Table of Contents Author Guidelines Submit a Manuscript
ISRN Applied Mathematics
Volume 2012 (2012), Article ID 251835, 16 pages
Research Article

Modeling Drought Option Contracts

1Department of Mathematics, University of British Columbia, Vancouver, BC, Canada V6T 1Z2
2Department of Mathematics, Trent University, Peterborough, ON, Canada K9J 7B8
3Department of Economics, Trent University, Peterborough, ON, Canada

Received 29 November 2011; Accepted 4 January 2012

Academic Editors: F. Hao and F. Lebon

Copyright © 2012 Jielin Zhu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.


We introduce a new financial weather derivative—a drought option contract—designed to protect agricultural producers from potential income loss due to agricultural drought. The contract is based on an index that reflects the severity of drought over a long period. By modeling temperature and precipitation, we price a hypothetical drought contract based on data from the Jinan climate station located in a dry region of China.

1. Introduction

Since the dawn of agriculture, farmers have been vulnerable to the effects of drought. These effects include not only any immediate financial losses that result from reduced crop production, but also the adverse effects that those losses may have on their operational viability going forward.

Historically, the only financial instruments available to mitigate the risks associated with drought were crop insurance contracts. Those contracts have the obvious advantage of minimizing the financial losses incurred by farmers, thus stabilizing their operations. But because the marginal benefits of farmers’ crop damage-mitigating strategies are offset by reductions in the expected value of the insurance claim, the private benefits they realize from those efforts will fall short of the social benefits. From a social perspective, the resulting moral hazard then leads to an inefficiently low level of damage-mitigating effort and an inefficiently high level of realized crop loss. Moreover, because of drought’s high positive spatial correlation, regional insurers may themselves have difficulty meeting their obligations in the event of drought, thus undermining the ability of insurance contracts to reduce the drought risks farmers face.

By the mid-1990s, however, new financial contracts known as weather derivatives had emerged as a way of reducing particular weather-related risks, including those resulting from uncertain temperature, precipitation, and frost. Because the payoffs of these derivatives depend only on weather variables over which the farmer has no control, these instruments have the advantage of sustaining the damage-mitigation incentives that crop insurance contracts may undermine, thus mitigating real crop losses [1]. Moreover, relative to an insurance market, a robust weather derivative market may better distribute correlated risks.

Weather derivatives are very similar to traditional derivatives on financial assets. But because the weather indices they depend on have no underlying measurable value [25], there is no underlying instrument with which to hedge, and the Black-Scholes sort of risk-neutral pricing models cannot be applied. As such, much of the literature on weather derivatives—including Garman et al. [6], Brody et al. [7], Richards et al. [8], Cao and Wei [4], Geman and Leonardi [9], and Taylor and Buizza [10]—has focused on the development of alternative models to price weather risk. Because the weather variables on which those models are based are likely to be imperfectly correlated with their impact on the value of the crop yield, however, a basis risk remains.

In this paper, we propose, model, and price a weather derivative, which we call a “drought option,” that is based not only on multiple weather factors (i.e., precipitation and temperature), but also on an agricultural drought index that depends on the particular crop growth being hedged as a way of reducing this basis risk.

The balance of the paper is organized as follows. In Section 2, we describe the structure of the drought option contract as well as the drought index upon which the contract depends. In Section 3, we introduce three alternative approaches to pricing the option. In Section 4, we use data from Jinan climate station in eastern China to simulate the option prices for each of our three approaches. Section 5 concludes with comparisons of the three approaches as well as a discussion of possible directions for future work.

2. Background

2.1. Drought Options

We model our derivative contract as a European put option on an agricultural drought index, 𝐼, with strike level 𝐾, and tip 𝑆, which reflects the relationship between the drought index and the loss of farmers’ income. At maturity, the payoff function is𝑓(𝐼)=𝑆max(𝐾𝐼,0).(2.1) If 𝐼—an index that takes on smaller values when a drought is more severe—is less than 𝐾, the holder will receive 𝑆(𝐾𝐼); if 𝐼 is larger than 𝐾, the holder would choose not to exercise the contract and the profit would be 0.

