝛽!, is simply the constant. Then, there is the 𝛽!, which is the coefficient of

interest. As explained on the first page, the explanatory variable 𝑋!! will represent the global stock

market index returns. The logic behind this is as following: Investors who expect a crisis in the (near) future are willing to invest in gold rather than gold futures. This will drive up the convenience yield and increase gold spot prices. Therefore, a positive relation with the independent variable is expected. Furthermore, control variables 𝑋!! up to 𝑋!" represent other possible influences on changes in the

convenience yield. An example is the dollar/euro exchange rate and dollar value (Sjaastad & Scacciavillani, 1996; Tully & Lucey, 2007). Since gold is (used as) a hedge against dollar movements (Capie et al., 2005; Reboredo, 2013), changes in the dollar value (relative to other currencies) could have an impact on the gold price. However, this impact might not be seen in the price of gold futures and therefore, a difference in the convenience yield could arise.

Next, Sari et al. (2010) state that both gold and silver are safe havens. There might be a negative effect in the convenience yield due to the two assets being substitutes. If the price of silver drops, investors may want to replace their silver by gold. It would make sense that the investors are changing their commodity silver for commodity gold. This would make the convenience yield go up. Another control variable will check if investors truly replace their stocks by gold. In order to find this, this research will look into the ratio between money invested in gold versus stocks [3].

!"#$%&'$"& !"#$ !"#$%& !"#. !"#$%& [3]

There is no simple data available on the amount of money invested in gold. However, in Thomson Reuters’ GFMS, the yearly change in gold bar demand for investment is reported. By combining this data with the total amount of gold used for investment purposes in 2015, a dataset that tracks the amount of gold invested for each year can be created. Then, these yearly amounts are multiplied with the average gold price in that specific year to obtain the total amount of money invested in gold. The average gold price is based on the monthly gold prices from the London Bullion Market. By dividing this amount of invested gold against the market capitalization of stocks, the gold/stock ratio is found. Positive changes in this ratio would imply that investors are, as causality predicts, replacing some part of their stock for gold. Since the data is yearly, using it in the daily regression (or even in the monthly robustness check) will cause a bias due to the serial correlation.

(8)

Therefore, a separate (monthly) regression will be carried out. The way this yearly data is converted to monthly data is by matching the yearly gold/stock ratio to all months in that specific year, what will most likely impose a bias. These regression results will then be compared to the original monthly regression, without gold/stock ratio. Notice that this data is available up to the end of 2015, but will be compared to regression results that work with data up to the end of 2016.

The last control variable is the US inflation rate. Worthington and Pahlavani (2007) find that “investments in gold can serve as an inflationary hedge.” A higher inflation rate therefore could lead to a higher demand of gold. This may enlarge the price difference between the asset and the futures contract, and therefore the convenience yield. Both the influences of the interest rate as well as the real dollar value on the gold price have been confirmed by the research of Akram (2009). However, only monthly data is available regarding the different inflation measures (CPI, PPI, etc.). This coefficient will therefore only be used in a robustness check that will be done by using monthly data.

3. Methodology

In this section, the methodology will be explained in further detail. First, the regression equation based on the information obtained in the literature review section will be presented. Hereafter, different policies on which periods to research will be discussed. This is followed by information regarding the data and this section will end by stating the hypothesis.

Regression equation

Taking together all the elements described in the literature review section, the regression equation [4] below is formed. Notice that one part of the regression equation is between parentheses. These parentheses indicate that the variables in between are not part of every regression. As explained in the literature section, the US inflation rate consists of only monthly data. Therefore, this coefficient will only be included in regressions that make use of monthly data. The gold/stock ratio will be solely part of a separate set of regressions (with monthly data) to look into the effects of this specific variable. The same goes for both variables regarding the storage costs. However, since the data on storage costs date back to 1993 instead of 1979, another set of monthly regressions without these storage costs is carried out in order to be able to compare them. The other coefficients will be part of both the daily regression as well as the monthly robustness check. The regression will be done by applying linear ordinary least squares in Stata. Since there might be an autocorrelation, due to leaving out the storage costs, the robust option will be used in every regression.

𝐶𝑜𝑛𝑣𝑒𝑛𝑖𝑒𝑛𝑐𝑒 𝑦𝑖𝑒𝑙𝑑 = 𝛽!+ 𝛽!∗ 𝑀𝑆𝐶𝐼 + 𝛽!∗!"##$%_!"#$ + 𝛽!∗ 𝑆𝑖𝑙𝑣𝑒𝑟 + (𝛽!∗ 𝑈𝑆 𝐼𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛 + 𝛽!∗_!"#$%!!"#$ 𝑟𝑎𝑡𝑖𝑜 + 𝛽!∗

(9)

Periods

Since the research question needs further refinement by determining the exact periods for investigation, another essential article to this research is that of Fabozzi and Francis (1977). In this article, they define multiple ways of identifying bull and bear markets. It is unnecessary to limit the research to only periods that are followed by one, since the above-explained causality is based on expectations of a crisis. Therefore, all periods with a stock market that is substantially increasing will be taken into account. However, to be able to conclude that this model truly confirms the explained mechanism, it is necessary to check for the same effect in a control group. The easiest measure to identify these periods is to use bull/bear market indicators. In the article of Fabozzi and Francis (1977), three different definitions of bull/bear markets are explained:

The first method, named ‘bull and bear’ (hereafter BB) puts all months with a declining stock market in the bearish category and puts most months with a rising stock market in the bullish category. However, if a bullish month lies amidst two bearish months, it is identified as a bearish month as well. This method will help to identify trends. The second method, called ‘up and down’ (hereafter UD), ignores trends and checks every month independently. This method is the most simplistic and divides months based upon its monthly return. When this is negative, the month is labeled bearish and when this is positive, the month is labeled bullish. The last method, ‘substantial up and down’ (hereafter SUD), is an expansion on the UD method. Now, instead of just looking at the sign of the market return, also the magnitude of the return is important. Months that have a return that is substantial will be labeled bullish or bearish, dependent on the market going up or down. To be put in the substantial category, the absolute market return of a month has to be larger than half of one standard deviation of the market’s return measured over the entire sample period [3]. The months that do not fulfill this criterion will be placed in a rest category, and will not be reviewed. Nonetheless, in this thesis, the MSCI world index will be used to identify the bull and bear market trends.

