Relevance of the state of the economy in the impact of macroeconomic data events on precious metal prices

(1)

RELEVANCE OF THE STATE OF THE

ECONOMY IN THE IMPACT OF

MACROECONOMIC DATA EVENTS ON

PRECIOUS METAL PRICES

Sameer Alsaadi

Supervisor: Daniel Dimitrov June 2018

Abstract

The study utilizes intraday data over the period 2008 till 2017 to investigate the impact of macroeconomic data announcements on the precious metals gold and silver. The scope of the research is to find evidence for whether the state of the economy has any relevance in this relationship. The paper succeeds in finding evidence for sensitivity of silver returns for the revelation of consumer confidence and nonfarm payroll data. Remarkably, exactly these data events were tested significant for the variable that denoted the state of the economy. Therefore the study concludes that the state of the economy is indeed relevant for the impact of the previously mentioned events on at least silver prices. The remainder of the macroeconomic data events provided no evidence for affecting gold and silver returns such that the findings of the study are not yet generalizable for all precious metals and macroeconomic data events. However, the results are notable and incentivize future research to elaborate on the study’s findings.

(2)

1

Statement of Originality

This document is written by Sameer Alsaadi who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

2

1. Introduction

Roache & Rossi (2010) state that commodity prices were subject to high volatility during the extensive period of financial turmoil after the global financial crisis of 2008. The pro-cyclical nature of demand drove the price movements of some commodities while commodities like precious metals that are seen as safe haven assets and store of value emphasized their roles during this period (Roache & Rossi, 2010). Precious metal markets became dominated by financial investors looking to increase diversification and returns in their portfolios which resulted in a major price increase of over 100% for gold and silver in the following years. The elevated financial and economic volatility together with the financialization of precious metals have sparked a debate about the prime movers of precious metal prices and the extent to which they may have changed (Mayer, 2008). Traditional precious metal price

determination merely relies on fundamentals (Borensztein & Reinhart, 1994) while recent research suggests that as precious metal markets get more financialized the influence of systematic shocks progressively increases (Hammoudeh, Yuan, McAleer, & Thompson, 2010). The increasing dependence of precious metals to the overall market combined with the findings of Elder, Miao, & Ramchander (2012), which imply that macroeconomic data

inbounds are of vital importance to the price forming and future price expectations of precious metals, incentivized researching whether the impact of macroeconomic data announcements on precious metals is greater in times of economic contraction.

The results of this research are relevant as previous literature that uses high-frequency intraday data only focuses on the significance of the impact of macroeconomic events. This paper builds on the present literature by studying whether precious metal price movements as a result of macroeconomic data announcements tend to show bias from economic cycles. This is in line with Elder, Miao, & Ramchander (2012) who suggested that the relationship

between economic asymmetries and the impact of macroeconomic announcements needs further research. Besides scientific relevance, the results can be used more practically as well by investors that hedge their portfolios with precious metals or traders who want to profit from the price jumps that occur from macroeconomic data announcements.

In conclusion, the study does suggest a bias in silver price forming during a contractionary economy for consumer confidence and nonfarm payroll data releases. However, for gold and silver, the remainder of the analyzed events, no interpretable results are obtained as the macroeconomic events did not provide significant evidence for being effective. It is remarkable that the events that proved to have a relationship to the precious metal price changes also tested significance for the state of the economy. The latter urges further research to be conducted such that the findings of this study could be generalized overall precious metals and types of macroeconomic data announcements.

The study will resume as follows. Section 2 reviews the literature and is followed by the methodology in section 3. Section 4 summarizes the data characteristics and presents the results, while section 5 discusses the limitations and concludes.

(4)

3

2. Literature review

An analysis of the literature in the precious metal markets unveils four dominating types of research: (i.) investigation of relationships between future and spot prices (Kocagil, 1997); (ii.) studies of precious metals as safe-haven assets for hedging purposes (Reboredo, 2013); (iii.) research of statistical characteristics of different metals (Arouri, Hammoudeh, Lahiani, & Nguyen, 2012); and (iv.) examination of relationships between the precious metals themselves or other assets (Escribano & Granger, 1998). As endorsed by Elder, Miao, & Ramchander (2012) the field of analysis of the effect on precious metal markets that arises from macroeconomic news events is relatively underresearched and therefore additional studies weigh in this new dimension of precious metal market research.

Even though preceding literature broadly focuses on the effect of macroeconomic events on exchange rates (Almeida, Goodhart, & Payne, 1998; Ehrmann & Fratzscher, 2005), stocks (Rangel, 2011) and bonds (Andritzky, Bannister, & Tamirisa, 2007), papers about the effect of macroeconomic news events on precious metal markets are limited. Part of this limited literature uses daily price data under the main assumption that direct effects from

macroeconomic news announcements are unbiased white noise (Roache & Rossi, 2010; Hess, Huang, & Niessen, 2008) which is proven to be a rather harsh assumption by intraday

literature as the use of intraday data generally provides more evidence of responsiveness from precious metal prices to macroeconomic news. The latter is confirmed as the findings of Roache & Rossi (2010) suggest that the daily price changes of precious metals remain unaffected by macroeconomic data announcements, while Hess, Huang, & Niessen (2008) find no significant daily price change responses from precious metals during economic expansions. Hess, Huang, & Niessen (2008) do find that daily precious metal price changes are only sensitive in a contractionary state of the economy which is especially relevant to consider for this research.