Because no-arbitrage pricing methods are unsuitable, we use the equilibrium approach and price the option as the present value of the expectation of its payoff [4, 11]. That expectation can be written as𝑉=𝐸𝑒𝜏0𝑟(𝑡)d𝑡𝑓(𝐼),(2.2) where 𝜏 is the contract’s duration, 𝑟(𝑡) is the risk-free interest rate, 𝐼 is the value of drought index, and is the probability density over all possible values of 𝐼.

To simplify the problem, we will assume that 𝑆=1 and that 𝑟(𝑡) is constant over the length of the contract, 𝜏. Equation (2.2) can then be rewritten as𝑉=𝑒𝑟𝜏𝐸[]max(𝐾𝐼,0).(2.3) Our problem is thus reduced to choosing an appropriate drought index, setting the strike level, and estimating the expectation.

2.2. Drought Index

Because our interest lies in developing a tool to help farmers hedge against the risks of reduced agricultural yield, our approach will utilize the Reconnaissance Drought Index (RDI) [12, 13], which measures the severity of drought in agricultural terms. For any period within year 𝑖, the RDI can be calculated asRDI𝑖=𝑚𝑗=1𝑃𝑖𝑗𝑚𝑗=1PET𝑖𝑗,𝑖=1,,𝑁,𝑗=1,,𝑚,(2.4) where 𝑃𝑖𝑗 and PET𝑖𝑗 are the precipitation and potential evapotranspiration, respectively, for the 𝑗th month of the 𝑖th year, 𝑚 is the number of observed months in year 𝑖, and 𝑁 is the number of observed years (Evapotranspiration (ET) is a measure of water movement from the Earth’s surface to the atmosphere; it is the sum of evaporation, such as that from soil and canopy, and plant transpiration. Potential Evapotranspiration (PET) is the evapotranspiration that could occur where there is sufficient water supply. Given the relationship between 𝑃 and PET, the RDI takes on smaller values when the drought is more severe, making our choice of a put option appropriate).

2.3. Potential Evapotranspiration and Actual Evapotranspiration

Potential evapotranspiration is difficult to measure because it depends on the type of plants, the type of soil, and the climatic conditions. For practical purposes, we will therefore use the actual rather than the potential evapotranspiration in our estimation of the RDI.

This gives us a new Adjusted-RDI, that can be written asRDI𝑖ad=𝑚𝑗=1𝑃𝑖𝑗𝑚𝑗=1ET𝑖𝑗,𝑖=1,,𝑁,𝑗=1,,𝑚,(2.5) where ET𝑖𝑗 is the actual evapotranspiration in the 𝑗th month of the 𝑖th year.

To estimate the ET, we will use the temperature-based methods of Blaney-Criddle [14] (Methods of estimating evapotranspiration can be divided into three categories [15]: (1) mass-transfer-based methods; (2) radiation-based methods; (3) temperature-based methods. Our use of a temperature-based method is necessitated by data limitations). The formula for the SCS Blaney-Criddle method isET=𝑘𝑐𝑘𝑡𝑓,(2.6) where𝑘𝑡=0.0311𝑇+0.24,𝑓=𝑑(1.8𝑇+32),100(2.7) and where the monthly ET measured in inches, 𝑘𝑐 is the consumptive crop coefficient for the SCS version, 𝑇 is monthly mean temperature, and 𝑑 is the percentage of daylight hours.

3. Drought Option Pricing Models

We now wish to calculate possible values of our Adjusted-RDI in order to estimate the option price. Common techniques to calculate index values include Historical Burn Analysis, Index Value Simulation, and Stochastic Simulation [11].

3.1. Historical Burn Analysis and Index Value Simulation

For Historical Burn Analysis, the basic assumption is that the historical data reasonably approximate future scenarios, allowing the price of the option to be calculated from data we already have. From (2.3), we can see that the price of a put option for drought is given by1𝑉=𝑁𝑒𝑟𝜏𝑁𝑖=1max𝐾𝐼𝑖,,0(3.1) where 𝐼𝑖 is the value of the drought index for year 𝑖 and 𝑁 is the number of years of historical data.

Our Index Value Simulation approach involves finding the best fit distribution for index values calculated using the Historical Burn Analysis approach and then sampling this distribution to produce possible future values of the index. The option price will then be calculated using (3.1).