𝑅_!"# > 0.5 ∗ 𝜎_{!"# !"#$%! !"#$%&} [3]

This research focuses on periods with an upward trend. Therefore, it is important to pick a bull/bear identifier that takes trends into account. Besides, expectations of an inflating bubble will be greater when market returns are substantial. It seems most appropriate to use a combination of the BB and SUD methods. At first a selection will be formed by using the BB method. This will cancel out all the bearish periods as well as bullish months that lie in between two bearish months. Then, the SUD method will be applied. It is expected that applying the SUD will remove ‘weak’ months that lie in the start of a bullish trend because markets need time to reverse. Further, it will cancel out the very weak trends that were not canceled out by the BB method. However, if there is a non-substantial

(10)

bullish month in the middle of a bullish trend, this month should not be removed since it is part of a larger upward trend. Deciding which periods to rule out by the SUD method and which to keep using, the following decision rules have been determined.

1. For trends that last two months containing:

• Two substantial months; the trend will be fully used in the sample. • Zero substantial months; the trend will be fully left out of the sample. • One non-substantial and one substantial month;

a) If the substantial month is the first; include both months in the sample.

b) If the substantial month is the second; include only the second month in the sample

This last decision rule needs some further explanation. The underlying reason for the difference in treatment between the (a) and the (b) situation is as following: When the substantial month comes right after a (substantial) negative month, this can surprise investors since markets usually need some time to reverse. This large change might trigger some fear for an inflating bubble. Since the up following month is increasing as well, although not substantial, this still could indicate a trend. Therefore, it will be useful to check both months. This differs for the reversed scenario. Expectations of a crisis are not likely to be formed in a non-substantial month that takes place after a bear period. However, when this is followed up by a substantial month, this could indicate a trend and therefore fear for a bubble can arise. Only the second month will be checked for in the sample. The same logic is used for creating the decision rules for more-than-two-month periods.

2. For trends that last more than two months:

• Non-substantial months in the beginning of a trend, up to the first substantial month, will be ruled out from the sample.

a) If a trend solely exists of non-substantial months, the entire trend will therefore be ruled out.

b) If only one substantial month remains due to the above correction, only this month will be taken into account

• Non-substantial months in the middle or at the end of a substantial trend will always be taken into account.

Next to the mix of the BB and SUD methods, the three basic methods obtained from Fabozzi and Francis (1977) will be tested. It is expected that the BB/SUD mix will give more significant results compared to the other bull/bear market identifiers (especially the simple UD) due to period selection that is more in line with the causality.

(11)

Finally, it will be useful to check for the same result within a control group. The control group needs to be a selection of periods that, according to the causality, should not have the effects that is searched for in the other groups. Considering this thesis works with expectations of and anticipation on a possible bubble and a subsequent crisis, these effects are not likely to be found in periods with a decreasing stock market. However, the substantially declining periods are left out of the control group as well. This is because of the safe-haven effect. Like explained before, this safe-haven effect can have a positive influence on the gold price in the short-term, due to a negative correlation between gold and stocks (Lucy & Baur, 2010). Since the regression is looking for long-term effects, these substantial down periods will create bias within the findings of the control group. Therefore, the control group will exist of non-substantially declining periods. After the above selection, the control group exists of 87 months.

Data:

All data is modified in such a way that the first data point matches the one of the restricting factor. In this research, the restricting factors were the datasets of the London Bullion Market (Gold and Silver prices). No other modifications have been applied to the data.

Implied convenience yield data Gold futures

Since gold futures mature only at specific dates, the exact prices for the remaining dates cannot be determined. Gorton and Rouwenhorst (2006) solve this data problem by selecting for each day a future contract with the closest expiration date (front-month contract). In other words, they choose the contract that has the shortest time of maturity. This approach is used in continuous future contracts as well. However, if the day and the expiration date of the selected contract lie in the same month, Gorton and Rouwenhorst (2006) use the up-following futures contract (the #2 contract or back-month contract). Using a continuous gold future will be the wisest option since replicating this method of matching dates with future contracts will be a huge amount of work. The CMX 100 OZ. Continuous Settlement price will be used in this research. This contract trades on the New York Mercantile index.

Data obtained from Datastream (daily/monthly)

Name: CMX-GOLD 100 OZ CONTINUOUS - SETT. PRICE (dollar denominated) Code: NGCCS00

Daily: 02-01-79 until 22-12-16 (9908 data points) Monthly: 31-01-79 until 30-11-16 (455 data points)

(12)

Gold commodity

For the spot price of gold, the Handy & Harman Base (hereafter HHB) or the London Bullion Market (hereafter LBMA) can be used. Both datasets have their advantages: HHB is based in the US whereas the LBMA is based in London. From this perspective, it will be most fitting to use the HHB. However, by using the LBMA, less negative results for the convenience yield are obtained. Also, the volatility of the convenience yield is lower by using the LBMA. All together, the LBM will be used due to the fact that using this dataset results in more sensible data for the convenience yield.

