Climate risks and market efficiency

Abstract

Climate science finds that the trend towards higher global temperatures exacerbates the risks of droughts. We investigate whether the prices of food stocks efficiently discount these risks. Using data from thirty-one countries with publicly-traded food companies, we rank these countries each year based on their long-term trends toward droughts using the Palmer Drought Severity Index. A poor trend ranking for a country forecasts relatively poor profit growth for food companies in that country. It also forecasts relatively poor food stock returns in that country. This return predictability is consistent with food stock prices underreacting to climate change risks.

Introduction

Regulators are increasingly worried about the extent to which stock markets efficiently price climate change risks. Most notably, Mark Carney, the head of the Bank of England, recently linked these risks to financial stability (Carney, 2015). Such risks include energy corporations’ exposure to carbon assets, which might be affected by future carbon prices or taxes. This so-called “stranded asset issue” has attracted the most discussion in regulatory and market circles at this point.¹ But climate change risks need not be so narrowly confined to carbon exposures. Vulnerability of corporations’ production processes to natural disasters amplified by climate change can impose significant damage to corporate profits, as we detail below. In particular, regulators are concerned that markets have had little experience in dealing with such risks and might not pay enough attention, and thereby underreacting to them as a result. Various regulatory bodies are promoting both voluntary and mandatory disclosures of corporations’ climate risk exposures to address this issue.² However, there is little systematic research on the topic of climate risks and market efficiency up to this point.

We tackle this important question by focusing on the efficiency with which the stock prices of food companies respond to trends in droughts across the world. The motivation for our study is that climate scientists have found that the trend increase in global temperatures exacerbates the risks of droughts, generating dispersion across countries with many potentially adversely affected while some might actually benefit (Trenberth et al., 2014). Among the natural disasters that might be amplified by climate change, including drought, heat waves, floods, and cold spells, drought is considered one of the most devastating for food production.³ The food industry in countries suffering adverse trends in droughts are likely to experience lower profits since this industry is the most reliant on water and hence the most sensitive to drought risk (Blackhurst et al., 2010). As we document below, most countries’ food industries are comprised of small to medium sized firms that are significantly exposed to the climate conditions of their country of origin.⁴ As a result, the food companies of a country with an adverse (favorable) drought trend are likely to experience relatively poor (good) subsequent profit growth.

We test the hypothesis of whether food stocks are efficiently pricing in such risks associated with these trends for future food industry cashflows. Using data from thirty-one countries with publicly traded equities in the food industry, we develop and test our hypothesis in three steps. First, we measure time trends in droughts across countries with publicly-traded equities in the food industry and categorize countries into those with negative (or adverse trends) versus those with non-negative (or in some instances even positive trends) by using publicly available data up to a given year $t$. Second, we then document that these trend rankings, measured using data only up to year $t$, can forecast the relative performance of food industry cashflows (in years $t+1$, $t+2$, …), i.e. the food industries in countries with negative trends experience subsequently poor profit growth relative to the food industries in countries with positive trends.

Third, we test the null hypothesis of market efficiency. These trends, which are publicly available information in a given year $t$, should not then forecast future food industry stock returns to the extent that markets have efficiently priced in the implications of these trends for future cashflows. On the other hand, to the extent negative (positive) trend rankings forecast poor (good) relative stock price performance for food industries in those countries points to markets not sufficiently pricing in the information contained in these trends for future cashflow growth, i.e. that stock markets are under-reacting to climate change risks.

We begin by estimating drought time trends by using the Palmer Drought Severity Index (PDSI), a widely used monthly metric in climate studies (Palmer, 1965). PDSI combines information such as temperature and the amount of moisture in the soil to create an index that does an accurate job of measuring drought intensity. Less positive values of PDSI are associated with more drought-like conditions. While not perfect, it is by far the most widely used in climate studies and the most readily available (Alley, 1984). Globally, it is available at the country level and goes back to the early 1900s.