3.2. Stochastic Simulation

Mean reversion stochastic processes are used to simulate the daily mean temperature and the speed of monthly rainfall over the entire year. The period’s rainfall and evapotranspiration are then obtained by summing the related simulated values. Finally, the option payoff is computed based on the simulation where the average discounted payoff is used as the option price.

3.2.1. Simulation of Mean Monthly Temperature

Although based on (2.6) and (2.7), only the mean monthly temperatures are needed to calculate the evapotranspiration, and those mean temperatures can be simulated directly.

Mean Reversion Process
Average daily temperature is, of course, subject to seasonal changes. For one specific day in each year, however, we will assume that the temperature fluctuates around some mean value, which means we can choose a mean reversion stochastic process to simulate the behavior of daily temperature [2, 16].
We use the following Stochastic Differential Equation (SDE) to model daily temperature:𝑑𝑇𝑡=𝑑𝑇𝑚𝑡𝑇+𝑎𝑚𝑡𝑇𝑡𝑑𝑡+𝜎𝑡𝑑𝑊𝑡.(3.2) The solution is 𝑇𝑡=𝑇𝑠𝑇𝑚𝑠𝑒𝑎(𝑡𝑠)+𝑇𝑚𝑡+𝑡𝑠𝑒𝑎(𝑡𝜏)𝜎𝜏𝑑𝑊𝜏,(3.3) where 𝑇𝑡 is the temperature at time 𝑡, 𝑇𝑚𝑡 is the mean daily temperature at time 𝑡, 𝑎 reflects the speed of the mean reversion, 𝜎𝑡 reflects the variation from the mean value at time 𝑡, and 𝑊𝜏 is a Wiener process.