Data obtained from Datastream (daily/monthly) Name: Gold Bullion LBM U$/Troy Ounce Code: GOLDBLN

Interest rate

The interest rate will be used to obtain the implied convenience yield of gold. In the formula, it corrects for the opportunity costs of investing in gold as a commodity rather than investing in gold future contracts (margin-investing). Since gold is a relative safe investment, it is most appropriate to use the risk-free interest rate for determining the opportunity costs. The interest rate that is perceived as closest to risk-free is the 3-month US Treasury bill. Therefore, the 3-month US Treasury Bill

(secondary market) middle rate will be used.

Data obtained from Datastream (daily/monthly)

Name: US T-BILL SEC MARKET 3 MONTH (D) - MIDDLE RATE Code: FRTBS3M

Storage costs

This data can be used to create dummies for periods with very high or very low storage costs. Information on the specific decision rule is explained in the Literature Review section.

(13)

Name: Producer Price Index by Industry: General Warehousing and Storage: General Warehousing and Storage Services, Index Jun 1993=100, Monthly, Not Seasonally Adjusted

Code: PCU4931104931101

Monthly: 01-06-1993 until 01-12-2016 (283 data points)

Regression equation data: Global stock market index

The MSCI world index is a good proxy for the global developed market and covers around 85% of the free float-adjusted market capitalization in these developed countries.

Data obtained from Datastream (daily/monthly).

Name: MSCI WORLD U$ - PRICE INDEX (dollar denominated) Code: MSWRLD$

Exchange rate Dollar/Euro

Data obtained from Datastream (daily/monthly) Name: US $ TO EURO (WMR&DS) - EXCHANGE RATE Code: USEURSP

US inflation

Data obtained from Datastream (monthly) Name: US CPI - ALL URBAN: ALL ITEMS SADJ Code: USCONPRCE

Monthly: 15-01-79 until 15-11-16 (455 data points)

Silver

Data obtained from Datastream (daily/monthly) Name: Silver, Handy&Harman (NY) U$/Troy OZ Code: SILVERHD

(14)

Gold/stock ratio Investment gold:

Data obtained from Datastream (yearly) and World Gold Council (amount of investment gold in 2015) Name: WD GOLD: DEMAND - BAR INVESTMENT, WORLD VOLN

Code: WDCGTDBY

Yearly: 1979 until 2015 (37 data points)

Market capitalization stocks:

Data obtained from The World Bank (yearly)

Name: Market capitalization of listed domestic companies (current US$) Code: CM.MKT.LCAP.CD

Yearly: 1979 until 2015 (37 data points)

Hypothesis 1 𝐻!: 𝛽!≤ 0

𝐻_!: 𝛽_!> 0

The logic behind the equation would now be as following: Controlled for other explanatory variables, the effect of stock market prices on the convenience yield will be measured. If this is a positive and significant effect (and we can therefore reject 𝐻!), this means that the convenience yield of gold will

be higher when stock market price level increases. This would imply the following: when rising stock market prices cause a higher convenience yield, investors are anticipating a future crisis.

Hypothesis 2 𝐻_!: 𝛽_!!"#/!! = 𝛽_!!"

𝐻_!: 𝛽_!!"#/!! > 𝛽_!!"#_/𝛽

!!! > 𝛽!!"

This hypothesis is looking into the different methods of identifying the periods to research. It is expected that by selecting periods based on a combination of criteria obtained from both the SUD and BB methods (explained in more detail in the methodology section ‘periods’), a more significant result will be found than by using the rather simple UD method or one of both ingredients of the combination itself. This is because of the period selection. When using the simple UD method, all upward periods are taken into the sample. However, it would make sense that investors’ expectations of a crisis are bigger in upward trends and substantial upward periods. The same goes for the SUD and

(15)

MSCI 0.0008094 0.0010941 0.0008403 0.000873 0.0012269 (4.04)*** (3.70)*** (2.95)** (2.50)** (2.32)** DOLLAR/EURO -0.7977806 0.1674906 -0.3203242 -0.2768189 -0.1440222 EXCHANGE RATE (-2.32)** (0.41) (0.67) (-0.47) (-0.30) SILVER -0.0105668 -0.083979 -0.0543592 -0.0584363 0.0029085 (-0.63) (-2.28)** (-1.95)* (-1.88)* (0.05) CONSTANT -0.174267 -0.867132 -0.3144028 -0.3631723 -1.087009 # OF OBSERVATIONS 5869 2329 2979 2415 1087 R-SQUARED 0.0085 0.0201 0.0117 0.0125 0.0118

the BB methods as separate identification methods, however, to a lesser extent then the simple UD method. By taking into account the trend as well as the substantial aspect instead of just one of both, a more significant (or less insignificant) effect is expected to be found. If the 𝐻! is indeed rejected, this

basically says that the causality makes sense since it implies that having a stronger focus on months that support the causality results in a larger effect.

4. Results and Analysis

In this section, the results will first be displayed and explained. Then, the results will be discussed. In table II, the coefficients for the regressions with daily data are presented. The first column contains the names of these coefficients. In the columns thereafter, the coefficients as well as their T-score and the level of significance are displayed. A single asterisk indicates a significance of 10%, whereas two and three asterisks respectively indicate a 5% and 1% level of significance. For every method of period selection, a different column is used. In the bottom of the table are two descriptive statistics that give an impression about the amount of data points used and about the amount of variance in the convenience yield explained by the model. The results will be discussed in the same order as they appear in Table II.

