Printer Friendly

Managing weather risks: the case of J. League soccer teams in Japan.


Weather-related risk exposures, such as varying temperature and precipitation, are important concerns for companies worldwide. According to the Chicago Mercantile Exchange (CME) Group, (1) one-third of businesses around the world are affected by adverse weather conditions. In particular, 80 percent of U.S. firms and 75 percent of Japanese firms have been exposed to weather-related risks to different extents (Myers, 2008; Yokoyama, 2014). Weather risks are of particular interest to the leisure and entertainment industries, such as ski resorts with snow risks (Brockett, Wang, and Yang, 2005), golf courses and beaches with precipitation risks (Smith, 1993; Leggio, 2007), and as we will focus on in our study, the sports industry with an assortment of different weather risks. Some initial evidence has been provided in the literature on varying effects of adverse weather for sports teams in different types of sports and geographic areas (e.g., Bird, 1982; Bruggink and Eaton, 1996; DeSchriver, 2007).

In most regions of the world, weather risks have been managed more or less by specialized insurance policies or other government-sponsored financial mechanisms (IPCC, 2012). In the United States, these risks are now primarily managed using weather derivative products since they became available on the CME in 1997. The trading volume on CME was $9.4 billion from April 2010 to March 2011 with a large number of trades (466,000 in total) (Weather Risk Management Association [WRMA], Annual Industry Survey, 2011). Although there has been a drop from the high of over $40 billion from April 2005 to March 2006, it is still of paramount significance. Temperature contracts account for the great majority of CME contracts and less than 5 percent are precipitation contracts. Temperature-based weather derivatives are primarily based on indices of heating degree days (HDD) and cooling degrees days (CDD) and are traded for 24 U.S. cities, 6 Canadian cities, 11 European cities, 3 Australian cities, and 3 Japanese cities. Precipitation-based contracts are available for March to October in the form of monthly contracts or seasonal contracts (2 to 8 consecutive months in length). Both futures and options contracts are available and they are based on a precipitation index reflecting the cumulative precipitation amount (measured in inches) for the desired time period. Futures contracts pay $500 times the respective index value, with a tick size of 0.1 inch. Option contracts have exercise prices ranging from 0 to 20 inches for monthly contracts and 0 to 60 inch. for seasonal contracts. The indices are currently available only for 10 U.S. cities and are not available for other countries. In fact, most of the CME contracts are used for weather events in different regions within North America and less than 5 percent are used for regions outside of North America and Europe.

According to WRMA annual industry surveys since 2001, the growing over-thecounter (OTC) market plays an active role in complementing the CME market. Of the 14 surveyed companies (WRMA members in various industries such as banking, insurance, and energy), a total of 998 OTC contracts were written from April 2010 to March 2011, a 160 percent increase from the previous year. These contracts are large with over $2 million notional value on average and in total, represent a volume of $2.5 billion in 2010/2011. This represents the highest total notional value since 2004/2005. Note that the percentage of OTC contracts with counterparties that do not participate in the survey has been consistently high over time (79 percent in the 2011 survey). In contrast to the CME market, precipitation contracts account for at least 20 percent of the OTC contracts and regions outside Europe and North America have a much better representation (over 10 percent in 2011 and over 20 percent in 2010).

Despite the increasing trend of contracts covering precipitation events and Asian markets in the recent years, the weather derivatives market is still greatly dominated by temperature contracts covering North American events. Consequently, existing academic research has focused on temperature-based weather risks and the U.S. market, with few exceptions (e.g., Martin, Barnett, and Coble, 2001; Skees et al., 2001; Leggio, 2007). (2) In this article, we fill in a gap by providing specific evidence on using precipitation-based weather derivatives for managing risk exposures in an international marketplace. Using data from the premier Japanese soccer league, we examine its financial susceptibility to precipitation related risk, propose a weather hedging mechanism for the league, and examine its value.

The Japanese sports industry, in particular soccer, is large and growing. Established in 1993 with only 10 teams in total, J. League now comprises two divisions, the premier division J1 and the lower division J2, with 38 teams in total. With a 2010 attendance of 7,928,976 (J1: 5,638,894; J2: 2,290,082), the total revenue for J. League in 2010 was US$880 million (J1: US$665 million; J2: US$215 million), or an average of US$38 million for J1 teams and US$11 million for J2 teams. Our analysis is mainly based on data from the J1 teams because of their popularity and financial significance.

Adverse weather conditions were not taken into consideration when most soccer stadiums were built in Japan. Of the 49 stadiums that hosted a J. League game in 2011, 9 (18.4 percent) have roofs covering the entire stadium, 18 (36.7 percent) have partial roofs, and 22 (44.9 percent) have no roofs at all. Since J. League never suspends a game because of rain, spectators must watch the game in the rain, and might even have to sit directly on the wet grass as some stadiums do not have seats. (3) Although soccer games are scheduled well in advance and teams are contractually bound to play rain or shine, spectators can and do rely on weather forecasts when making plans to attend a game. This exposes soccer teams to significant financial risks in ticket sales and related revenues that are their primary sources of income. Left unmanaged, inclement weather can be a substantial contributing factor to the average operating loss of US$500,000 for J1 teams in 2010. This volatile financial performance might ultimately threaten continuing corporate sponsorship, which is key to the survival of sports teams worldwide.