Parameter Estimation
𝑇𝑚𝑡 in (3.2) represents simulated mean daily temperature. If the mean temperature is assumed to be a continuous variable then it is possible to simulate this curve by a sinusoidal function of the form: sin(𝜔𝑡+𝜑),(3.4) where 𝑡 denotes the time, measured in days.
To account for local warming trends observed in the data, 𝑇𝑚𝑡 is modeled as having the form:𝑇𝑚𝑡=𝐴+𝐵𝑡+𝐶sin(𝜔𝑡+𝜑).(3.5) Expanding (3.5) gives us 𝑇𝑚𝑡=𝑎1+𝑎2𝑡+𝑎3sin(𝜔𝑡)+𝑎4cos(𝜔𝑡).(3.6)
We then estimate the parameters in (3.6) by using the Gauss-Newton algorithm to solvemin𝜉𝑇𝑡𝑋2,(3.7) where 𝑋 is the data vector and 𝜉 is the parameter vector (𝑎1,𝑎2,𝑎3,𝑎4).
Estimates of 𝜉 then also allows us to calculate the parameters in (3.5) as follows:𝐴=𝑎1,𝐵=𝑎2,𝐶=𝑎23+𝑎24,𝑎𝜑=arctan4𝑎3𝜋.(3.8)
The term 𝜎𝑡𝑑𝑊𝑡 in (3.2) reflects the discrepancy of real temperature from the simulated mean temperature. We simplify the function 𝜎𝑡 as a piecewise constant function. Here we assume the value of 𝜎𝑡 is a constant number for each month.
As we see, 𝜎𝑡 reflects the variation of daily temperature, the first estimator is based on the quadratic variation of 𝑇𝑡: 𝜎2𝜇=1𝑁𝜇𝑁𝜇1𝑗=0𝑇𝑗+1𝑇𝑗2,(3.9) where 𝑁𝜇 is the number of days in month 𝜇.
If we discretize (3.2), for a given month 𝜇, we have𝑇𝑗=𝑇𝑚𝑗𝑇𝑚𝑗1+𝑎𝑇𝑚𝑗1+(1𝑎)𝑇𝑗1+𝜎𝜇𝜖𝑗1,(3.10) with 𝑗=1,,𝑁𝜇, where {𝜖𝑗}𝑁𝜇1𝑗=1 follow the standard normal distribution. If we let 𝑇𝑗 be 𝑇𝑗(𝑇𝑚𝑗𝑇𝑚𝑗1), then (3.10) becomes 𝑇𝑗=𝑎𝑇𝑚𝑗1+(1𝑎)𝑇𝑗1+𝜎𝜇𝜖𝑗1,(3.11) which we can see as a regression of today’s temperature on yesterday’s temperature. Therefore, the second estimator of 𝜎𝜇 is 𝜎2𝜇=1𝑁𝜇2𝑁𝜇𝑗=2𝑇𝑗̂𝑎𝑇𝑚𝑗1(1̂𝑎)𝑇𝑗12,(3.12) where ̂𝑎 is an estimator of 𝑎.
To estimate the mean reversion parameter 𝑎, we use the martingale estimation functions method. After collecting observations during 𝑛 days, an efficient estimator ̂𝑎𝑛 of 𝑎 can be obtained by solving the equation𝐺𝑛̂𝑎𝑛=0,(3.13) where 𝐺𝑛(𝑎)=𝑛𝑖=1̇𝑏𝑇𝑖1;𝑎𝜎2𝑖1𝑇𝑖𝑇𝐸𝑖𝑇𝑖1,(3.14) and where ̇𝑏(𝑇𝑖1;𝑎) denotes the derivative with respect to 𝑎 of the term 𝑏𝑇𝑡=;𝑎𝑑𝑇𝑚𝑡𝑇𝑑𝑡+𝑎𝑚𝑡𝑇𝑡.(3.15) Therefore, we have ̇𝑏𝑇𝑖1;𝑎=𝑇𝑚𝑡𝑇𝑡.(3.16)
To then solve (3.14), all we need is the formula 𝑇𝑖𝐸[𝑇𝑖𝑇𝑖1]. As we know the solution of (3.2) is (3.3), and the expectation of the integration part is 0, when 𝑡=𝑖 and 𝑠=𝑖1, we have𝐸𝑇𝑖𝑇𝑖1=𝑇𝑖1𝑇𝑚𝑖1𝑒𝑎+𝑇𝑚𝑖.(3.17) So we have 𝐺𝑛(𝑎)=𝑛𝑖=1𝑇𝑚𝑖1𝑇𝑖1𝜎2𝑖1𝑇𝑖𝑇𝑖1𝑇𝑚𝑖1𝑒𝑎𝑇𝑚𝑖.(3.18) It is easy to estimate the parameter 𝑎 from ̂𝑎𝑛=log𝑛1𝑖=1𝑌𝑖1𝑇𝑖𝑇𝑚𝑖𝑛𝑖=1𝑌𝑖1𝑇𝑖1𝑇𝑚𝑖1,(3.19) where 𝑌𝑖1𝑇𝑚𝑖1𝑇𝑖1𝜎2𝑖1,𝑖=1,2,𝑛.(3.20) Based on (3.15), we can see that ̂𝑎𝑛 is the unique estimation.

Numerical Solution
Because it is difficult to use the integral solution in (3.3), we employ the implicit Milstein numerical method [17, 18]. The problem is then reduced to the following iterative procedure:𝑇(0)𝑛+1=𝑇𝑛+𝑇𝑚𝑛𝑇+𝑎𝑚𝑛𝑇𝑛Δ𝑡+𝜎𝑘𝑇Δ𝑊,(1)𝑛+1=𝑇𝑛+𝑇𝑚𝑛+1𝑇+𝑎𝑚𝑛+1𝑇(0)𝑛+1Δ𝑡+𝜎𝑘Δ𝑊,(3.21) where 𝑇(1)𝑛+1 is the simulated point we get, 𝜎𝑘 is the value of diffusion parameter depending on the 𝑛th point located in month 𝑘.

3.2.2. Simulation of the Speed of Monthly Rain

Transforming Monthly Rain to a Continuous Process
While precipitation does not behave continuously, the speed of precipitation can be assumed to change continuously in time. When rain starts, the speed of precipitation increases continuously from zero to its peak and decrease continuously to zero where it remains at zero until it rains again. In this formulation, the total amount of precipitation can be obtained by integrating the speed of precipitation over the period of interest, as given by the formula: 𝑃𝑡1,𝑡2=𝑡2𝑡1𝑥(𝑡)𝑑𝑡,(3.22) where 𝑥(𝑡) represents the speed of rainfall at time 𝑡 and 𝑃𝑡1,𝑡2 represents the amount of precipitation during the period from time 𝑡1 to time 𝑡2.
The monthly rainfall, then, is essentially the monthly mean speed of precipitation with units (mm/m2·month). In order to simulate monthly rainfall, it will be sufficient to simulate the speed of rainfall using a mean reversion process.

