key: cord-0681728-3rwus3bp
authors: Salisu, Afees A.; Akanni, Lateef; Raheem, Ibrahim
title: The COVID-19 global fear index and the predictability of commodity price returns
date: 2020-07-22
journal: J Behav Exp Finance
DOI: 10.1016/j.jbef.2020.100383
sha: 73db8c2c350717b355b213188a57af442dd96c78
doc_id: 681728
cord_uid: 3rwus3bp

In this paper, we subject the global fear index (GFI) for the COVID-19 pandemic to empirical scrutiny by examining its predictive power in the predictability of commodity price returns during the pandemic. One of the attractions to the index lies in its coverage as all the countries and by extension regions and territories in the world are considered in the construction of the index. Our results show evidence of a positive relationship between commodity price returns and the global fear index, confirming that commodity returns increase as COVID-19 related fear rises. By way of extension, we further establish that commodity market offers better safe-haven properties than the stock market given the negative association between GFI and the latter. Finally, the GFI series improves the forecast accuracy of the predictive model for commodity price returns and its forecast outcome outperforms the historical average (constant returns) model both for the in-sample and out-of-sample forecasts. Our results are robust to alternative measures of pandemics.

The coronavirus outbreak was first reported on the 31 st December, 2019 in Wuhan, Hubei province in China. Three months later, it has generated into a pandemic outbreak and has posed adverse consequences on the global economy. Business activities are most affected by this shock. For instance, there is a wide disruption in the global supply chains, weaker demand for goods and services, significant decline in tourism and business travels, lower consumer and business confidence, increase in the rate of risk aversion in the financial system and fall in equity prices, among others. Human normal routines are not spared, as virtually all countries have put in place some forms of restrictions on movements 1 . It is safe to assert that the world is experiencing a global crisis, something different to what we are used to such as the financial, debt, and currency crises. Historically, global crises have attendant effect on global variables, such as commodity prices. For instance, Sanderson, Raval and Sheppard (2015) note that commodity prices have fallen to their lowest level since the financial crisis and -by at least one measure -to the lowest this century. Also, Zhang and Broadstock (2019) documents a dramatic change in the nature of connectedness in global commodity prices following the 2008 global financial crisis. They show that co-dependence in price-changes among seven major commodity classes goes from a pre-crisis average of 14.82% to a strikingly larger average of 47.87% in the period following the crisis, and which has endured until now.

Thus, as COVID-19 has been declared to be pandemic, its impact is expected to transcend a single country, region or continent. In other words, it is expected to affect the global demand and supply of goods and services, chief among which are commodity prices and stock prices.

Since the outbreak of the pandemic, oil prices seem to have declined significantly 2 while stock, such as the EOG Resources stock which is oil-related, has dropped by 50%. The importance of accurate prediction of commodity prices in times of crisis cannot be overemphasized and can be explained based on the two reasons. First, Zhang and Broadstock (2019) showed that there is high connectedness between commodity markets and other markets. Relatedly, the concept of commodity financialization has raised interest among academics and policymaker to the extent that commodity is now considered as a core asset classification by financial investors (Cheng and Xiong, 2014 and Zhang, Yongzeng and Jianghong, 2017) . This has also led to increase in speculative activities (Hamilton and Wu, 2014) . The same argument goes for commodity currency (Chen and Rogoff, 2003) . Hence, the inability to accurately predict commodity prices might lead to commodity price crash and thus spillover to other markets, which can further exacerbate the intensity of the already existing crisis. However, taking this information on board helps investors makes decision of portfolio adjustments and asset pricing.

It also helps academics develop models that will enhance forecasting.

Secondly, evidences have shown that some of the post-crisis consequences are the high volatility and reversal in commodity prices and uncertainty in financial markets (Antonakakis, Chatziantonioub and Filis, 2017) . No doubt, at the moment, there has been reversal in the prices of some commodities. The stability can be explained under the guise of lower demand attributed to the "global lockdown" policies set by government. The benefit of accurate prediction in bear times is of utmost importance to both investors and policymakers. Starting with the investors, studies have shown that the inclusion of commodity assets in the portfolios of investors is able to hedge against risk, thus offering higher returns, especially when the market is in bear mode (Öztek and Öcal, 2017; Zhang et al., 2017; and Aït-Youcef, 2018) show the relationship between stock markets and agricultural commodity is significant during turbulent times, such as the 2008 financial crisis.