Thus far three studies using intraday data have been published that are devoted to the effect of macroeconomic news announcements on precious metal markets. Christie-David,

Chaundry, & Koch (2000) conclude in their 15-minute interval research over the period from 1992 till 1995 that both metals are responsive to surprise data regarding inflation and

unemployment but in general are less responsive than interest rate futures. Cai, Chueng, & Wong (2001) use 5-minute intervals for gold futures over the period from 1994 till 1997 and find responsiveness to unemployment, inflation and personal income data while the analyzed gold futures seem less sensitive to this data than the bond and foreign exchange markets. Finally, Elder, Miao, & Ramchander (2012) discover in their analyses, in which future prices are sampled at 1-minute discrete intervals, that metal futures respond swiftly and significantly to surprise data over the period from 2002 till 2008. Notably, nonfarm payrolls and durable goods resulted in precious metal responsiveness.

This thesis distinguishes itself from the aforementioned researches through various aspects. First of all, the study is concerned with intraday data instead of daily data as previous literature shows evidence for intraday reactions of precious metals on economic data

surprises. Furthermore, the timeframe of the dataset is extended over a longer period than any other study, namely 10 years, and the data comprises the most recent developments in the precious metal markets as the time frame is extended till 2017. The long timeframe ensures

(5)

4 that a large sample is used for the study and different economic cycles are covered within the dataset. The first part of the dataset of gold and silver shows an uptrend-cycle until

September 2011 which is followed by a downtrend-cycle until November 2015 and a stable period until the end of the time series. A final and critical point of distinction is the different scope of this study as it does not only search for relationships between precious metal prices and macroeconomic news events as there are enough studies supporting the connection but it does also research whether the state of the economy influences this relationship.

3. Methodology

Data

Price data for the precious metals gold and silver from the 1st_{of January, 2008 till the 31}st_of

December, 2017 is analyzed. For the determination of the time period of the research the quarterly GDP per capita growth is examined. Many negative growth observations were registered from 2008 (Figure 1,2 and 3) due to the Global Financial Crisis which are relevant for the study to include in the dataset. Furthermore, previous literature does not include contemporary data which incentivized the extension of the dataset to the most recent year available.

Ideally, future prices of gold and silver would be studied rather than spot prices as some literature suggests a positive correlation between future and spot pricing with a 1-day lag which would imply a delayed effect from the macroeconomic news announcements (Roache & Rossi, 2010). This is contradicted by Silvapulle & Moosa (1999) who found a bidirectional effect by nonlinear causality testing and therefore conclude that spot and futures markets react simultaneously to new information. Although, future prices are generally used in academia as previous literature states that future markets lead the developments in spot markets (e.g., Yang, Balyeat, & Leatham, 2005; Antoniou & Foster, 1992). Due to a lack of intraday data for futures this paper will focus on spot prices.

To decide over which time interval price changes as a result of macroeconomic data announcements should be measured several research methods of previous papers are compared. A 10-minute interval with price changes measured from 5 minutes prior to the event until 5 minutes after, a 5 minute-interval starting from the announcement and finally a 1-minute interval starting from the announcement are all three tested to find the sensitivity of gold and silver prices to the core CPI announcement. The test regressions did not provide statistical evidence for a significant relationship between core CPI data announcements and the precious metal prices for any of the three time intervals. Despite these results the 5-minute time interval starting from the announcement is used for this study as in the research of Cai, Chueng, & Wong (2001) this interval proved to bring the desired outcome in news event significance.

The selection of different types of news announcements is based on past literature on the impact of news on precious metal prices (Christie-David, Chaundry, & Koch, 2000; Cai, Chueng, & Wong, 2001) and other types of assets (Ehrmann & Fratzscher, 2005). Unemployment and inflation data announcements provided enough evidence to cause fluctuations in gold and silver prices such that these events cannot be omitted. Other events

(6)

5 such as ECB and BoE interest rate decisions came out as influential as well from the

aforementioned researches and are therefore included in this study.

Regression Model

As stated by Roache & Rossi (2010) financial assets, including gold and silver, incorporate expectations in their pricing. Therefore the surprise component in the news should be targeted when the impact of news announcements is regarded (Roache & Rossi, 2010). To measure the surprise factor the actual outcome of a data announcement is deducted by the expected outcome. Finally, this is normalized by its standard deviation to make a meaningful

comparison of the estimated news impact across gold, silver and the different news releases (Elder, Miao, & Ramchander, 2012). The standardization of the surprise elements looks as follows:

𝑍_𝑖,𝑡 =𝑋𝑖,𝑡− 𝐸(𝑋𝑖,𝑡) 𝜎_𝑖

Let 𝑋_𝑖,𝑡 denote the actual outcome of an announcement of type 𝑖 at time 𝑡 and the consensus estimate 𝐸(𝑋𝑖,𝑡) standardized by the standard deviation σ of the announcement 𝑖.

As discussed by Reboredo (2013) there is an indirect effect from the precious metals’ role as a hedge against U.S. dollar depreciation and lower interest rates. This effect might be exerted by the impact of macroeconomic news (Roache & Rossi, 2010). To address the latter the U.S. dollar index changes over the same 5-minute interval will be added to the regression model as an exogenous variable denoted by Δ$ such that purely gold and silver price changes are measured. It is assumed that all causality runs from the U.S. dollar which is an

uncontroversial assumption according to Roache & Rossi (2010) as commodity prices, including gold and silver, are influenced by exchange rates.

Finally, to address the research objective a variable that reflects the state of the economy is added to the model. As stated by Burns & Mitchell (1946) short-run economic activity is measured by discrete regimes, namely, contractions and expansions. Inspired by the research of Blomberg, Hessa, & Weerapana (2004) in which the relationship between terrorism and the state of the economy is tested, the former two economic states are defined as negative and non-negative growth of real GDP per capita respectively. The latter will be processed through a dummy variable denoted by 𝐶𝑡 that is 0 when the real GDP growth per capita is positive

such that there is no economic contraction and 1 when the real GDP growth per capita is negative to represent a contracting economy.