TABLE II

THE EFFECT OF GLOBAL STOCK MARKETS ON THE CONVENIENCE YIELD (DAILY) DEPENDENT VARIABLE = IMPLIED CONVENIENCE YIELD

OLS WITH ROBUST STANDARD ERRORS T-SCORES REPORTED IN PARATHESES

(16)

The first method that will be discussed is labeled ‘Simple’ since it makes use of the most simplistic period identification method. The MSCI coefficient of 0.0008094 gives a T-score of 4.04, which is significant at the 1% level. This coefficient is positive as well, which is exactly like theory predicts. Then, the Dollar/Euro exchange rate has a coefficient of -0.797806 and a negative T-score of -2.31, which is significant at the 5% level. Again, as described in the literature review section, this is in compliance with theory. An increase in the dollar/euro exchange rate means that the value of the dollar relative to the euro drops. Since gold is usually seen as a hedge against dollar value (see literature review), this should increase the price of gold while this might not be true for gold futures. Thus, the convenience yield should increase. Silver prices do not seem to influence the convenience yield in this sample, as the coefficient is not significantly different from zero for any significance level. The constant is -0.1743.

The second method, named ‘Substantial’, brought up a similar result regarding the MSCI coefficient as the ‘Simple’ method did. The coefficient is a little higher in the substantial situation, namely 0.0010941. The T-score of 3.70 however indicates a little lower significance, although still significant at the 1% level. Regarding the Dollar/Euro exchange rate, the substantial method came up with a positive, although not significant coefficient. The silver price on the other hand was significant on the 5% level and had a negative coefficient of -0.083979. This is in accordance with the theory in the literature review. Since gold and silver are both precious metals, they could be substitutes when thinking about investing in precious metals. Now, a drop in the price of silver might lead to more gold demand by investors, which increases the price of gold. This will then lead to an increase in the convenience yield. The constant is -0.8671.

The third method is named ‘Trend’ and focuses on periods that appear to be trends. Again, this period selection results in a positive and significant MSCI coefficient of 0.0008403. However, unlike the previous two methods, the T-score of 2.95 indicates a significance level of 5% per cent instead of 1%. Just like the substantial method, this selection does not result in a significant coefficient for the Dollar/Euro exchange rate. But, on the 10% significance level, there is some more prove for a substitute effect with the silver market. The constant is -0.3144.

The fourth method, labeled as ‘Mix’ uses the decision rules described in the methodology section to create a mixture between the trend and the substantial rule. As stated in the second hypothesis, it was expected that this way of choosing periods would result in more significant results than by using the simple decision rule. However, the results indicate the opposite. According to the regression, the MSCI coefficient of 0.000873 has a T-score of only 2.50. This is still significant on the 5% level, but it is the least significant result up to now. The Dollar/Euro exchange rate is negative but not significant. The silver price has a negative effect on the gold price with a significance level of 10%, just like the ‘Trend’ and ‘Substantial’ methods indicated. The constant is -0.3632.

(17)

MSCI 0.000000314 0.0002102 0.0005767 0.000615 0.0015976 (0.00) (0.04) (0.11) (0.11) (0.18) DOLLAR/EURO 3.295371 0.1674906 1.895209 0.6559029 -1.734125 EXCHANGE RATE (1.03) (1.07) (0.52) (0.16) (-0.29) SILVER 0.4523276 0.5467168 0.5478453 0.6005652 0.0956105 (1.91)* (1.79)* (2.11)** (2.02)** (0.34) CPI -0.0559257 -0.0772996 -0.066528 -0.0794946 -0.00494 (-0.86) (-0.94) (-0.94) (-1.03) (-0.05) CONSTANT 1.98303 2.800122 4.151058 7.277 0.877092 # OF OBSERVATIONS 270 165 225 180 87 R-SQUARED 0.0928 0.1509 0.1128 0.1354 0.0095

The fifth and last method that is tested is the control group. According to the causality, the control group should not have any anticipation effect and therefore, the MSCI coefficient is expected to be not different then zero. However, if we look at the results, the model also works fine for this specific group. The MSCI coefficient of 0.0012268 with a T-score of 2.32 is significant at the 5% level. Although this coefficient is somewhat lower than in the other groups, this is some important information to keep in mind before drawing any conclusions. The Dollar/Euro exchange rate coefficients well as the silver price coefficient are not significant when using this specific time periods. The constant is -1.097009.

For all 5 regressions, the R squared is really low. This means that the model does not explain a lot of the variance in the convenience yield. However, since the MSCI coefficients are all significant, this is not that much of a problem. The main conclusions can still be drawn from the coefficients, which is the most important objective of this thesis. To check for robustness, it makes sense to do the same regressions again, but now with monthly data. Besides, this allows for using the CPI as a control variable as well, which might change some of the results. The results are displayed in Table III.

TABLE III

THE EFFECT OF GLOBAL STOCK MARKETS ON THE CONVENIENCE YIELD (MONTHLY) DEPENDENT VARIABLE = IMPLIED CONVENIENCE YIELD

SIMPLE SUBST. TREND MIX CONTROL

(18)