Mean Reversion Process
In the previous model of monthly temperature, diffusion was modeled as a piecewise function 𝜎𝑡, where 𝜎𝑡 was assumed constant for each month. In order to make the simulation of the speed of monthly rain more realistic, the diffusion can be modeled as a function of 𝑋𝑡 according to the following SDE:𝑑𝑋𝑡𝜃=𝑑𝜃(𝑡)+𝑎(𝑡)𝑋𝑡𝑑𝑡+𝜎𝑋𝑝𝑡𝑑𝑊𝑡,(3.23) with 𝑡0, 𝑎0, and 𝜃(𝑡)>0. As described in (3.2), 𝜃(𝑡) is the simulated mean speed of monthly rain at time 𝑡 over a year and 𝑎 is the mean reversion parameter. Here 𝜎 and 𝑝 together reflect the volatility which is now dependent on previous precipitation.

Parameter Estimation
The data we will utilize are for the Jinan climate station in eastern China, obtained from the China Meteorological Data Sharing Service System ( for the period from January 1st, 1951 to December 31st, 2006.
Figure 1 depicts the mean monthly rain speed over the 56-year period. As in the temperature model, the real mean speed of monthly precipitation will be assumed to fit a sinusoidal function of the form [19]:𝜃(𝑡)=𝑚+𝛼sin2𝜋(𝑡𝜐)12,(3.24) where 𝑚 is the mean of the sine curve, 𝛼 determines the oscillation, and 𝜐 is the horizontal shift. To make the simulated curve closer to the historical experience, the simulation can be further improved by expanding 𝜃 in terms of a Fourier series [19]: 𝜃(𝑡)=𝑚+𝑛𝑖=0𝛼𝑖sin(2𝑖+1)𝜋(𝑡𝜐)6.(3.25)
The parameters in 𝜃(𝑡) can then be estimated using the Gauss-Newton method to solve the least squares problem. To do this, we choose 𝑛=1.
We now wish to find an unbiased estimator of the mean reversion parameter 𝑎. The original estimation of 𝑎 is [19]: 1̂𝑎=𝑛𝑛1𝑖=0𝑋𝑖+1𝑋𝑖𝜃(𝑖+1)+𝜃(𝑖)𝜃(𝑖)𝑋𝑖Δ,(3.26) where Δ is the time increment. Because this estimation would change due to different lengths of data, we have the modification: 𝑎𝑏=1||𝐼𝑏||𝑖𝐼𝑏𝑋𝑖+1𝑋𝑖𝜃(𝑖+1)+𝜃(𝑖)𝜃(𝑖)𝑋𝑖Δ,(3.27) with 𝐼𝑏={𝑖=1,,𝑛|𝜃(𝑖)𝑋𝑖|>𝑏}, 𝑏 a positive real number. As 𝑏 gets larger, 𝑎𝑏 is convergent. Here we choose 𝑏=50 for a suitable estimation.
The diffusion parameters can be estimated by first squaring equation (3.23) to get𝑑𝑋𝑡2=𝑎2𝜃𝑋𝑡2𝑑𝑡𝑑𝑡+𝑎𝜃𝑋𝑡𝜎𝑋𝑝𝑡𝑑𝑡𝑑𝑊𝑡.(3.28) Then, based on Itô’s integration rule, we have ln𝑑𝑋𝑡2𝑋=2ln(𝜎)+ln(𝑑𝑡)+2𝑝ln𝑡,(3.29) such that ln(𝑑𝑋𝑡)2 and ln(𝑑𝑋𝑡) have a linear relationship. The data can now be substituted into (3.29). We use linear regression method to estimate 𝜎 and 𝑝.

Figure 1: Mean speed of monthly rain over 56 years.