Combining the previously described variables results in the following multilinear regression model with the change in spot prices from gold and silver as the dependent variable ∆𝑃_𝑡:

∆𝑃_𝑡= 𝛽₀+ 𝛽_𝑖𝑍_𝑖,𝑡+ 𝛾𝐶_𝑡+ 𝜃∆$_𝑡+ 𝜀_𝑡

Depending on the significance of the macroeconomic event parameter 𝛽𝑖 the parameter 𝛾 will

be tested for significance through a Student’s t-test. The results would imply whether there is a structural additional change in the prices of gold and silver in times of a contractionary economy.

(7)

6

Hypothesis

The expectation of the outcome of this research is driven by the previously discussed

statements by Roache and Rossi (2010). In their conclusion, Roache and Rossi (2010) discuss asymmetries in how commodity prices respond to news, especially for gold and silver as these act as safe-haven assets for negative economic news. This in combination with their findings for a procyclical (but biased) nature formed the hypothesis that in periods of economic contraction macroeconomic news announcements have a greater impact on precious metal prices than in times of economic expansion. Formulated in a testable hypothesis this gives:

𝐻0: 𝛾 = 0

𝐻₁: 𝛾 ≠ 0

Where the parameter 𝛾 stands for the relationship between whether there is an economic contraction or not and the precious metal 5-minute returns after a macroeconomic event.

4. Data characteristics and results

Data characteristics

The data on asset prices consists of intraday, 5-by-5 min, spot market prices for the period January 2008 through December 2017. The data is obtained from the Swiss broker

Dukascopy and comprises a dataset of 1,048,572 observations after calculating the 5-min

returns. The relevant observations are manually linked to the events which resulted in the final sample of this study (Table 1). Gold and silver pricing could directly be downloaded, while the U.S. dollar index is derived from a basket of foreign currency pairs. The following formula is used to reconstruct the U.S. dollar index:

𝑈𝑆𝐷𝑋 = 50.14348112 ∗ (𝐸𝑈𝑅 𝑈𝑆𝐷) −0.576 ∗ (𝑈𝑆𝐷 𝐽𝑃𝑌) 0.136 ∗ (𝐺𝐵𝑃 𝑈𝑆𝐷) −0.119 ∗ (𝑈𝑆𝐷 𝐶𝐴𝐷) 0.091 ∗ (𝑈𝑆𝐷 𝑆𝐸𝐾) 0.042 ∗ (𝑈𝑆𝐷 𝐶𝐻𝐹) 0.036

Spot prices are traded by independent dealers with various types and contract sizes of precious metals. Spot markets are unregulated and dealers set their own prices and policies. In some cases, these prices may not reflect current global spot prices, although it seems like Dukascopy is a reliable source. Another characteristic of spot data is that trading hours are limited to the operation hours of the dealer. As a result, on top of the regular holidays and the daily hourly break from 23:00-00:00 the following dates were excluded from the dataset: 15-02-2009, 07-04-2013, 03-04-2015, 02-01-2017, 05-03-2017 and 14-04-2017.

Table 1 presents the summary statistics for the 5-minute return sample of Gold, Silver and the U.S. dollar index. Only the returns after the occurrence of an event are relevant for this research and therefore merely these returns are taken into account in the table. Gold has a 5-minute mean return of 0.007% and a standard deviation of 0.173% in the sample while silver and the U.S. dollar index have a mean return of 0.039% and 0.004% and a standard deviation of 0.454% and 0.072% respectively. The distribution of the returns for every asset shows a positive skewness and an excessive kurtosis.

(8)

7

Table 1. Total summary statistics of 5-min returns after analyzed macro events for the period 2008-2017.

Gold Silver USDX

Observations 876 876 876

Obs. in expanding economy 674 674 674

Obs. in contracting economy 202 202 202

Mean 0.007% 0.039% 0.004% Standard Deviation 0.173% 0.454% 0.072% Minimum -0.96% -2.07% -0.32% Maximum 1.20% 3.35% 3.35% Skewness 0.764 1.990 0.376 Kurtosis 11.361 17.801 8.415

Other than the total number of observations the number of observations for the different states of the economy are specified in the table as well. Although the statistics are derived from the total observations.

To discern between an expansionary and contractionary state of the economy a boxplot is composed of the aforementioned data. Figure 1 and 2 present the spread of gold, silver and the U.S. dollar index in the expansionary and contractionary state respectively. The only significant difference that can be observed with the naked eye is the larger amount of outliers in the expansionary state. This can simply be explained by the larger amount of observations compared to the contractionary state.

Figure 1. Boxplot of the studied assets for the observations in an expansionary state of the economy. 0

-2% -1.5% -1% -0.5% 0.5% 1% 1.5% 2% 2.5% 3% 3.5%

ΔP(GLD) ΔP(SLVR)

(9)

8

Figure 2. Boxplot of the studied assets for the observations in a contractionary state of the economy.

The data on macroeconomic news announcements comprises 8 different events. For every announcement, both the actual outcome and the consensus are obtained from Econoday. The events are released either quarterly or monthly on a pre-planned schedule and are published through newswires and other data providers (Elder, Miao, & Ramchander, 2012). In table 2 all the researched events are presented together with the corresponding standard deviations of the differences between the actual values and the consensus.

Table 2. List of macroeconomic news announcements over the period 2008-2017.

