Risk management of climatic volatility

This Commentary outlines a risk assessment approach for climatically exposed sectors like farmland, water, ski-tourism, insurance and travel. Project and risk models for these sectors are traditionally calibrated to historic occurrences of extreme events, such as drought, storm, and flood. Unfortunately frequencies of such events are now increasing, resulting in weather-related losses growing at a rate greater than inflation. This shift is becoming a source of error in investment risk models that assume the continuation of past climatic patterns. For example, insurance policies calibrated to long-term climate data would not have correctly priced recent climate-related losses (see Figure 1).

Figure 1. Economic losses from climate-related disasters have increased, with large spatial and interannual variations

Sources: Munich Re 2011, IPCC

In this Commentary we:

1. present data on changing weather event frequencies,
2. review traditional climate-exposed project risk models (with a particular focus on agricultural projects), and
3. overview an approach for adjusting traditional frameworks for changing frequencies of weather events.

We conclude with some comments on the likely impacts of these changes on the relative performance on different classes of farmland.

1. More frequent extreme climate events

In the past five years we have seen: 2008-2009: Severe droughts in Argentina, contributing to a 35% drop in crop production;

2010: Droughts in the Horn of Africa; south-western China and much of Russia;

2011-2013: Persistent droughts throughout Southern USA;

2012: Drought in the US Corn Belt, the worst for 50 years causing a 25% shortfall in  yields; drought continuing in south-western China; wettest summer weather in England  for 100 years;

2013: Drought continuing in the Horn of Africa, the longest recorded; and drought in the North Island of New Zealand, the worst for 70 years;  a “once in a century” widespread flood in Germany and Austria, three times the expected levels, exceeding the levels of the previous “once in a century” flood experienced in 2002.

These changes in the frequency of extreme weather are now statistically observable. A paper by James Hansen and colleagues (2012) observes that in continental areas the annual chance of a three standard deviation heat event (an extreme heat wave or severe drought) has increased from 1% in the 1951–1980 base period to 15% now – a 15 times increase (see Figure 2).²

Figure 2. The increase in areas with “hot events” over a period of 60 years

 

Source: Perceptions of Climate Change, Hansen et al (2012)

Forward looking models predict that the observed increase in occurrence of extreme weather events (cold as well as hot) have a high likelihood of continuing due to climate change (see Figure 3).

Figure 3. Increased Variability

Source: IPCC

¹ Exploring industry impacts of climatic variability is not the same as a theory of impacts of global warming in the sense of average temperature rise. Most farmland investments can cope with the rises in average temperatures we have seen thus far (0.7% Celsius over past 100 years). That level of warming, in itself, is manageable within the tolerances of most crop types and farming systems. However increases in climatic volatility are far more damaging than increases in average temperature. E.g. in maize crops exposed to heat but with optimal moisture will typically lose 1% of budgeted yield for each day that temperatures are over 30 degrees celcius. Moreover, days where the temperature exceeds 32°C do twice the harm of those at 31°C. And during a drought, things are worse still. Then, yields take a hit of 1.7% per day over 30°C (see Economist, March 17th, 2011, quoting research by Lobell).
² Perception of climate change. Proceedings of the National Academy of Sciences of the United States of America, 109(37), E2415–23. doi:10.1073/pnas.1205276109, http://www.epa.gov/climatechange/ghgemissions/gases.html, http://www.fao.org/docrep/017/aq191e/aq191e.pdf
³ See e.g.: IPCC, 2012: Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation and GFDRR/World Bank 2008: Disaster Risk Management and Climate Change Adaptation in Europe and Central Asia

Let us assume that a typical farm in a particular region has generated average EBITDA cash yields on total assets of 5% in a regular non-drought climatic season, 6% in a very favourable year, and 0% in a drought year.

Traditional cash flow models for such farm will estimate weighted annual average EBITDA, based on historically observed production from that farm and comparable farms in the region.

Assume the past 40 years of data for that region record a returns-eliminating drought every 6 years. This farm might then be budgeted to generate (5%*5 + 0%*1) / 6 = 4.2% annual EBITDA returns (we assume away the occasional “good” year).⁴

The above calibration to historic data is the first-order risk management step for a climate-exposed project. However some second-order adjustments are also often made. The next sections outline three additional risk adjustments.