Some studies have argued to account for the extreme periods when modelling financial series (Andersen, Bollerslev and Diebold, 2017) . We take a more radical, but similar approach.

Specifically, we hypothesize that focusing solely on the extreme periods will reveal the true relationship among the series in the model. Whereas, combining bear and bull periods while modelling will distort some inherent information present in the data. Another justification for the sole reliance on the bear period is due to the current "lockdown" restrictions in most countries of the world. Hence, it could be subtly argued that financial and economic series are inoperative. What's new, some markets have been officially shut down 3 and new policy guidelines have been put in place in the financial markets 4 . 3 On the 26th March, 2020, the UK government suspended the housing and property market in the country. It was announced that this suspension will, at least, last during the stay-at-home order. 4 In the UK, a three months moratorium has been placed on private loans and mortgages in March, 2020 and has been further extended by another quarter. The Reserve Bank of Australia announced short term funding for the banking system. Financial regulators in China pledge to enhance liquidity support for the market, increasing credit support and extending repayment period for enterprises. Bank of Japan to provide loans against corporate debt J o u r n a l P r e -p r o o f

Based on the foregoing, we offer the following innovations to the existing literature on the predictability of commodity prices. First, we construct a predictive model for commodity prices that accounts for the impact of the COVID-19 pandemic. This is a major contribution to the literature as forecasting economic and financial series during crisis is usually a daunting task for economic, financial and policy analysts. To the best of our knowledge, we are not aware of similar studies that had conducted similar exercise, especially during the recent crisis period.

To achieve this objective, we select twenty-four (24) commodities covering precious metals, agriculture and industrial inputs that are allowed to trade during the period of the pandemic.

Second, we utilize the global fear index (GFI) constructed by Salisu and Akanni (2020) to capture the fear/panic associated with the pandemic. One of the strengths of the index lies in its coverage as all the countries and by extension regions and continents in the world are considered in the construction of the index. Third, we test the predictive power of GFI in the predictability of commodity prices. Several plausible scenarios are rendered to validate our results. Fourth, we made a safe-haven comparison analyses between commodity prices and stock markets in order to examine the best investment options for policymakers and investors.

The recency of COVID-19 has led to the emergence of studies that had examined how the pandemic has impacted the commodity markets. For instance, Sharif, Aloui, and Yarovaya A glimpse of our results suggests a high predictive power of the GFI in the predictability of commodity price returns. The inclusion of the index further improves the predictive model for commodity price returns and more specifically, its forecast outcome outperforms the historical and 0% interest rate. The European Commission provides guarantees, to member countries, to ensure banks keep providing loans to the customers who need them.

J o u r n a l P r e -p r o o f Journal Pre-proof average (constant returns) model. Further results suggest that commodity market appears to offer better safe-haven properties than the stock market during the pandemic.

The rest of the study is structured as follows: Section 2 provides the methodology; the results are presented and discussed in Section 3 while Section 4 concludes.

We formulate a predictive model for commodity price returns where the global fear index serves as a predictor. The underlying intuition is that as the panic associated with the pandemic becomes severe (implying an increase in the index), the global economic activity reduces, which would lower trading in the commodity market and by extension returns. Since our focus is on the COVID-19 period and given the need to consider a long range of data sample, we utilize panel data where the various commodities are pooled over the available data period for COVID-19. In order to account for any inherent heterogeneity in the selected commodities, we favour the heterogeneous panel models (see for example, Robertson and Symons, 1992; and Pesaran and Smith, 1995) . The analyses using heterogeneous panel can be done using each commodity's time series regression, or using various estimation methods described in the earlier papers (see Baltagi, 2008; Eberhardt, 2012; Chudik and Pesaran, 2015; Chudik et al., 2016; Westerlund and Narayan, 2016; Salisu and Isah, 2017; Ditzen, 2018; and Salisu and Ndako, 2018) . Notwithstanding the inherent heterogeneity in the commodities, a number of studies have also demonstrated the need to account for unobserved common factors (see Chudik and Pesaran, 2015; Chudik et al., 2016) . In other words, regardless of the specificities of these commodities, they may be driven by some unobserved common factors. Thus, following the studies of Chudik and Pesaran, (2015) , Chudik et al. (2016) , and Ditzen (2018), we construct a predictive panel data model for commodity price returns 5 where the GFI series is the only predictor: 6

5 In addition to the suitability of the model for long T, it also helps resolve any inherent nonstationarity which is a suspect when dealing with long T. It also accommodates mixed order of integration and facilitates the estimation of long run and short run dynamics including the speed of adjustment. 6 We are grateful to Ditzen (2018 Ditzen ( , 2019 for providing the relevant codes for the estimation of dynamic panel data models with dynamic common correlated effects. Thus, in addition to allowing for heterogeneity in the predictability, we also incorporate unobserved common factors for the selected commodities.