For each of the 31 countries in our international sample, we construct a new measure of a country’s vulnerability to droughts as a result of climate change. Recall that the premise of our measure is that climate studies point out that there is a time trend in global temperatures (see Fig. 1) leading to potentially differential trends in droughts across countries over time. Using long time series of PDSI for each country going back to the early 1900s, we can calculate $Tren{d}_{i,t}$, the time trend of drought for each country $i$ using data up to a given year $t$. We estimate this time trend using a trend-stationary model: an AR(1) model for drought (PDSI) that is augmented with a linear deterministic time trend.⁵ Consistent with earlier climate studies, we find that there is significant dispersion in trends towards droughts, with a more significant left-tail, i.e. more countries with statistically significant negative trends in drought than countries with positive or improving trends.

We sort countries based on their estimated trends in any given time $t$ into quintile groups, with the bottom or Quintile 1 group comprised of the negative trending countries and the top or Quintile 5 group comprised of the positive trending countries. That is, we use these time trends to rank which countries are most vulnerable to droughts (i.e. the negative time trends and rising risk) and least vulnerable to droughts (i.e. the positive time trends and falling risk). These drought trend rankings are stable over time and capture the long-run effect of climate change on a country’s drought vulnerability.

Our focus is on the spread in future performance of the food industries in the Quintile 1 or rising drought-risk group of countries relative to the Quintile 5 or falling drought-risk group of countries as opposed to the mean performance of the overall food sector (or the middle Quintiles 2–4 group of non-trending countries). The rationale is that the overall effect of climate change on global food production or crop yields is ambiguous Mendelsohn et al. (1994), while the spread in performance as driven by sensitivity to drought risk is clear cut. In other words, we are implementing a difference-in-difference estimate of the differential impact of drought trends on the stock market.

To this end, we then examine the extent to which these simple cross-sectional country rankings at year $t$ can forecast changes in food industry profitability (net income divided by total assets) and food industry stock returns across countries over the sample period of 1985 to 2014. Our dependent variable of interests are the change in profitability ratios and the returns of the FOOD industry of each country.⁶ FOOD combines food processing, beverage and agricultural companies. We focus on this aggregated industry portfolio as opposed to the finer industry classifications, which separate FOOD into smaller components. The reason is that drought is likely to have a direct impact on the profits of both food processing and agricultural companies.⁷ We confirm below that adverse trends in droughts have significant impact on all three sub-sectors of the Food industry.

We show that there is strong forecastability of changes in food industry profitability out a number of years. Countries with the negative time trends experience subsequently lower growth in profits than countries with positive time trends. For countries in the negative trend group, the mean cumulative change in profits from year $t$ to year $t+3$ is −0.46%. For those in the positive trend group, the corresponding mean is 0.61%. The difference has a $t$-statistic of 2.2. This spread of 1% is ten times larger than the unconditional sample mean and one-third of the unconditional standard deviation of annual changes in profitability ratios. We show that this spread persists even after accounting for different industry and country characteristics.

We can re-run our analysis by using only trend rankings calculated at the end of 1984. That is, for the remaining years of our sample from 1985–2014, we fix the rankings and just track this set of countries over time. Given how persistent the trend rankings are, we get similar results, which accentuates the point that our findings reflect long-run drought trends for long-run food industry profitability.

In an efficient market, such publicly available rankings, even though they forecast profits, should not then be able to forecast stock returns years out if the stock market is efficient. We show that these same rankings, however, do forecast stock returns. The food stocks in the negative trend group have an excess return of .33% per month. The stocks in the positive trend group have an excess return of .89% per month. The difference is 0.56% per month (or around 7% annualized) with a $t$-statistic of 2.03. This 7% annual difference is reasonable given the substantial spread in changes in profits across the two groups, as we explain below.

The results are similar whether we adjust the return spread using the global Sharpe (1964) CAPM, Carhart (1997) four factor model, or the currency factor model of Lustig et al. (2011). These adjustments make clear that while the mean returns of the middle group of non-trending countries is sensitive to the model of risk, the spread between the Quintile 1 group and Quintile 5 group is robust. Food stocks in the Quintile 1 group under-perform the stocks in the middle group, while stocks in the Quintile 5 group out-perform stocks in the middle group.