⁴ Caution must be taken to correct information biases in farm financial yield data. These biases may in some instances arise from “disaster myopia”. Or from non-objective incentives (toward governments where farmers seek to minimise tax in their published accounts, and toward banks or other transaction counterparties where they have more panglossian tendencies). Non-accounting survey data from farmers will often estimate “average” yields solely from median years, failing to sufficiently weight frequency and payoff’s of e.g. drought years. It is desirable, in generating farmland returns data, to explicitly record physical yields from both median and abnormal seasons. And to match these pay-offs to event frequencies. A similar approach is used in hydrology; where proposed water yields (the first factor)  are “back tested” against actual historic rainfall data (a second and independent factor) in order to estimate future performance.

i Estimating volume-price feedback effects

A key complication in measuring the risk of climatic volatility is that prices may rise as a result of falling supply intersecting with largely constant demand. Consumers fill up their supermarket trolleys with roughly the same amount of calories each week, irrespective of price. Although there may be some substitution between expensive and cheap calories such income effects will not greatly knock the overall income of the farming industry. In fact there is evidence that falling supply (from e.g. a widespread drought) will often lift prices by proportionally more than the fall in quantity grown. E.g. in our August, 2012 Commentary we noted the average US farmer produced 25% less maize in 2012, an approximate 4% fall in global supply, but that this lifted global maize prices by 35%. One way to model this feed-back effect is to assume industry-wide farm revenues remain constant for each major crop-type, and then to estimate a particular farm’s participation in those revenues. Cash flow on farms with access to subsidies of various types, including crop insurance (particularly in the US where crop insurance is heavily subsidised) may also be modelled and will attenuate cash flows in poor production years. However cash flow benefits of subsidies should be discounted at a higher rate than returns from core food production (there is risk they may be scaled back). Also crop-insurance premia should be included.

Farm location and crop insurance/subsidies are not the only important inputs in modelling volume-price feedback effects. Farm quality (in particular soil moisture retention and access to irrigation) should also be included. A sophisticated model should be able to reflect how more reliable or non-climatically correlated farms may produce the volume to capture “the cream” from the high prices of drought years. Meanwhile more marginal farms (those with weaker, less moisture-retentive soils or in marginal climatic bands) should be (and are) penalised in models and by the farmland market for their greater vulnerability to weather events. Such marginal farms have an opportunity cost. They will “let you down in a drought year – just when prices are high”. In this Commentary we explore a number of reasons for penalising weaker farms with a discount rate penalty or other financial penalty over-and-above their arithmetically average agronomic assumed production. The above opportunity cost is the first of them.

ii Adjustments for production volatility

A second complexity relates to variable levels of production irrespective of price feed-backs. Volatility of production is inefficient because fixed costs such as machinery are not used as effectively in volatile systems. Farmers (and their suppliers) prefer not to hold harvesting, labour, storage, transporting, financing and processing capacity idle in order to cope with volume shifts between boom and bust years. Thus, farms with higher volatility are likely to experience average lower EBITDA per kg of produce (and higher interest charges, and lower ability to take on capital debt) as a result of regularly idle capacity.

Furthermore, farm returns are often asymmetric. The financial downside of the drought year is worse than the upside of the “very good” climatic season. This asymmetry adds risks of inconvenience and leverage to the earlier mentioned “opportunity cost” of climate exposed farms, such as those with lighter soils, higher altitude and poorer or more variable average rainfall. This is a further reason farmers (and their bankers) will typically apply an additional discount rate premium to budget cash flows of farms that have higher expected variability and asymmetry of returns.

iii Tracking errors of climatically variable farms are high

A third nuance relates to tracking error. Many investors purchase farmland as a hedge against inflation, viewing farmland as a way of creating a claim on future streams of real commodity income. Some investors view farmland as the ultimate income producing “real” (and low beta) asset. However, care needs to be taken to avoid climatic volatility undermining this inflation hedge. Poor farmland selection may result in investments that are more exposed to the production losses from climate than positive global price effects from resulting reduced supply.