The findings in Table III differ from the daily regression results in Table II. The first and most important difference is that the MSCI coefficients are not significant anymore in every method of period selection. With extremely low T-values that are just a little above 0 (or 0), this is in contrast with the findings from the daily regression that brought up highly significant results (at the 5% level at least). Another big contradiction is the fact that the silver coefficients in the montly robustness check (Table III) are significant for almost all methods (except the control group) with a positive coefficient, whereas in Table II, we see that these coefficients are significantly negative (except the control group and the simple method). According to theory, these coefficients should have a negative sign. The positive effect might be explained by the fact that gold and silver, next to substitutes, are both precious metals. So, if a shock takes place and investors turn to precious metals, silver and gold may be affected in a similar, overarching way that undermines the fact that they are substitutes as well. Still, it is remarkable that the two regressions show such contradicting results. Finally, the last take away point of this regression compared to the daily regression can be found in the presence of the CPI coefficient. For every period, this coefficient is negative, however not significant. Since the coefficients are all non-significant, it seems that leaving them out of the daily regression will not cause a lot of bias. Further, it is interesting to look at the negativity of these coefficients since it contradicts theory: An increase in inflation means that money becomes less valuable and thus, investors or consumers might turn to a substitute for money; gold. Therefore, the convenience yield should increase. Concerning the dollar/euro exchange rate coefficient, there are no significant or interesting findings. For the first four monthly regressions, the R-squared is higher compared to their daily equivalents. However, for the control group, the R-squared is lower. The exact cause of this difference is not clear.

In the following paragraph, the findings in Table IV, which can be found on the next page, will be discussed and compared to the findings in Table III. The regression that this table underlies differs from that of Table III in two ways. Namely, the gold/stock ratio is included and the regression ranges only until the end of 2015 due to a lack of gold/stock ratio data. Despite these differences, most results in Table IV and Table III are quite similar and adding the gold/stock ratio does not seem to have much influence on the other coefficients. Again, both the MSCI and the Dollar/Euro exchange rate coefficients are insignificantly different from zero and the CPI coefficients are negative, but insignificant. Also, the R-squares are more or less the same. The silver coefficients are still positive but slightly lower than in Table III, which causes them to be non-significant. Of course, the most interesting part of this regression is the gold/stock ratio. Despite the deeply negative coefficients, they are not significant due to equally large standard deviations. The sign of the coefficients is not in accordance with theory, which may be caused by serial correlation (by the way the yearly data is converted to monthly data, see Literature Review), measurement errors that arose by the way the datasets are constructed or causality that is wrong/not strong enough.

(19)

MSCI -0.0002652 -0.0002575 0.0003602 -0.0003956 0.0011138 (-0.06) (-0.05) (0.07) (-0.08) (0.12) DOLLAR/EURO 4.692345 6.466726 4.296259 3.831016 1.492021 EXCHANGE RATE (1.20) (1.18) (0.89) (0.68) (0.26) SILVER 0.6132767 0.7429888 0.8181677 0.8679233 0.412886 (1.33) (1.34) (1.62) (1.59) (0.92) CPI -0.1001795 -0.131451 -0.1385363 -0.1485521 -0.1144751 (-0.84) (-0.90) (-1.06) (-1.08) (-1.05) GOLD/STOCK -49.62186 -64.44343 -79.2115 -82.50711 -104.4192 (-0.67) (-0.71) (-0.94) (-0.91) (1.64) CONSTANT 7.950262 10.09234 13.44231 15.88994 15.53436 # OF OBSERVATIONS 264 163 222 177 83 R-SQUARED 0.1039 0.1618 0.1340 0.1601 0.0410 TABLE IV

GOLD/STOCK RATIO ADDED

The last paragraphs will look into the different results that might arise by adding dummies for periods with high storage costs and periods with low storage costs. The exact decision rules concerning these dummies can be found in the ‘Literature Review’ section. The results of the regression including these dummies are presented in Table V on the next page. Further, since data on storage costs is available from 1993, it is not appropriate to compare this regression results with one of the previous described tables. Therefore, a separate regression without storage dummies that starts in 1993 is created. The results of this regression can be found in Table VI. Tables V and VI together will provide insights about the importance of the storage costs and how including them will affect the other variables. First, the storage costs coefficients itself will be reviewed. Then, the effect that they appear to have on the other variables will be discussed.

(20)

MSCI -0.0041209 -0.0042132 -0.0029115 -0.002582 -0.0012149 (-0.87) (-0.69) (-0.53) (-0.44) (-0.12) DOLLAR/EURO -5.69899 -11.28226 -4.82615 -6.430551 -32.88871 EXCHANGE RATE (-0.94) (-1.36) (-0.61) (-0.72) (-1.72)* SILVER 0.1643251 0.3678152 0.215959 0.2697306 0.638633 (0.85) (1.55) (1.07) (1.22) (0.95) CPI 0.0231308 -0.0516918 0.0115354 -0.0103826 -0.1125344 (0.34) (-0.58) (0.15) (-0.13) (-0.65) High Storage Costs 5.546373 9.037765 4.693289 4.242617 11.60591

(1.87)* (2.62)*** (1.34) (1.12) (1.37) Low Storage Costs -1.963955 -3.973644 -4.365055 -11.68432 4.030515

(-0.25) (-0.29) (-0.46) (-0.80) (0.49)

CONSTANT 5.562399 23.43459 5.196084 10.4491 53.45617

# OF OBSERVATIONS 164 96 136 105 54

R-SQUARED 0.0490 0.1189 0.0552 0.0850 0.0702

The ‘High Storage Costs’ coefficient changes through the various methods of period identification and will therefore be handled in the same order as these methods appear in the table. In the simple identification method, a coefficient of 5.546373 is found. With a T-score of 1.87, this is significant on the 10% level, indicating that periods with high storage costs have a significantly higher convenience yield. This is in line with the Implied Convenience Formula [2], stating that storage costs have a positive relationship with the convenience yield. For the substantial period selection method, this effect is even bigger. With a coefficient of 9.037765 and a T-score of 2.62 (significant at the 1% level), the high storage costs appear to have a large impact on the convenience yield. In the three up-following methods, the coefficients are still positive, however not longer significantly different from zero. The cause for this difference is not obvious.