Note that we allow for up to five lags given the underlying frequency for our analysis which is daily and therefore the proximity of the data points can be exploited to account for more dynamics in the predictive model. Thus, in addition to the behaviour of the individual parameters, testing for the overall sign and significance of these parameters jointly is crucial to arrive at a distinct conclusion on the predictability of GFI on the return series. The testable null hypothesis of no predictability can therefore be expressed as We account for an important feature of price returns which include commodities and stock prices. These prices series tend to exhibit day-of-the-week effect (for a review of literature, see Zhang et al., 2017) . In order to account for this feature, we regress the return series on dummy variables constructed for the five days of the week, to obtain the day-of-the-week adjusted returns (See Salisu and Akanni, 2020) J o u r n a l P r e -p r o o f the VIX index for different markets thus allowing us to narrow the systemic risk to commodity market. On this basis, the single predictor model in [1] is extended to become:

where it z is a measure of systemic risk, and  i is the corresponding coefficient for each crosssection while equation [2] remains the same. As a result, we are able to capture both observed and unobserved common factors in equation [3] . We favour an approach that allows us to use the risk-adjusted returns rather than estimating the direct impact of the systemic risk on returns.

This also helps circumvent any potential endogeneity bias that may result from the correlation between the two predictors. Therefore, we first regress the return series on z , that is, where t h f is the forecast horizon;ˆr MSE and ˆu MSE respectively are the squared errors of restricted and unrestricted predictive models and they are respectively computed as: ; P is the amount of predictions that the averages are computed. Lastly, the statistical significance of regressing t h f on a constant confirms the CT test.

Our data set consists of commodity prices (in USD) of the 24 major commodities traded globally, commodity-related volatility indexes, including crude oil gold and silver, and the global fear index (GFI) 8 . Daily price series for each of the commodities were retrieved from www.investing.com historical data while the volatility index data were collected from the Chicago Board of Exchange (CBOE) database (http://www.cboe.com/). On the other hand, the GFI series was obtained from Salisu and Akanni (2020) 9 . The start period was selected to conform with the declaration of the COVID-19 as a global pandemic by the World Health Organisation (WHO) 10 . Hence, the daily data was collected between March 11 and May 18, 2020, spanning 46 observations. Some graphical representations are rendered to highlight the co-movements between each of the commodity prices and the global fear index (see Figure 1 ).

The graphical illustration reveals that there are co-movements in the commodity prices and the global fear index. 

Table 2 summarizes the estimated predictability results for the global fear index predictability of commodity price returns. As discussed in the methodology section, we considered two GFI predictability models, the baseline model with a single factor predictive model and the macroeconomic volatility adjusted model. Following the prominent feature of traded stock and J o u r n a l P r e -p r o o f commodity prices as documented in empirical literature (see Zhang et al., 2017; Akanni, 2020), we adjust the commodity price returns for day-of-the-week effects. The estimated joint coefficient of five-period lag of the global fear index as summarised in Table 2 shows a positive and significant relationship with commodity price returns. In addition, the estimated coefficient after accounting for macroeconomic volatility effect shows a positive and significant impact of global fear index on commodity price returns. These positive results conform with findings in empirical economics and finance literature on the hedging features of commodities including metals such as gold, silver, platinum and palladium (Chen, Turnovsky, & Zivot, 2014; Salisu, Raheem, and Ndako, 2020) and agricultural commodities such as coffee and cocoa (see Tule, Salisu, & Chiemeke, 2019) . These studies found that investors exploit the wealth protection features of commodities by holding and spreading their investment portfolios from stocks and bonds especially during crisis, such as the COVID-19 pandemic. indicates sum of coefficients of the natural log of global fear index up to five period lags. The statistical significance for the joint coefficients is computed using the Wald test for coefficient restriction. The F-statistic for the Wald tests are in parenthesis while the corresponding probability values are reported in squared brackets. Asterisks *** and ** denote statistical significance at 1% & 5% levels, respectively.