For example, in extreme continental climates, droughts can span multiple years. Such prolonged droughts can significantly knock down the price of land through distressed sales, and therefore total returns, even when, because of the drought, commodity prices are high. This lack of tracking may be exhibited as low or even negative correlations of total return with commodity prices.

Investors who seek inflation protection may penalise farmland with a further risk premium if they identify a lack of correlation of total returns of a proposed investment with global food commodity prices. The most likely reason for such lack of correlation is poor relative quality of the farmland, e.g. a marginal climatic area and/or weak soils. However it is worth noting that a number of other factors (not all of them “bads”) such as bank credit cycles, government interference in markets, changes in productivity and fluctuating alternate-use (“amenity”) values of land may also reduce correlations between farmland total returns and commodity prices.

Summary of the traditional approach

The above sections reviewed, first,  traditional “climate event weighted arithmetic average returns” project models and then, second, elaborated the framework with typical adjustments for i) volume-price-feedbacks, ii) the inefficiencies and asymmetries of return resulting from climatic volatility, and iii) issues with tracking. These “traditional” approaches are centrally calibrated to past climatic patterns. It follows that they are at risk of mis-specifying risk if the frequencies of events are changing. In the next section we attempt to extend these approaches to allow for changing frequencies of drought and flood.

3. Re-calibrating agricultural risk management financial models to account for increasing climatic volatilities

Imagine if, in our stylised example, which saw a severe drought every six years, we may now encounter a drought of similar severity at twice this rate, i.e. every three years.

Other things being equal this would affect the EBITDA calculation (weighted arithmetic average returns would fall from 4.2% to 3.3%). Also, if the farm was judged more risky than average in terms of i. volume-price-feedback, ii. inefficiency-aysmmetry or iii. tracking error, then higher risk premiums might also be levied.

The above adjustment to the model simply re-weights the average by introducing a new frequency of extreme events. Since climatic volatility has recently become a systemic factor (by growing worldwide), it is likely to both reduce supply of all agricultural commodities and lift prices. It follows that forecasters may introduce a further volume-price feedback effect, which in our example would be raising prices to keep the yield at 4.2%. This is an exploratory and schematic thought and we have not implemented this in Craigmore’s models. It may be judged more conservative to recalibrate models to reflect likely falls in production and not to adjust prices. Anyway, actual project models, to incorporate these types of assumptions, will need to be multi-factor⁵ (Figure 4). Which is in itself a challenge since traditional farmland valuation models have assumed all factors of production advance into the future in their previous ratios.

⁵ It is partly in order to have better ability to specify shifts in impacts of contributing factors that Craigmore and a number of our peers are building a “Map of Agriculture” time series database of farmland input factors (as well as production and financial data) in multiple regions and crop types

Figure 4. Geo-spatial soil quality indicators are one of a number of factors necessary

Source: Map of Agriculture

Simple diversification may no longer be as effective

One implication of a systemic global increase in frequency of extreme events may be a change in optimal portfolio formation decisions. Diversification is often touted as a “free lunch” in risk and investment management circles. It is believed that, as non-correlated risks are added to a portfolio, their addition should decrease portfolio volatility while, one hopes, maintaining returns. Following this line of thought agricultural investments in disparate regions and crop types are often thought to represent risk reduction. Similarly farming families, who normally own only one farm often plan to own it for a long time, assume mean-reverting climate patterns will diversify returns over time. It remains the case, of course, that some climate risk reduction can be achieved through these geographic, crop selection, and holding period diversifications. However, the greater frequency and severity of extreme weather events introduce an element of systemic exposure to risk that will be difficult to hedge via diversification. E.g. a portfolio that systematically sought out “high return, high risk” farmland in a series of geographically disparate regions may experience more systemic exposure to deteriorating climatic conditions on those higher risk farms than would have been the case if drought frequencies were not increasing.⁶ Allocation strategies should seek out the islands of not only diverse but also stable weather patterns amidst a rising tide of climatic volatility.