Using cross-country Fama and MacBeth (1973) regressions, we show that this excess return predictability remains even after we control for additional country and industry characteristics. This predictability is also significant even if we re-run our analysis by using only trend rankings calculated at the end of 1984. This predictability is present across sub-samples of 1985–1999 and 2000–2014. Nonetheless, we want to be modest about our excess predictability results since our international sample only has 31 countries. Given the food stocks in our sample are mostly small to medium sized firms, however, arbitrage would be very costly so that the large alpha of our long/short strategy does not mean there is easy money to be made.

We next conduct a placebo analysis where we repeat these exercises for all other industries and find that drought is uniquely tied to the FOOD industry. The next largest industry, which is however not statistically significant, is UTILITIES. It is known to be next to FOOD a highly water-reliant industry. This placebo test serves as a way to show that we are identifying climate change risks related to drought and our main results are not driven by unobserved differences in country characteristics (i.e. to address omitted-variables concerns in cross-country regressions).

In our robustness analyses (available in our Supplementary Internet Appendix), we consider a downside-risk CAPM and construct an alternative drought ranking measure as a country’s 36-month moving average of the PDSI (denoted as PDSI36m) net of the long-run mean of that country divided by the standard deviation of PDSI, with the mean and standard deviation estimated using data from 1900 to 1939. The cross-sectional rankings of this standardized PDSI36m measure are correlated with the drought trend rankings but are less persistent as they also capture prolonged droughts. We obtain similar results in both instances.

Our findings are related to the recent literature on attention and return predictability (see, e.g., Hong et al. (2007), DellaVigna and Pollet (2007), and Cohen and Frazzini (2008)), whereby the market underreacts to many types of value relevant information such as industry news, demographic shifts, and upstream–downstream relationships. Even for these types of obviously relevant news, the market can be inattentive.

Our first set of results on cashflows differs from prior work using weather shocks to estimate the damage to crops from climate change (Deschenes and Greenstone (2007), Schlenker and Roberts (2009), Dell et al. (2014)). This weather-economy literature argues that short-run temperature shocks estimated in a panel regression with location fixed effects are useful from an identification perspective to measure potential damages to food production from temperature increases. But the extrapolation to climate-change damages is uncertain given adaptation in the long run and potential intensification effects not captured in local weather shocks. Our drought-trends approach, along with a placebo analysis using other non-agricultural industries to address omitted variables in cross-country regressions, complements this literature in better measuring the long-run effects of climate change on agricultural industry profits.

Our second set of results on excess return predictability distinguish our work from earlier work on the pricing of weather derivatives, which focuses again on only short-term fluctuations in weather (see, e.g., Roll (1984), Campbell and Diebold (2005)). Our study of climate change risks and market efficiency helps characterize the nature of the potential inefficiencies, which might inform regulatory responses and be useful for practitioners interested in the construction of quantitative risk-management models (Shiller (1994)).

There is a large literature on the economic analysis of how to design government policies to deal with climate change (see, e.g., Stern (2007), Nordhaus (1994)), be it through emissions trading (Montgomery (1972)) or taxes (Golosov et al. (2014)). In contrast, our analysis highlights the role of markets in potentially mitigating the risks brought on or exacerbated by climate change. Understanding the role of financial markets in pricing climate risks is a natural one, though work is limited at this point with some notable exceptions. Bansal et al. (2014) argue that long-run climate risks as captured by temperature are priced into the market. Daniel et al. (2016) and Giglio et al. (2015) show how stock and real estate markets might help guide government policies assuming markets efficiently incorporate such climate risks. Our analysis suggests, however, that such climate risk information, at least when it comes to natural disasters, are incorporated into stock prices with a significant delay.

Our paper proceeds as follows. We present our data and discuss the PDSI metric in Section 2. In Section 3, we present the results of time trends in droughts. In Section 4, we present the results of drought trend rankings and predictability of changes in food industry profitability. In Section 5, we present the results of drought trend rankings and food stock excess return predictability. We conclude in Section 6.