Now, the Low Storage Costs will be discussed. In agreement with theory, all coefficients have a negative sign. However, none of them is significantly different from zero. A possible explanation is that the decision rule for this group has been eased (relative to the high storage cost category). This could have resulted in a less strict selection of periods and eventually in less significant results.

TABLE V

DATA FROM 1993, WITH STORAGE DUMMIES

(21)

MSCI -0.004442 -0.0047491 -0.003263 -0.0031317 0.0001446 (-0.92) (-0.77) (-0.62) (-0.55) (0.01) DOLLAR/EURO 1.437862 1.967425 1.708564 -0.5362053 -12.74767 EXCHANGE RATE (0.22) (0.22) (0.22) (-0.06) (-0.83) SILVER 0.1058489 0.2129773 0.1963513 0.2387645 0.2753954 (0.48) (0.78) (0.85) (0.97) (0.44) CPI 0.0397463 0.0004487 0.0111966 -0.0100414 -0.0231647 (0.58) (0.00) (0.14) (-0.13) (-0.14) CONSTANT -4.500304 1.500883 -1.553307 4.593525 17.27472 # OF OBSERVATIONS 164 96 136 105 54 R-SQUARED 0.0177 0.0321 0.0215 0.0192 0.0214 TABLE VI

DATA FROM 1993, NO STORAGE DUMMIES

In order to see the effects that adding storage costs dummies to the regression have on the other variables, Tables V and VI should be compared. Starting with the MSCI coefficients, there is no visible difference between the regressions. The T-scores ranging from -0.87 to -0.12 in Table V are quite similar to the T-scores in Table VI, ranging from -0.92 to 0.01. It is remarkable to see that in both regressions, the coefficients for the control group are the least negative/most positive. Next, the storage costs seem to have some negative influence on the Dollar/Euro exchange rate, whereas the coefficients all become (more) negative, with the coefficient of the control group even being significant at the 10% level (-32.88871, T-score of -1.72). Further, the silver price seems not to be influenced. All coefficients remain positive but not significantly different from zero. The same goes for the CPI. Although the sign in the substantial period selection method changed, this all revolves around zero and is not significant. Notice that the R-squares are higher in Table V.

(22)

5. Conclusion

This thesis investigated the effect of the global stock market on the gold price in the long-term by using the implied convenience yield as dependent variable. The economical mechanism behind this works with investors’ expectations of a crisis, which could influence their investment patterns. In 5 different regression equations; the silver price, the dollar/euro exchange rate, the inflation rate, the gold/stock ratio and two dummies indicating high or low storage costs are interchangeably used as control variables. Next to this, five different methods of period selection have been identified, resulting in a total of 25 unique regressions. In order to answer the research question, two hypotheses have been posed. The first hypothesis checks whether the coefficients of the global stock market differ significantly from zero, whereas the second hypothesis checks the correctness of the causality. In the next paragraph, two main findings will be drawn from the regression results in order to answer these hypotheses. Then, some points of criticism are expressed, which on its turn lead to recommendations for future research. Finally, the research question will be answered.

This paragraph will identify and explain two main findings, which both will be linked to a specific hypothesis. The first main finding regards the coefficients of the MSCI world stock index and is linked to the first hypothesis. As can be seen in Table II, the MSCI coefficients in all four methods have a positive and significant effect on the convenience yield: Simple (4.04)***, Substantial (3.70)***, Trend (2.95)** and Mix (2.50)**. These results give enough reason to reject the null hypothesis and thus confirm some positive effect of the global stock market on the convenience yield. However, it is important to notice the following things: The null hypothesis is not only rejected in the four methods above that fit the explained economical mechanism, but also in the control group (2.32)**. This gives ample reason to question the underlying causality. Furthermore, in Tables III up to and including VI, most of the MSCI coefficients are close to zero or even negative, which increases doubts about the correctness even more.

The second main finding, which is linked to the second hypothesis, focuses on the differences in results that emerge between methods of period selection. If the economical mechanism is right, a more significant (or less insignificant) global stock market effect should be encountered in methods that make use of a decision rule that filters out non-substantial months or weak trends. Therefore, it was expected that the MSCI coefficients of the ‘mix’ method would exceed those of the ‘substantial’ method and the ‘trend’ method, which on their turn would exceed the MSCI coefficients of the simple method. The results in Table II, however, show something completely different. Here, the sequence of significance is completely turned around. The most significant coefficient is found in the ‘Simple’ method (4.04), followed by ‘Substantial’ (3.70) and Trend (2.95). The least significant result is found in the ‘Mix’ method (2.50). Again, for robustness purposes, it is useful to check the remaining tables. In

(23)

Tables III, V and VI, some evidence supporting the hypothesis can be found. However, since these tables do not include any coefficients significant different from zero, it is hard to draw conclusions. Taking all this information together, there is not enough hard evidence to reject the null hypothesis. Therefore, the underlying economical mechanism should be questioned.

The first point of criticism is on the influence of the silver price on the convenience yield. As can be seen in the ‘Results’ section, the silver price coefficients tend to be rather volatile. In the daily regression (Table II), 4 of 5 coefficients have a significant negative effect whereas in the monthly basic regression (Table III), 4 of 5 coefficients have a significant positive effect. Although both signs can have a plausible explanation, this points out that more research is needed. In order to use silver prices as a valid variable, it is necessary to have a better understanding of the causal relationship between gold and silver prices. In what specific situations do silver prices lead the gold price? How big is the distortion created by the fact that both assets are in the ‘precious metal investment class’?