Despite significant weather risks, very few teams have taken effective measures to manage these risks. Among the 38 J. League teams, only 3 teams (Teams Shimizu, C Osaka, and Hiroshima) purchased precipitation-based insurance contracts after it was first introduced in Japan in 2002 (Nihon Keizai Shimbun, 2002a, 2002b). However, according to personal interviews conducted by the authors in 2011 with J. League managers, four out of seven showed interest in weather risk management. The interviews also revealed concerns about the currently available risk management methods that are expensive and time consuming. Teams can either build a new stadium with a roof or retrofit an existing stadium, or purchase an insurance policy with high loadings. (4) These findings call for better risk management options designed based on empirical evidence that documents and measures the financial impact of adverse weather. This article examines the use of weather derivatives as a potential risk management solution for the Japanese Soccer League teams.

Using two matched data sets composed of J1 teams' game attendance and corresponding weather conditions between 1993 and 2010, we are able to develop quantified measures of the teams' precipitation-related weather exposures, design a hedging mechanism that caters to the quantified exposure, and carry out subsequent analyses to evaluate this risk management option for the teams. In particular, we use the Wang transform model (Wang, 2000) to arrive at valuations that address the incompleteness of this market. We specifically account for the risk preferences of the decision makers through a survey-based measure. Our regression and simulation analysis results suggest that the J. League soccer teams have a significant precipitation exposure. We also find that the proposed hedging mechanism is valuable based on the estimated managerial risk preferences, even after taking into account transaction costs and premium loadings associated with the derivatives.

This article contributes to the growing literature on capital market solutions to managing physical risks, such as financial product innovations as new or alternative approaches in catastrophic risk, and longevity/mortality risk management. In particular, weather risk products in an incomplete capital market post additional challenges in designing, pricing, and evaluating different risk management solutions. This is one of the first papers in the peer-reviewed literature, to our knowledge, that provides a complete analysis of the impact of hedging weather risk and its financial consequences for a specific potential weather derivative user in a well-defined industry. By focusing on precipitation risk and a less explored industry with a significant risk exposure, we complement the existing literature that considers primarily temperature derivative contracts for the energy and agriculture industries. Although our analysis is done using data from the Japanese soccer industry, it provides insights for the sports industries around the world. Just like in most of these cases, corporate sponsorship plays a major role in J. League clubs' financial viability. Properly managing weather risks and hence reducing the likelihood of unfavorable financial outcomes and the uncertainties therein will help cultivate continued support from the corporate sponsors. In addition, the proposed design and evaluation of the hedging instrument from a managerial perspective are more generally applicable to businesses in various sports and entertainment industries for their weather risk exposures.

The rest of the article is organized as follows. The "Literature Review" section discusses the related literature. The "Quantifying Precipitation Risk Exposure for J. League Teams" section provides a comprehensive set of empirical analyses that document and estimate the weather exposures of J. League teams. In the "Simulation Analyses for the Impact of Precipitation on Financial Performance" section we use simulations to calibrate the impact of the weather exposures on the teams' financial performance based on empirically estimated parameters. In the "Weather Risk Hedging and Its Contribution to the Corporate Value" section we evaluate the contribution of a proposed weather derivative hedging instrument to a team's corporate value using the Wang transform model with survey-based risk aversion parameters. The "Conclusion" section concludes the article.


Starting with Noll (1974), a plethora of papers focus on estimating the demand function for the sports industry, a nice review of which is provided by Borland and MacDonald (2003). Different factors are found to be significant contributors to the demand function, including the winning rate of the team (Noll, 1974; Davis, 2008), the presence of super stars (DeSchriver, 2007) and terrorist risks (e.g., Toohey and Taylor, 2008; Kalist, 2010), and the competitiveness of the games (Coates and Humphreys, 2010). Management of these risk factors is also extensively studied in the literature (see, e.g., Fuller and Drawer, 2004; Lee, Farley, and Kwon, 2010).

However, there have been only limited studies of the effect of weather conditions on the demand for sports teams or its risk management, even though the impact is suspected to be significant and unpredictable. In addition, these studies provide mixed findings. For example, major sport events in the United States such as Major League Baseball (MLB) (Bruggink and Eaton, 1996; Butler, 2002) or National Football League (NFL) games (Welki and Zlatoper, 1999) have been significantly affected by excessive precipitation and extreme temperature. However, U.S. soccer teams do not seem to be affected by weather conditions (DeSchriver, 2007). Significant weather effect was not found for European sports teams in general (Bird, 1982; Hynds and Smith, 1994; Carmichael, Millington, and Simmons, 1999), except for cricket whose attendance is negatively affected by the rain (Schofield, 1983; Morley and Thomas, 2007). In particular, Kawai and Hirata (2008) estimate, for the first time, the demand function for J. League soccer games and report that the teams experience significantly lower attendance when it rains. Their study used one dummy variable based on precipitation observed at the stadium at the beginning of the game, and used data from 1993 to 2005. Using data from 1993 to 2010, we design and examine a comprehensive set of precipitation variables to thoroughly capture the effect weather has on attendance rates and, moreover, measure its impact on the teams' financial outcomes. More importantly, we further investigate how a hedging mechanism can stabilize cash flows, thereby improving the team value.