Macroeconomic event (Abbreviation) Total Std. dev. surprise factor

ADP Change in Nonfarm Payrolls (ADP NFP) 118 54.3718

BoE Interest Rate Decision (BoE) 109 0.12%

CB Consumer Confidence (CC) 119 5.4667

Core Consumer Price Index (CPI) 118 0.09%

ECB Interest Rate Decision (ECB) 101 0.07%

Fed Interest Rate Decision (Fed) 75 0.07%

Nonfarm Payrolls (NFP) 117 63.2512

Producer Price Index (PPI) 119 0.38%

The standard deviations are denoted in their reporting unity. This results in standard deviations designated in absolute values while others are designated in relative values.

0

-2% -1.5% -1% -0.5% 0.5% 1% 1.5% 2% 2.5% 3% 3.5%

ΔP(GLD) ΔP(SLVR)

(10)

9 Table 3 reports the mean percentage returns and the standard deviations for each event

separately. The means of the expansionary and contractionary economy are tested against each other through the Welch’s t-test. Nonfarm payrolls show a significant difference in the mean return for gold and silver in times of economic contraction while the consumer

confidence event strongly confirms a significant difference for the mean return of silver. The ECB and Fed interest rate decision events provide statistical evidence for a deviation of both gold and U.S. dollar index mean returns in a contractionary state. Therefore additional attention will be given to these events in the rest of the study.

Table 3. Statistics and Welch’s t-test results for each event separately over the period 2008-2017.

Gold ADP NFP BoE CC CPI ECB Fed NFP PPI

Total Mean 0.009% 0.021% -0.007% 0.005% 0.016% 0.023% -0.001% -0.005% Std. dev. 0.140% 0.123% 0.133% 0.226% 0.142% 0.237% 0.255% 0.119% Exp. Mean 0.009% 0.016% -0.004% 0.005% 0.001% 0.030% -0.013% -0.006% Std. dev. 0.139% 0.089% 0.105% 0.183% 0.107% 0.212% 0.237% 0.089% Con. Mean 0.011% 0.026% -0.022% 0.007% 0.031%b _-0.020%a _0.062%c _-0.004% Std. dev. 0.146% 0.153% 0.241% 0.395% 0.171% 0.360% 0.218% 0.228%

Silver ADP NFP BoE CC CPI ECB Fed NFP PPI

Total Mean -0.012% 0.049% -0.005% 0.003% 0.016% 0.081% 0.169% 0.023% Std. dev. 0.309% 0.238% 0.272% 0.381% 0.266% 0.618% 0.847% 0.363% Exp. Mean -0.012% 0.052% 0.015% 0.005% 0.033% 0.089% 0.192% 0.034% Std. dev. 0.310% 0.223% 0.228% 0.259% 0.185% 0.642% 0.873% 0.333% Con. Mean -0.012% 0.046% -0.113%c _-0.005% _-0.002% _0.035% _0.044%b _-0.041%b Std. dev. 0.311% 0.257% 0.439% 0.782% 0.330% 0.476% 0.693% 0.510%

USDX ADP NFP BoE CC CPI ECB Fed NFP PPI

Total Mean 0.003% -0.001% 0.009% 0.003% -0.003% -0.007% 0.015% 0.006% Std. dev. 0.057% 0.042% 0.055% 0.062% 0.071% 0.104% 0.113% 0.049% Exp. Mean 0.004% -0.001% 0.008% 0.005% 0.002% -0.001% 0.017% 0.008% Std. dev. 0.055% 0.036% 0.047% 0.060% 0.068% 0.100% 0.118% 0.048% Con. Mean -0.004% -0.002% 0.017% -0.004% -0.009%a _-0.042%c _0.005% _-0.006%c Std. dev. 0.067% 0.048% 0.090% 0.076% 0.074% 0.125% 0.086% 0.052%

The statistics are multiplied by 100 to represent percentages. Exp. and Con. represent an expansionary and contractionary economy respectively. The Welch’s t-test validates whether the mean return in the

contractionary economy is significantly different from the mean return in the expansionary economy.

Furthermore, the Welch’s test statistics can be found in Appendix A and the statistically significant results are marked in bold.

'a' implies significance at the 10% level. 'b' implies significance at the 5% level. 'c' implies significance at the 1% level.

(11)

10 Finally, the data for the state of the economy was obtained from the Oxford Economics

database. The quarterly GDP per capita of the United States, the United Kingdom and the Eurozone over the period 2008-2017 are used. Furthermore, the growth in GDP per capita is calculated to determine whether the economy during the period was in a contractionary or expansionary phase.

In figure 3 to 5, the GDP per capita growth for all three economies is displayed graphically. It can be observed that all economies were in a contractionary state from end-2008 till mid-2009. This can be linked to the Global Financial Crisis in 2008. Also, both the United Kingdom and the Eurozone economies were in a contractionary phase more regularly than the United States in the following years. This is most likely caused by events such as the European Sovereign Debt crisis in 2010 and Brexit in 2016.

Figure 3. United States GDP per capita growth in 2008-2017.

Figure 4.United Kingdom GDP per capita growth in 2008-2017. -3 -2 -1 0 1 2

Oct 06 Feb 08 Jul 09 Nov 10 Apr 12 Aug 13 Dec 14 May 16 Sep 17 Feb 19

United States GDP: PER CAPITA (%QOQ)

-20 -15 -10 -5 0 5 10

United Kingdom GDP: PER CAPITA (%QOQ)

(12)

11

Figure 5. Eurozone GDP per capita growth in 2008-2017.

Results

The standard errors of the ordinary least squares regression are tested for autocorrelation through the Durbin-Watson test. The Durbin-Watson statistics are presented in table 4. None of the results provide significant evidence for autocorrelation and therefore it will be assumed that all the performed linear regressions are free of autocorrelation.