Numerical Solution
After the parameter estimation, based on the implicit Milstein method, the monthly precipitation can be simulated by the following: 𝑋(0)𝑛+1=𝑋𝑛+𝜃(𝑛)+𝑎𝜃(𝑛)𝑋𝑛(0)Δ𝑡𝑛+𝜎𝑋𝑛𝑝(1)Δ𝑊𝑛+12𝜎2𝑝𝑋𝑛(1)2𝑝1Δ𝑊2𝑛Δ𝑡𝑛,𝑋(1)𝑛+1=𝑋𝑛+𝜃(𝑛+1)+𝑎𝜃(𝑛+1)𝑋(0)𝑛+1Δ𝑡𝑛+𝜎𝑋𝑛𝑝(1)Δ𝑊𝑛+12𝜎2𝑝𝑋𝑛(1)2𝑝1Δ𝑊2𝑛Δ𝑡𝑛,(3.30) where 𝑋(1)𝑛+1 is the simulated point at the (𝑛+1)th step, 𝑛=0,,𝑁 and 𝑁 is the step size for one simulation process.

3.3. Correction of the Model
3.3.1. Correction for the Mean Curve

For both temperature and precipitation models, the distance between the average of historical records and the simulated mean curve can also be simulated with another sinusoidal function like (3.25), with the exception that the period of this function may be different depending on the climate conditions of the location.

Figures 2 and 3 compare the simulated mean curve with and without the simulation of distance. It is clear that the inclusion of simulated distance provides results that are much closer to the historical record than the original one.

Figure 2: Simulated daily temperature comparison.
Figure 3: Simulated mean speed of rain comparison.
3.3.2. Maintaining the Positivity of the Rain Speed

Since 𝑋𝑡 represents the speed function, it must be nonnegative to have a meaningful interpretation. Moreover, according to (3.23), negative 𝑋𝑡 leads to no value for 𝑋𝑡2𝑝1. However, there is no guarantee that the simulation of (3.23) yields positive values, especially when 𝑋𝑡 is close to 0 at the beginning of the year [17]. If we examine Figure 3, we see that the mean speed of rain is under the time axis at the beginning of the year.

If 𝑋𝑡 represents a stochastic process with𝑋Prob𝑡>0𝑡=1,(3.31) the stochastic integration scheme possesses an eternal life time if𝑋Prob𝑛+1>0𝑋𝑛>0=1.(3.32) Otherwise, we say it has a finite lifetime.

If we can find a numerical method to solve this model which has an eternal life, as long as the initial value of historical precipitation is positive, we can make sure that all the following simulated points are positive. This is another reason to choose the implicit Milstein method to do the simulation [19].

For the mean reversion process given by𝑑𝑋𝑡=𝛼𝛽𝑋𝑡𝑑𝑡+𝜎𝑋𝑝𝑡𝑑𝑊,(3.33) with 𝛼,𝛽,𝜎,𝑝𝑅+ and 𝑝>1/2, the Milstein method provides numerical positivity with the following restriction:1Δ𝑡<𝜎2.(3.34)

Compared to (3.23), (3.33) will guarantee positivity with𝛼=𝑎𝜃(𝑡)+𝜃(𝑡)𝑅+.(3.35)

We now use the daily data from Jinan station with 𝑛=1. As Figure 4 shows, the graph of 𝜃(𝑡) for the whole year based on (3.35), the simulation of the mean monthly precipitation does not stay positive due to the value 𝛼 which is not always positive. Moreover, for the model whose diffusion part has term 𝑋𝑡, once a negative 𝑋𝑡 value is obtained, according to (3.30), the simulated value for the next step could never be calculated since 𝑝<1/2, which would lead to 2𝑝1<0 and therefore no value for 𝑋𝑛2𝑝1.

Figure 4: Simulated speed of monthly rain in Jinan.

Therefore, in order to ensure positivity, 𝜃(𝑡) must be modeled appropriately. Since 𝜃(𝑡) is a continuous sinusoidal function and the speed of mean reversion 𝑎 is assumed constant, there should be a boundary for the value of 𝛼 in (3.35) that ensures 𝛼>0. Thus we require,𝜃(𝑡)>𝜃(𝑡)𝑎,𝑡=𝑡0,,𝑡𝑁.(3.36) If we only add a constant on the right hand of (3.25) to make sure the new 𝜃(𝑡) satisfies (3.35), we still have the same 𝜃(𝑡). Therefore we must find max𝑡=𝑡0,,𝑡𝑁(𝜃(𝑡)/𝑎), denoted by 𝑀, so that the adjusted simulated mean speed of precipitation can be expressed as 𝜃ad(𝑡)=𝜃(𝑡)+𝑀+1, and 𝜃ad(𝑡)=𝜃(𝑡).