Next, we carried out the forecast evaluation of the predictive models by dividing the data into in-sample and out-of-sample periods. For the in-sample predictability evaluation, 75% of the entire data sample is used, while the out-of-sample periods include 7-day and 14-day ahead forecast horizons, for robustness purposes. The forecast performance evaluation is carried out using both the CT (2008) and CW (2007) tests. We evaluate the predictability performance of the baseline single predictor model against the constant returns historical average model. In addition, we also compared the predictability performance of the volatility adjusted model with the historical average model. The results are summarised in Table 3 

For robustness purpose, we conduct two sets of additional analyses. First, we evaluate the robustness of the COVID-19 global fear index as an alternative measure of the pandemic in the predictability commodity price returns. To do this, we employ the Infectious Disease Equity Market Volatility (ID-EMV) data which captures a broad range of pandemics (see Baker et al., 2020) . The ID-EMV dataset comprehensively covers epidemics and pandemics that predate the emergence of the COVID-19 some of which include the Severe Acute Respiratory Syndrome, Ebola virus, Middle East Respiratory Syndrome and Zika virus.

Hence, it is the most suitable alternative index to measure uncertainties related to fear the COVID-19 health risks. In the same vein as the main results, we evaluate the predictability performance of the single-predictor model, i.e ID-EMV against the gfi-based and historical average models.

The estimated predictability as well as the in-sample and out-of-sample forecast evaluation results are summarised in J o u r n a l P r e -p r o o f , 1 i t emv  are the coefficients of health-news and stock market volatility index predictors respectively. The C-T stat indicates the Campbell-Thompson (2008) test statistics. The asterisk *** indicates statistical significance at 1% level.

For the second additional results for robustness, we follow the empirical literature that a vast support for the hedging potential and characteristics of commodities for stock and other financial markets. These hedging characteristics can be said to be more relevant during crisis period. Several factors characterise the desirable features of these commodities, particularly gold and other precious metals, and these include: (i) the intrinsic value of most commodities neither depends on prospective cash flows nor carries a default risk; (ii) precious metals universally acceptable and scarce; (iii) most commodities have relative supply inelasticity and their observed counter-cyclical demand characteristics make them store of value; (iv) their protection attributes and properties are commonly referred to by many investors, individuals and the media (see Arnold and Auer, 2015) . Therefore, to further evaluate the hedging features of commodities during the current volatile and turbulent global economy as a result of the COVID-19 pandemic, we extend the analysis to evaluate the global fear index predictability of OECD countries stock returns since the declaration of COVID-19 as a global pandemic by the WHO.

The estimated stock returns predictability results summarised in the upper pane of Table 5 reported a negative and statistically significant coefficient. The joint coefficient of the fivedays lagged coefficient of the global fear index has a negative sign indicating that an adverse effect of COVID-19 on stock returns. contrary to the positive coefficient reported for commodity price returns predictability, the result for stock returns, as expected, shows that inverse relationship. By implication, as investors' fear associated with the COVID-19 pandemic increase, stock returns decline, while commodity price returns increase (see Table   J o u r n a l P r e -p r o o f 2). We further applied the two criteria for evaluating the forecast performance as discussed in the methodology section, that is the CT (2008) and CW (2007) tests. The in-sample period and two out-of-sample forecast horizons confirm that the global fear index predictability of stock returns outperforms the historical average constant stock returns model. to five period lags. The statistical significance for the joint coefficients is computed using the Wald test for coefficient restriction. While the F-statistic for the Wald test is reported in parenthesis, the probability value is in squared brackets. The forecast performance of the GFI-predictor stock returns predictability model is compared with the constant returns model. The asterisk *** denotes statistical significance at 1% levels.

The central theme of this study is to evaluate the predictability of commodity prices during economic and financial crisis, particularly the current unprecedented COVID-19 pandemic. In addition, we evaluate whether investing in commodities during these trying times serves as a good hedge to investors against volatilities and decline witnessed conventional assets, especially equities. We employed the global fear index (GFI) recently constructed by Salisu and Akanni (2020) in a predictive model for 24 major commodity prices. Our results show evidence of positive relationship between commodity price returns and the global fear index, confirming that commodity returns increase as COVID-19 related fear rises. The predictability performance further confirms that the outperformance of the predictive model above the constant return predictor, both in-sample and out-of-sample.