Table 4. Durbin-Watson test statistics to test for autocorrelation.

ADP NFP BoE CC CPI ECB Fed NFP PPI

Gold 2.06 1.94 2.14 1.87 2.24 2.34 2.17 2.18

Silver 2.00 2.25 2.15 2.13 2.09 1.92 1.94 2.13

The values represent Durbin-Watson test statistics from the ordinary least squares residuals of the model.

Subsequently, the residuals are tested for heteroscedasticity through the Breusch-Pagan test for which the outcomes can be found in table 5. The test suggests heteroscedasticity for the majority of the regressions. Hence, for the results that comply with significance at a 10% level the Huber-White’s robust standard errors approach is used in the linear regression.

Table 5. Results of the Breusch-Pagan test for heteroscedasticity.

Gold 0.000c _0.000c _0.000c _0.000c _0.000c _0.035b _0.004c _0.016b

Silver 0.254 0.000c _0.081a _0.078a _0.075a _0.999 _0.000c _0.683

The presented outcomes are p-values of the chi-square distribution. The Breusch-Pagan test is performed over the ordinary least square residuals of the model. The statistically significant results are marked in bold. 'a' implies significance at the 10% level.

'b' implies significance at the 5% level. 'c' implies significance at the 1% level.

-15 -10 -5 0 5 10

Eurozone GDP: PER CAPITA (%QOQ)

(13)

12 The estimates for the sensitivities of gold and silver prices to the U.S dollar index 5 minutes post the release of the macroeconomic news announcements are presented in table 6. As stated by Capie, Mills, & Wood (2005) the U.S. dollar serves as a strong hedge to gold and as discussed in the research of Escribano & Granger (1998) gold and silver are generally correlated in the long term. The results are in line with the prior findings and it seems successful to control for the U.S. dollar effect in the model. For almost every announcement the effect is significant and therefore the effect from the U.S. dollar is canceled out from the effect of other regression variables. These empirics prove that the assumption about the U.S. dollar index is indeed uncontroversial.

Table 6. Gold and silver sensitivities to the U.S. Dollar index after macroeconomic news announcements.

Gold -1.097c _-0.902b _-1.405c _-1.752c _-0.888c _-1.098c _-1.207c _-1.202c

Silver -1.626c _-1.247 _-2.658c _-2.819c _-1.219c _-2.229c _-2.090c _-1.540b

The coefficients represent absolute values for the relationship between the precious metals and the U.S. dollar index. Furthermore, the raw regression outputs can be found in Appendix B and the statistically significant results are marked in bold.

Furthermore, the sensitivities to the macroeconomic events themselves are presented in table 7. Just as in previous literature (e.g., Christie-David, Chaundry, & Koch, 2000; Elder, Miao, & Ramchander, 2012) not all the news events show a significant short-term effect on gold and silver pricing. Not every event is considered to be significantly important such that a surprise shock causes an immediate price movement. Possibly could research that uses other time intervals find more responsiveness from the macroeconomic news announcements stated in Table 7. As macroeconomic data sensitivity is crucial for the interpretation of the results the focus for the rest of this research will mostly lie on the events that are tested to show significance. Consumer confidence, ECB interest rate decision and nonfarm payroll events provide significance by sensitivities of -0.052%, -0.105% and 0.215% respectively.

Table 7. Gold and silver sensitivities to the macroeconomic announcements.

ADP NFP BoE CC CPI ECB Fed NFP PPI

Gold 0.001% -0.014% 0.001% -0.017% -0.047% -0.044% 0.037% -0.006% Silver -0.031% -0.026% -0.052%b _-0.001% _-0.105%c _-0.065% _0.215%c _-0.035%

The coefficients are multiplied by 100 and represent the percentage change in the precious metal for one unexpected standardized deviation. Furthermore, the raw regression outputs can be found in appendix B and the statistical significant results are marked bold.

(14)

13 The gold and silver sensitivities to the state of the economy are displayed in table 8. In line with the hypothesis the coefficients for economic contraction from the regressions in which the effect of the news announcement seemed to have a significant effect turn out to be statistically significant as well. Only for the central bank interest rate decisions for which there is evidence for an effect on the precious metal prices the economic contraction variable did not seem to explain the price changes in gold and silver. This might be due to central banks only changing the interest rate unexpectedly in urgent situations such as crises. No non-zero surprise factors were found in the central bank interest rate decision data during expansionary phases and therefore it is assumed that this bias causes the disparity in the results.

The results comply to economical feasibility as a -0.334% structural additional price change in the price of silver in a Nonfarm Payroll announcement or an additional price change of -0.113% in the price of silver after Consumer Confidence announcements during a

contractionary economy are convincing.

Table 8. Gold and silver sensitivity to the state of the economy

Gold -0.007% 0.050% -0.004% -0.003% 0.014% -0.135%a _0.116% _-0.016%

Silver -0.014% -0.016% -0.113%a _{-0.033% -0.064% -0.203%} _-0.334%a _-0.098%

The coefficients are multiplied by 100 and represent the structural additional percentage change that can be expected in times of economic contraction. Furthermore, the raw regression outputs can be found in Appendix B and the statistically significant results are marked in bold.

The results indicate that the state of the economy indeed does have an effect on the price changes of silver for at least the consumer confidence and nonfarm payroll data

announcements. For the latter two events the null-hypothesis is therefore rejected and the previously stated proposition that the procyclical bias and the larger influence of negative data imply a greater impact of macroeconomic events holds. The remainder of the

macroeconomic events did not provide evidence to affect gold and silver prices. Further research in which the remaining events prove significance has to be conducted to confirm the findings of this study for gold and the other events for silver.