Continuing with our use of the Jinan data as an example, the results are shown in Figure 5 and the corrector for 𝜃(𝑡) is −40.977.

Figure 5: Adjusted simulated speed of monthly Rain in Jinan.

To use the new 𝜃 in the algorithm, note that all the simulated points, 𝑋𝑡, have mean value 𝜃ad(𝑡). This change could also affect the diffusion part. In order to keep the process similar to the original one, we modify the algorithm as follows.(1)𝐸=max(𝑋𝑛(1)𝑀1,𝑣), where 𝑣 is a uniform random variable in (0, 0.1). If we use 0 to replace 𝑣, the diffusion part could be 0 if 2𝑝1>0 and infinity if 2𝑝1<0.(2)Change (3.30) to 𝑋(0)𝑛+1=𝑋𝑛+𝜃(𝑛)+𝑎𝜃(𝑛)𝑋𝑛(1)Δ𝑡𝑛+𝜎𝐸𝑝Δ𝑊𝑛+12𝜎2𝑝𝐸2𝑝1Δ𝑊2𝑛Δ𝑡𝑛,𝑋(1)𝑛+1=𝑋𝑛+𝜃(𝑛+1)+𝑎𝜃(𝑛+1)𝑋(0)𝑛+1Δ𝑡𝑛+𝜎𝐸𝑝Δ𝑊𝑛+12𝜎2𝑝𝐸2𝑝1Δ𝑊2𝑛Δ𝑡𝑛.(3.37)(3)After calculating all the simulated points, we need to change the whole list back down to the original level: 𝑋𝑛=𝑋𝑛(1)𝑀1.(3.38) Note that if 𝑝<1/2, then positivity is not yet guaranteed for a large size simulation.

4. Analysis of Results

Using temperature and precipitation data from the Jinan climate station, and the curve of consumptive crop coefficient based on McGuinness and Bordne [14], we price a drought put option contract based on three different within-year data periods: (1) January to December (Table 1); April to August (Table 2); April to June (Table 3). In each case, we set the strike level at 𝐾=0.7 and use each of our three methods. Each of Tables 1 through 3 shows the option prices based on data from four different historical spans: 1951–2006, 1967–2006, 1977–2006, and 1987–2006.

Table 1: Option price comparison for months 1–12.
Table 2: Option price comparison for months 4–8.
Table 3: Option price comparison for months 4–6.

Comparing the option prices for the full 56-year span, we see that the results from the Historical Burn Analysis and the Index Value approaches are very similar to one another, while the Stochastic Simulation method yields different results. This difference is attributable to the inaccuracy of the simulation of the speed of monthly rain. The Stochastic Simulation method does, however, yield results that have the lowest variance across the four columns.

Looking at the 1987–2006 results in the last column in each of Tables 13, we see that, for each method, the results are quite different from those obtained using the full 56 years of data. This points to the overall sensitivity of pricing to the historical span of time upon which the pricing is based.

5. Future Work

To simplify our analysis, we have ignored a number of financial factors. Future work could incorporate the relationship between a drought index and farmers’ profits, the possibility of changing interest rates, and the potential market price of risk [20].

In addition, a number of improvements could be made to the weather models we have used to price drought contracts. In particular, the price of the drought option clearly depends on the joint distribution of the temperatures and the precipitation at maturity. To simplify our analysis, we have simulated these two variables independently. In future work, a joint model could be developed. We have also not considered the location limitations of the climate models. In particular, the climate stations which collect precipitation data are typically located in large urban areas, far from the rural areas where farming is concentrated. Because the levels of precipitation might be quite different between even neighboring urban and rural areas, future work should incorporate this spatial basis risk.