For robustness purpose and in order to further confirm the hedging potential of commodities during crisis, we extend the analysis to evaluate the performance of OECD stocks during the pandemic period. The estimated result report a negative relationship between stock returns and the global fear index with statistically significant forecast performance, in-sample and out-of-J o u r n a l P r e -p r o o f sample. Thus, the results support findings in several previous empirical literature that investing in commodity, especially during crisis and turbulent periods, serves as a goo hedge against the volatility and declines in stock markets.

How index investment impacts commodities: A story about the financialization of agricultural commodities

COVID 19's impact on crude oil and natural gas S&P GS Indexes

Roughing it up: Including jump components in the measurement, modelling, and forecasting of return volatility

Oil shocks and stock markets: Dynamic connectedness under the prism of recent geopolitical and economic unrest

What Do Scientists Know about Inflation Hedging?

The unprecedented stock market impact of COVID-19 (No. w26945)

Econometric Analysis of Panel Data

Forecasting inflation using commodity price aggregates

The financialization of commodity markets

Predicting excess stock returns out of sample: Can anything beat the historical average?

Approximately normal tests for equal predictiveaccuracy in nested models

Common correlated effects estimation of heterogeneous dynamic panel data models with weakly exogenous regressors

Long-Run Effects in Large Heterogeneous Panel Data Models with Cross-Sectionally Correlated Errors Essays in Honor of Aman Ullah

The Contagion Effects of the COVID-19 Pandemic: Evidence from Gold and Cryptocurrencies

Estimating Dynamic Common Correlated Effects in Stata

Estimating long run effects in models with cross-sectional dependence using xtdcce2

Estimating panel time series models with heterogeneous slopes

Risk premia in crude oil futures prices

Financial crises and the nature of correlation between commodity and stock markets

Some strange properties of panel data estimators

Estimating Long-Run Relationships from Dynamic Heterogeneous Panels

Revisiting the oil price and stock market nexus: A nonlinear Panel ARDL approach

Modelling stock price-exchange rate nexus in OECD countries: A new perspective

Constructing a global fear index for the COVID-19 pandemic. Emerging Markets Finance and Trade, forthcoming

The inflation hedging properties of gold, stocks and real estate: A comparative analysis

Revisiting oil-stock nexus during COVID-19 pandemic: Some preliminary results

Why commodities have crashed

COVID-19 pandemic, oil prices, stock market, geopolitical risk and policy uncertainty nexus in the US economy: Fresh evidence from the wavelet-based approach

Can agricultural commodity prices predict Nigeria's inflation?

Analysis of the impact of COVID-19 on the correlations between crude oil and agricultural futures

Testing for predictability in panels of any time series dimension

The effects of a 'black swan' event (COVID-19) on herding behavior in cryptocurrency markets: Evidence from cryptocurrency USD, EUR, JPY and KRW markets

The Day-of-the-Week Effects of Stock Markets in Different Countries

Global financial crisis and rising connectedness in the international commodity markets

The COVID-19 Global Fear Index (GFI) seeks to measure daily concerns and emotions on the spread and severity of COVID-19 since the pandemic declaration. Excessive fears could have significant implications on investment sentiments and decisions, and as such affecting prices such as stocks and oil prices. Relying on the official reports of COVID-19 cases and deaths across the globe, the GFI is a composite index of two factors; Reported Cases and Reported Deaths, on a scale of 0 to 100, respectively indicating no fear to extreme fear/panic. An index value of 50 is considered neutral, while anything higher signals more fear than usual. We employed what we referred to as the incubation period expectation in daily reported cases and deaths in constructing the index. The incubation period expectation is defined as the time expectation between when a person could be exposed to the Coronavirus and emergence of symptoms of the disease. According to the World Health Organization (WHO), most estimates of the incubation period for COVID-19 range from 1-14 days (WHO, 2020).

(RCI) which measures how far peoples' expectations on reported cases in the preceding 14days period veered from the present day's reported case. The choice of 14-day expectations represents the highest number of days for COVID-19 incubation period as defined by the WHO. The RCI for a given day is computed as the ratio of COVID-19 reported cases globally on the day under consideration and the sum of reported cases for the current day and reported cases at the start of the incubation period. The ratio is then multiplied by 100 to give the index on a scale between 0 and 100.The second component is the Reported Deaths Index (RDI) and measures how far peoples' expectations from reported deaths in the preceding 14-days period veered from the present day's reported deaths. The RDI, just as the RCI, is computed as the ratio of COVID-19 related reported deaths across the world and the sum of reported deaths on the day under consideration