The second critique regards the gold/stock ratio. As described in the ‘Results’ section, this gold/stock ratio does not seem to have any effect. However, this thesis has failed in collecting proper monthly/daily data on both the total amount of gold above ground as well as the amount of money invested in gold. To solve this problem, a yearly dataset has been transformed into a monthly dataset, leading to both autocorrelation and bias. Therefore, it is not appropriate to label the gold/stock ratio as irrelevant. More research needs to be conducted including a correct gold/stock ratio.

Another weak point can be found in the (lack of) storage costs data. Since this research uses the implied convenience yield as dependent variable, having a correct and complete dataset is of major importance. Table V and VI show that storage costs, in some cases, do affect the convenience yield, as well as the coefficients for silver prices. For future research on this topic, it is a must to include proper storage costs data.

Research question to be answered:

Did the MSCI world returns in bull-market periods from January 1979 until December 2016 have a positive effect on the gold price in the long-term via an increase in the convenience yield of gold and therefore can be identified as an indicator variable for the long-term price of gold?

Summarizing the main findings with the criticisms discussed above, there is too much ambiguousness to answer this research question with a simple yes or no. Some results (Table II) show that indeed, in bull-market periods, the global stock market has a positive effect on the convenience yield and thus on the gold price in the long-term. However, other results do not (fully) support or even contradict this effect. As stated before, more research needs to be conducted before being able to validly answer this question.

(24)

6. Bibliography

Akram, Q. F. (2009). Commodity prices, interest rates and the dollar. Energy economics, 31(6), 838-851.

Baur, D. G., & Lucey, B. M. (2010). Is gold a hedge or a safe haven? An analysis of stocks, bonds and gold. Financial Review, 45(2), 217-229.

Baur, D. G., & McDermott, T. K. (2010). Is gold a safe haven? International evidence. Journal of Banking & Finance, 34(8), 1886-1898.

Capie, F., Mills, T. C., & Wood, G. (2005). Gold as a hedge against the dollar. Journal of International Financial Markets, Institutions and Money, 15(4), 343-352.

Casassus, J., & Collin-Dufresne, P. (2005). Stochastic convenience yield implied from commodity futures and interest rates. The Journal of Finance, 60(5), 2283-2331.

Demidova-Menzel, N., & Heidorn, T. (2007). Gold in the investment portfolio. Frankfurt School of Finance & Management.

Fabozzi, F., & Francis, J. (1977). Stability Tests for Alphas and Betas Over Bull and Bear Market Conditions. The Journal of Finance, 32(4), 1093-1099.

Fama, E., & French, K. (1987). Commodity Futures Prices: Some Evidence on Forecast Power, Premiums, and the Theory of Storage. The Journal of Business, 60(1), 55-73

Gibson, R., & Schwartz, E. S. (1990). Stochastic convenience yield and the pricing of oil contingent claims. The Journal of Finance, 45(3), 959-976.

Gorton, G., & Rouwenhorst, K. G. (2006). Facts and fantasies about commodity futures. Financial Analysts Journal, 62(2), 47-68.

Lee, C. F., & Lee, A. C. (Eds.). (2006). Encyclopedia of finance (Vol. 855). New York: Springer. Narayan, P. K., Narayan, S., & Zheng, X. (2010). Gold and oil futures markets: Are markets

efficient? Applied energy, 87(10), 3299-3303.

Reboredo, J. C. (2013). Is gold a safe haven or a hedge for the US dollar? Implications for risk management. Journal of Banking & Finance, 37(8), 2665-2676.

Sari, R., Hammoudeh, S., & Soytas, U. (2010). Dynamics of oil price, precious metal prices, and exchange rate. Energy Economics, 32(2), 351-362.

(25)

Siegel, J. J. (2003). What is an asset price bubble? An operational definition. European financial management, 9(1), 11-24.

Sjaastad, L. A., & Scacciavillani, F. (1996). The price of gold and the exchange rate. Journal of international Money and Finance, 15(6), 879-897.

Tully, E., & Lucey, B. M. (2007). A power GARCH examination of the gold market. Research in International Business and Finance, 21(2), 316-325.

Worthington, A. C., & Pahlavani, M. (2007). Gold investment as an inflationary hedge: Cointegration evidence with allowance for endogenous structural breaks. Applied Financial Economics Letters, 3(4), 259-262.

Data World Gold Council. Retrieved from: http://www.gold.org/gold-mining/what-gold-mining-companies-produce/how-much-gold-has-been-mined

Data World Bank. Retrieved from: http://data.worldbank.org/indicator/CM.MKT.LCAP.CD

(26)

7. Appendix

With daily data (Table II) Simple Substantial Trend _cons -.174267 .3889653 -0.45 0.654 -.9367823 .5882483 SILVER -.0105668 .0167193 -0.63 0.527 -.0433429 .0222092 DollarEuro -.7977806 .3457788 -2.31 0.021 -1.475634 -.1199267 MSCI .0008094 .0002005 4.04 0.000 .0004164 .0012024 ConvYield Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 4.3588 R-squared = 0.0085 Prob > F = 0.0000 F( 3, 5865) = 9.64

Linear regression Number of obs = 5869

_cons -.867132 .5049165 -1.72 0.086 -1.857266 .1230017 SILVER -.0683979 .0299611 -2.28 0.023 -.1271511 -.0096447 DollarEuro .1674906 .409674 0.41 0.683 -.635874 .9708551 MSCI .0010941 .0002958 3.70 0.000 .0005141 .0016742 ConvYield Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 4.2216 R-squared = 0.0201 Prob > F = 0.0000 F( 3, 2325) = 10.72