(15)

14

5. Discussion and conclusion

Limitations

The research is restricted through the dataset as intraday precious metal pricing is not widely available for free. Instead of future prices that are set by exchanges less regulated spot prices are used. Another consequence of the intraday data availability is the omission of the metal copper from the study. Copper has a different role in financial markets than gold and silver which have common characteristics. It could have been relevant to compare the results of metals with different traits.

An additional potential limitation is the dominance of observations in an expansionary state of the economy (Table 1) in the sample. This is not surprising as usually economies are situated in a growing state although it allows for more outliers (Figure 1 & 2) which might affect the results. Secondly, the sample for central banks interest decisions shows a bias as there appear to be no surprise factor inputs in times of economic expansion such that the outcomes are uninterpretable. A final limitation of the sample is that the greater part of the tested macroeconomic data announcements does not provide evidence to have a relationship with precious metal price movements which makes the conclusion not generalizable for all precious metals and types of macroeconomic data announcements.

Furthermore, a limitation of the model is the assumption that all causality runs through the U.S dollar index. Although the assumption is supported by previous literature and the results of this study it can be arguable that some causality is excluded from the U.S. dollar index and other explanatory variables are omitted from the regression.

Conclusion

The study examined the relation between returns of gold and silver to 8 different

macroeconomic events in specific states of the economy. The model used 5-minute interval intraday data over a 10 year period from 2008 till 2017. Variables for the surprise elements of the news announcements, control for the U.S. dollar effect and a dummy variable for the state of the economy were combined to construct the multilinear regression model. Although economic reasoning suggests differently not all macroeconomic events showed to have a significant effect on gold and silver prices which is in line with previous research. Only the events that proved to have a significant effect (Consumer Confidence, Nonfarm payrolls and ECB/Fed interest rate decision) were relevant to the research as macroeconomic data

sensitivity is essential for the interpretation of the results. Furthermore, the results did show convincing evidence for the U.S. dollar index being an explanatory variable of gold and silver prices. These empirics prove that the assumption of all causality running through the U.S. dollar index is undisputed. To finally address the objective of the paper the dummy variable for the state of the economy was analyzed. In line with the hypothesis, the variables that were tested significantly for the macroeconomic news events showed significance for a

contractionary state of the economy as well except for the central bank interest rate decision events for which the disparity in the results is assumed to be caused by a bias in the sample. Therefore the alternative hypothesis was only accepted for silver during the consumer confidence and nonfarm payroll events such that the research does suggest an additional

(16)

15 structural change in times of economic contraction. No conclusion could be made for gold and the remainder of the events for silver.

Although there is enough statistical evidence for the previously mentioned events more research has to be conducted to confirm the results. It is remarkable that exactly the events that have a significant effect on precious metals also show significance for the economic state dummy variable but most of the macroeconomic announcements that were tested turned out to have no relation with gold and silver price changes and therefore were neglected. A recommendation for future research is to analyze different events that seem to have a

relationship with precious metal price movements or analyze the effect on other asset classes to find further confirmation for the proposition. Conducting the research in another time interval could enrich the findings of this study as well.

(17)

16

Appendix A

Welsch’s t-test statistics

Gold ADP NFP BoE CC CPI ECB Fed NFP PPI

σew 0.012% 0.011% 0.017% 0.029% 0.013% 0.027% 0.018% 0.016%

t-value -0.173 -0.885 1.032 -0.070 -2.358 1.879 -4.202 -0.122

degrees of freedom 318 243 224 227 250 244 355 219

p-value 0.860 0.377 0.303 0.945 0.019b _0.061a _0.000c _0.903

Silver ADP NFP BoE CC CPI ECB Fed NFP PPI

σew 0.025% 0.020% 0.032% 0.056% 0.024% 0.042% 0.059% 0.038%

t-value 0.000 0.300 3.986 0.179 1.441 1.297 2.499 1.968

p-value 1 0.765 0.000c _0.858 _0.151 _0.195 _0.013b _0.050b

USDX ADP NFP BoE CC CPI ECB Fed NFP PPI

σew 0.005% 0.004% 0.007% 0.006% 0.006% 0.010% 0.008% 0.004%

t-value 1.548 0.274 -1.367 1.545 1.887 4.270 1.586 3.415

p-value 0.123 0.780 0.173 0.124 0.060a _0.000c _0.114 _0.001c

The coefficients in the table represent the statistics that are used to conduct the Welch’s t-test. ‘σew‘ denotes the Welch’s standard deviation of the error term. The statistically significant results are marked in bold.

Appendix B

Raw regression data for ADP nonfarm payroll

_cons .000136 .0001196 1.14 0.258 -.0001008 .0003729 C -.0000737 .0003644 -0.20 0.840 -.0007955 .0006482 Δ -1.097155 .2664819 -4.12 0.000 -1.625053 -.569256 ZADP 9.17e-06 .000123 0.07 0.941 -.0002346 .0002529 ΔPGLD Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00127 R-squared = 0.1995 Prob > F = 0.0002 F(3, 114) = 7.08 Linear regression Number of obs = 118 . regress ΔPGLD ZADP Δ C, vce(robust)

(18)

17

Raw regression data for BoE interest rate decision

_cons -.0000195 .0002989 -0.07 0.948 -.0006117 .0005727 C -.0001433 .0007614 -0.19 0.851 -.0016518 .0013651 Δ -1.62614 .4827348 -3.37 0.001 -2.582434 -.6698463 ZADP -.0003127 .0002746 -1.14 0.257 -.0008567 .0002312 ΔPSLVR Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .001116673 117 9.5442e-06 Root MSE = .00297 Adj R-squared = 0.0760 Residual .001005363 114 8.8190e-06 R-squared = 0.0997 Model .000111309 3 .000037103 Prob > F = 0.0073 F(3, 114) = 4.21 Source SS df MS Number of obs = 118 . regress ΔPSLVR ZADP Δ C