M. Pollanen and K. Abdella are partially supported by NSERC Canada.


  1. T. J. Gronberg and W. S. Neilson, Incentive under Weather Derivatives vs. Crop Insurance,, Texas A and M University, 2007.
  2. P. Alaton, B. Djehiche, and D. Stillberger, “On modeling and pricing weather derivatives,” Applied Mathematical Finance, vol. 9, pp. 1–20, 2002. View at Google Scholar
  3. M. Mraoua and D. Bari, Temperature Stochastic Modeling and Weather Derivative Pricing, Empirical Study with Moroccan Data, Institute for Economic Research, OCP Group, Casablanca, Morocco, 2005.
  4. M. Cao and J. Wei, “Weather derivatives valuation and market price of weather risk,” Journal of Futures Markets, vol. 24, no. 11, pp. 1065–1089, 2004. View at Publisher · View at Google Scholar · View at Scopus
  5. S. Yoo, Weather Derivatives and Seasonal Forecast, Cornell University, Ithaca, NY, USA, 2003.
  6. M. Garman, C. Blanco, and R. Erikson, “Weather derivative modeling and valuation: a statistical perspective,” in Climate Risk and the Weather Market: Financial Risk Management with Weather Hedges, 2002. View at Google Scholar
  7. D. Brody, J. Syroka, and M. Zervos, “Dynamical pricing of weather derivatives,” Quantitative Finance, vol. 2, pp. 189–198, 2002. View at Google Scholar
  8. T. J. Richards, M. R. Manfredo, and D. R. Sanders, “Pricing weather derivatives,” American Journal of Agricultural Economics, vol. 86, no. 4, pp. 1005–1017, 2004. View at Publisher · View at Google Scholar · View at Scopus
  9. H. Geman and M. Leonardi, “Alternative approaches to weather derivatives pricing,” Managerial Finance, vol. 31, pp. 46–72, 2005. View at Google Scholar
  10. J. W. Taylor and R. Buizza, “A comparison of temperature density forecasts from GARCH and atmospheric models,” Journal of Forecasting, vol. 23, no. 5, pp. 337–355, 2004. View at Publisher · View at Google Scholar · View at Scopus
  11. O. Mußhoff, M. Odening, and W. Xu, “Modeling and hedging rain risk,” in Proceedings of the Annual Meeting of the American Agricultural Economics Association, Long Beach, Calif, USA, 2006.
  12. G. Tsakiris and H. Vangelis, “Establishing a drought index incorporating evapotranspiration,” European Water, no. 9-10, pp. 3–11, 2005. View at Google Scholar
  13. G. Tsakiris, D. Pangalou, and H. Vangelis, “Regional drought assessment based on the Reconnaissance Drought Index (RDI),” Water Resources Management, vol. 21, no. 5, pp. 821–833, 2007. View at Publisher · View at Google Scholar · View at Scopus
  14. J. L. McGuinness and E. F. Bordne, “A comparison of lysimeter-derived potential evapotran-spiration with computed value,” USDA Technical Bulletins 1452, 1972. View at Google Scholar
  15. C. Y. Xu and V. P. Singh, “Cross comparison of empirical equations for calculating potential evapotranspiration with data from Switzerland,” Water Resources Management, vol. 16, no. 3, pp. 197–219, 2002. View at Publisher · View at Google Scholar · View at Scopus
  16. F. E. Benth and J. S. Benth, “The volatility of temperature and pricing of weather derivatives,” Quantitative Finance, vol. 7, no. 5, pp. 553–561, 2007. View at Publisher · View at Google Scholar · View at Zentralblatt MATH · View at Scopus
  17. C. Kahl, Positive numerical integration of stochastic differential equations, Diploma Thesis, University of Wuppertal, Wuppertal, Germany, 2004.
  18. S. Han, Numerical solution of stochastical differential equations, M.S. thesis, University of Edinburgh and Heriot-Walt University, Scotland, UK, 2005.
  19. C. van Emmerich, Modelling and simulation of rain derivatives, M.S. thesis, University of Wuppertal, Wuppertal, Germany, 2005.
  20. M. Cao, A. Li, and J. Wei, “Precipitation modeling and contract valuation: a frontier in weather derivatives,” The Journal of Alternative Investments, vol. 7, pp. 93–99, 2004. View at Google Scholar