. log close _cons -.3144028 .5599506 -0.56 0.575 -1.412332 .7835269 SILVER -.0543592 .0278581 -1.95 0.051 -.1089823 .0002639 DollarEuro -.3203242 .4750435 -0.67 0.500 -1.251771 .6111229 MSCI .0008403 .000285 2.95 0.003 .0002815 .0013991 ConvYield Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 4.4994 R-squared = 0.0117 Prob > F = 0.0001 F( 3, 2975) = 7.51

(27)

Mix

Control

With monthly data (Table III) Simple _cons -.3631723 .6074873 -0.60 0.550 -1.554424 .8280791 SILVER -.0584363 .0310102 -1.88 0.060 -.1192457 .0023731 DollarEuro -.2768189 .5921989 -0.47 0.640 -1.43809 .8844527 MSCI .000873 .0003489 2.50 0.012 .0001887 .0015572 ConvYield Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 4.6579 R-squared = 0.0125 Prob > F = 0.0002 F( 3, 2411) = 6.69

_cons -1.087009 .8516032 -1.28 0.202 -2.757988 .5839698 SILVER .0029085 .0561984 0.05 0.959 -.1073615 .1131786 DollarEuro -.1440222 .4797711 -0.30 0.764 -1.085408 .797364 MSCI .0012268 .0005297 2.32 0.021 .0001874 .0022662 ConvYield Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 3.8208 R-squared = 0.0118 Prob > F = 0.0765 F( 3, 1083) = 2.29

. log close _cons 1.98303 5.706918 0.35 0.729 -9.253643 13.2197 SILVERH .4523276 .2365399 1.91 0.057 -.0134091 .9180644 USEURSP 3.295371 3.199127 1.03 0.304 -3.003571 9.594312 USCONPRCE -.0559257 .0651135 -0.86 0.391 -.1841313 .07228 MSWRLD 3.14e-07 .0046216 0.00 1.000 -.0090993 .0091 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 11.579 R-squared = 0.0928 Prob > F = 0.2102 F( 4, 265) = 1.47

(28)

Substantial Trend Mix _cons 2.800122 7.749467 0.36 0.718 -12.50431 18.10456 SILVERH .5467168 .3059302 1.79 0.076 -.0574652 1.150899 USEURSP 4.474549 4.18388 1.07 0.286 -3.788202 12.7373 USCONPRCE -.0772996 .0822522 -0.94 0.349 -.2397397 .0851404 MSWRLD .0002102 .0057301 0.04 0.971 -.0111061 .0115266 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 12.204 R-squared = 0.1509 Prob > F = 0.1332 F( 4, 160) = 1.79

. log close _cons 4.151058 6.598649 0.63 0.530 -8.853597 17.15571 SILVERH .5478453 .2597199 2.11 0.036 .035988 1.059703 USEURSP 1.895209 3.636556 0.52 0.603 -5.271736 9.062155 USCONPRCE -.066528 .0704874 -0.94 0.346 -.205445 .0723891 MSWRLD .0005767 .0051715 0.11 0.911 -.0096152 .0107687 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 11.988 R-squared = 0.1128 Prob > F = 0.2400 F( 4, 220) = 1.39

. log close _cons 7.277 7.576648 0.96 0.338 -7.676367 22.23037 SILVERH .6005652 .2967472 2.02 0.045 .0149013 1.186229 USEURSP .6559029 4.050138 0.16 0.872 -7.337501 8.649307 USCONPRCE -.0794946 .0771121 -1.03 0.304 -.2316841 .0726948 MSWRLD .000615 .0056705 0.11 0.914 -.0105764 .0118063 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 12.301 R-squared = 0.1354 Prob > F = 0.2326 F( 4, 175) = 1.41

(29)

Control

With gold/stock ratio (Table IV) Simple Substantial _cons .877092 10.43528 0.08 0.933 -19.882 21.63619 SILVERH .0956105 .2845971 0.34 0.738 -.4705437 .6617648 USEURSP -1.734125 5.97941 -0.29 0.773 -13.62908 10.16083 USCONPRCE -.00494 .0917911 -0.05 0.957 -.1875419 .1776618 MSWRLD .0015976 .0088477 0.18 0.857 -.0160033 .0191985 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 11.973 R-squared = 0.0095 Prob > F = 0.9476 F( 4, 82) = 0.18

_cons 7.950262 12.73295 0.62 0.533 -17.12348 33.024 Goldstock -49.62186 73.57282 -0.67 0.501 -194.5016 95.25785 SILVERH .6132767 .4594689 1.33 0.183 -.2915101 1.518063 USEURSP 4.692345 3.901791 1.20 0.230 -2.991066 12.37576 USCONPRCE -.1001795 .1192636 -0.84 0.402 -.3350335 .1346745 MSWRLD -.0002652 .004602 -0.06 0.954 -.0093274 .008797 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 11.521 R-squared = 0.1039 Prob > F = 0.0231 F( 5, 258) = 2.66

_cons 10.09234 15.43941 0.65 0.514 -20.40341 40.5881 Goldstock -64.44343 90.19455 -0.71 0.476 -242.5947 113.7079 SILVERH .7429888 .5562138 1.34 0.184 -.3556386 1.841616 USEURSP 6.466726 5.465701 1.18 0.239 -4.329067 17.26252 USCONPRCE -.131451 .146432 -0.90 0.371 -.4206819 .1577799 MSWRLD -.0002575 .0053121 -0.05 0.961 -.0107499 .010235 CY Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = 12.17 R-squared = 0.1618 Prob > F = 0.0182 F( 5, 157) = 2.82

Does the global stock market influence the price of gold In the long run?