(19)

18

Raw regression data for consumer confidence

Raw regression data for CPI

_cons .0000675 .0001103 0.61 0.542 -.0001511 .0002861 C -.0000405 .0003543 -0.11 0.909 -.0007423 .0006614 Δ -1.405199 .3798407 -3.70 0.000 -2.15759 -.6528073 ZCBcc .0000131 .0001126 0.12 0.908 -.0002099 .0002361 ΔPGLD Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .0011 R-squared = 0.3362 Prob > F = 0.0005 F(3, 115) = 6.38 Linear regression Number of obs = 119 . regress ΔPGLD ZCBcc Δ C, vce(robust) _cons .0003886 .0002198 1.77 0.080 -.0000468 .0008239 C -.0011273 .000671 -1.68 0.096 -.0024564 .0002017 Δ -2.657669 .6275372 -4.24 0.000 -3.9007 -1.414639 ZCBcc -.0005198 .0002339 -2.22 0.028 -.0009831 -.0000565 ΔPSLVR Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00218 R-squared = 0.3722 Prob > F = 0.0004 F(3, 115) = 6.51 Linear regression Number of obs = 119 . regress ΔPSLVR ZCBcc Δ C, vce(robust) _cons .0000863 .0001904 0.45 0.651 -.0002908 .0004634 CCPI -.0000276 .0007891 -0.03 0.972 -.0015907 .0015355 ΔCPI -1.751849 .3987577 -4.39 0.000 -2.541785 -.9619131 ZCPI -.0001742 .0002516 -0.69 0.490 -.0006725 .0003242 ΔPGLDCPI Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00199 R-squared = 0.2415 Prob > F = 0.0000 F(3, 114) = 9.00 Linear regression Number of obs = 118 . regress ΔPGLDCPI ZCPI ΔCPI CCPI, tsscons vce(robust)

(20)

19

Raw regression data for ECB interest rate decision

_cons .0001766 .0002702 0.65 0.515 -.0003587 .0007118 CCPI -.0003321 .0016778 -0.20 0.843 -.0036557 .0029916 ΔCPI -2.818571 .7600054 -3.71 0.000 -4.324136 -1.313006 ZCPI -.0000131 .000436 -0.03 0.976 -.0008768 .0008507 ΔPSLVRCPI Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00343 R-squared = 0.2130 Prob > F = 0.0030 F(3, 114) = 4.93 Linear regression Number of obs = 118 . regress ΔPSLVRCPI ZCPI ΔCPI CCPI, tsscons vce(robust)

_cons .0000251 .0001493 0.17 0.867 -.0002711 .0003213 CEZ .0001424 .000228 0.62 0.534 -.0003101 .0005949 Δ -.8883447 .2049422 -4.33 0.000 -1.295098 -.4815911 ZECB -.0004076 .0003023 -1.35 0.181 -.0010077 .0001924 ΔPGLD Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00118 R-squared = 0.3315 Prob > F = 0.0003 F(3, 97) = 6.77 Linear regression Number of obs = 101 . regress ΔPGLD ZECB Δ CEZ, vce(robust)

_cons .0003438 .0002733 1.26 0.211 -.0001987 .0008862 CEZ -.0006345 .0004479 -1.42 0.160 -.0015236 .0002545 Δ -1.219336 .3481513 -3.50 0.001 -1.91032 -.5283525 ZECB -.0010469 .0002914 -3.59 0.001 -.0016253 -.0004685 ΔPSLVR Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00225 R-squared = 0.3086 Prob > F = 0.0000 F(3, 97) = 9.99 Linear regression Number of obs = 101 . regress ΔPSLVR ZECB Δ CEZ, vce(robust)

(21)

20

Raw regression data for Fed interest rate decision

Raw regression data for nonfarm payrolls

_cons .0002653 .0002235 1.19 0.239 -.0001804 .000711 C -.0013505 .0007135 -1.89 0.062 -.0027731 .0000721 Δ -1.098153 .3737433 -2.94 0.004 -1.843376 -.3529301 ZFed -.0004368 .0005307 -0.82 0.413 -.0014949 .0006214 ΔPGLD Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00201 R-squared = 0.3051 Prob > F = 0.0015 F(3, 71) = 5.72 Linear regression Number of obs = 75 . regress ΔPGLD ZFed Δ C, tsscons vce(robust)

_cons .000829 .0007211 1.15 0.254 -.0006088 .0022668 C -.0020361 .0019925 -1.02 0.310 -.0060091 .001937 Δ -2.228907 .6733992 -3.31 0.001 -3.571627 -.886187 ZFed -.0006455 .000732 -0.88 0.381 -.0021051 .0008141 ΔPSLVR Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .002825375 74 .000038181 Root MSE = .00576 Adj R-squared = 0.1311 Residual .002355516 71 .000033176 R-squared = 0.1663 Model .000469859 3 .00015662 Prob > F = 0.0046 F(3, 71) = 4.72 Source SS df MS Number of obs = 75 . regress ΔPSLVR ZFed Δ C _cons .0000803 .0002041 0.39 0.695 -.0003241 .0004848 C .0006698 .0004777 1.40 0.164 -.0002766 .0016162 Δ -1.24996 .2011323 -6.21 0.000 -1.64844 -.8514809 ZNFP -.000091 .0002046 -0.44 0.657 -.0004964 .0003144 ΔPGLD Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00189 R-squared = 0.3727 Prob > F = 0.0000 F(3, 113) = 14.37 Linear regression Number of obs = 117 . regress ΔPGLD ZNFP Δ C, vce(robust)

(22)

21

Raw regression data for PPI

_cons .0023451 .0008727 2.69 0.008 .0006162 .004074 C -.0033429 .0017792 -1.88 0.063 -.0068678 .000182 Δ -2.09014 .7893819 -2.65 0.009 -3.654048 -.5262321 ZNFP .0021451 .0007739 2.77 0.007 .0006119 .0036783 ΔPSLVR Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00788 R-squared = 0.1555 Prob > F = 0.0012 F(3, 113) = 5.63 Linear regression Number of obs = 117 . regress ΔPSLVR ZNFP Δ C, vce(robust) _cons .0000457 .0000752 0.61 0.544 -.0001033 .0001947 C -.0001578 .0004891 -0.32 0.748 -.0011266 .0008109 Δ -1.202417 .1948056 -6.17 0.000 -1.58829 -.816545 ZPPI -.0000597 .0000804 -0.74 0.460 -.000219 .0000997 ΔPGLD Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .00104 R-squared = 0.2520 Prob > F = 0.0000 F(3, 115) = 13.61 Linear regression Number of obs = 119 . regress ΔPGLD ZPPI Δ C, vce(robust)

_cons .000482 .0003592 1.34 0.182 -.0002296 .0011936 C -.0009766 .0009171 -1.06 0.289 -.0027933 .00084 Δ -1.513977 .6834938 -2.22 0.029 -2.867846 -.1601073 ZPPI -.0003453 .0003326 -1.04 0.301 -.0010041 .0003134 ΔPSLVR Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .001558589 118 .000013208 Root MSE = .00357 Adj R-squared = 0.0371 Residual .001462615 115 .000012718 R-squared = 0.0616 Model .000095974 3 .000031991 Prob > F = 0.0618 F(3, 115) = 2.52 Source SS df MS Number of obs = 119 . regress ΔPSLVR ZPPI Δ C

(23)

22

Bibliography

Almeida, A., Goodhart, C., & Payne, R. (1998). The Effects of Macroeconomic News on High Frequency Exchange Rate Behavior. Journal of Financial and Quantitative Analysis, 33(3), 383-408.

Andritzky, J., Bannister, G., & Tamirisa, N. (2007). The impact of macroeconomic announcements on emerging market bonds. Emerging Markets Review, 8(1), 20-37.

Antoniou, A., & Foster, A. J. (1992). The effects of futures trading on spot price volatility: Evidence for Brent crude oil using GARCH. Journal of Business Finance & Accounting, 19(4), 473-484.

Arouri, M., Hammoudeh, S., Lahiani, A., & Nguyen, D. (2012). Long memory and structural breaks in modeling the return and volatility dynamics of precious metals. The Quarterly Review of

Economics and Finance, 52(2), 207-218.

Blomberg, B. S., Hessa, G., & Weerapana, A. (2004). Economic conditions and terrorism. European

Journal of Political Economy, 20(2), 463-478.

Borensztein, E., & Reinhart, C. (1994). The Macroeconomic Determinants of Commodity Prices. IMF

Economic Review, 41(2), 236-261.

Burns, A. F., & Mitchell, W. C. (1946). Measuring Business Cycles. National Bureau of Economic

Research, 46(1), 115-202.

Cai, J., Chueng, Y.-L., & Wong, M. (2001). What moves the gold market? The Journal of Futures

Market, 21(3), 257-278.

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.

Christie-David, R., Chaundry, M., & Koch, T. W. (2000). Do macroeconomics news releases affect gold and silver prices? Journal of Economics and Business, 52(5), 405-421.

Ehrmann, M., & Fratzscher, M. (2005). Exchange rates and fundamentals: new evidence from real-time data. Journal of International Money and Finance, 24(2), 317-341.

Elder, J., Miao, H., & Ramchander, S. (2012). Impact of macroeconomic news on metal futures.

Journal of Banking & Finance, 36(1), 51-65.

Escribano, A., & Granger, C. W. (1998). Investigating the Relationship between Gold and Silver Prices. Journal of Forecasting, 17, 81-107.

Hammoudeh, S., Yuan, Y., McAleer, M., & Thompson, M. (2010). Precious metals–exchange rate volatility transmissions and hedging strategies. International Review of Economics &

Finance, 19(4), 633-647.

Hess, D., Huang, H., & Niessen, A. (2008). How do commodity futures response to macroeconomic news? Financial Markets and Portfolio Management, 22, 127-146.

Kocagil, A. (1997). Does futures speculation stabilize spot prices? Evidence from metals markets.

Applied Financial Economics, 7(1), 115-125.

Mayer, J. (2008). The financialization of commodity markets and commodity price volatility. The

(24)

23 Rangel, J. (2011). Macroeconomic news, announcements, and stock market jump intensity dynamics.

Journal of Banking & Finance, 35(5), 1263-1276.

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.

Roache, S. K., & Rossi, M. (2010). The effects of economic news on commodity prices. The

Quarterly Review of Economics and Finance, 50(3), 377-385.

Silvapulle, P., & Moosa, I. A. (1999). The Relationship between Spot and Futures Prices: Evindence from the Crude Oil Market. The Journal of Futures Markets, 19(2), 175-193.

Yang, J., Balyeat, B., & Leatham, D. (2005). Futures trading activity and commodity cash price volatility. Journal of Business Finance & Accounting, 32(1), 297-323.

Relevance of the state of the economy in the impact of macroeconomic data events on precious metal prices