There has been only limited research on the design and effectiveness of weather derivatives and most of the studies focus on the temperature exposure in the United States. Vedenov and Barnett (2004) examine the effectiveness of weather derivatives in hedging temperature and precipitation risks affecting agriculture yields in the United States. They use semivariance, value-at-risk, and certainty-equivalent measures to evaluate the reduction of crop yield variations. Golden, Wang, and Yang (2007) and Golden, Yang, and Zou focus on temperature-related weather exposure and examine hedging effectiveness of CME-traded weather derivatives in the U.S. power industry using variance as the measure, while Yang, Li, and Wen study that for three European countries. Perez-Gonzalez and Yun (2013) show a significant and positive relationship between the temperature-based weather exposure of energy firms and the adoption of weather derivatives when they became available in 1997. They also provide evidence that the use of weather derivatives improves firm value. Leggio (2007) and Martin, Barnett, and Coble (2001) study the design of precipitation derivatives for golf courses and the agricultural business in the United States, respectively. Skees et al. (2001) is one of the few papers that propose precipitation risk management in the international market-place. It studies an index-based drought insurance product for cereal producers in Morocco and examines the impact of basis risk. Most of these studies do not focus on assessing the value of weather hedging and most times do not address market incompleteness. The popularity of soccer games in Asian and European countries and the lack of evidence in the international markets led us to take advantage of our unique data sets and focus on the much less explored precipitation risk and its management.

Because weather is not a traded asset, we need to address the incompleteness of the market when evaluating the hedging mechanism. Various valuation approaches have been proposed, including the Esscher transform (Gerber and Shiu, 1994, 1995), the Wang transform (Wang, 2000, 2002), the relative entropy optimization (McLeish and Reesor, 2003), and the indifference pricing method (Brockett et al., 2009). See Brockett, Wang, and Yang (2005) for an overview of these approaches. In this article, we adopt the Wang transform (Wang, 2000, 2002) to evaluate the contribution of a proposed weather hedging mechanism to the corporate value of J. League soccer teams. The Wang Transform provides a universal framework for pricing contingent claims. It has clear economic interpretations that capture the risk preference of the decision-maker without imposing a functional form for the utility. (5) Furthermore, it can recover the Black-Scholes formula and the capital asset pricing model (CAPM) under appropriate assumptions (Wang, 2000). The Wang transform has been applied to many contingent claim pricing scenarios (e.g., Chen and Cox, 2009; Godin, Mayoral, and Morales, 2009), empirical estimation of the risk aversion coefficient (Jones and Zitikis, 2007), and derivation of optimal economic capital for insurers (Hurlimann, 2004). The Wang transform has also been extended to multivariate risk settings (Wang, 2007; Kijima and Muromachi, 2008).

Using the Wang transform model, we calculate the expected cash flows under the distorted probability measure considering the risk preferences of the managers. One challenge is to estimate the risk aversion parameter. We follow the recent literature to use a set of lottery-based questionnaires distributed to the J. League managers to obtain this estimate (see, e.g., Pennings and Smidts, 2000; Donkers, Melenberg, and Soest, 2001 ; Holt and Laury, 2002; Andersen et al., 2008). We discuss the design of our questionnaires and summarize the results in the "Weather Risk Hedging and Its Contribution to the Corporate Value" section.


Taking advantage of the available data sources, we investigate the weather risk exposures of J. League soccer teams by empirically examining the impact of weather conditions on game attendance. Game attendance is closely related to direct ticket sales and other indirect revenue sources based on popularity (e.g., concessions and other sales in the stadium, advertisement and commercial revenues, etc.). These constitute a significant portion of revenue for the teams. (6) We run a set of regressions for each team using game-by-game J. League attendance data matched with corresponding weather data from 1993 to 2010. Our main interest is in precipitation since preliminary analyses suggest that excessive precipitation is the most plausible weather concern for low attendance. We explore two different types of precipitation measures in our analysis. The first type uses the raw value of the precipitation amount and the second one is a dummy variable with different threshold values of raw precipitation.

Data Description and Preliminary Data Processing

We obtain the game attendance data from J. League and focus only on J1 teams, which have much higher attendance and revenues than J2 teams. If a team is relegated from J1 to J2 in our sample period, we exclude this team from the analysis in the subsequent years. We include the team again if it returns to J1 and, similarly, include the corresponding data points if a new team is upgraded to J1. In our analysis, we focus on teams whose main stadium is "nonroofed," that is, the roof covers no greater than 50 percent of the spectator area, as they are likely to have a higher weather exposure. (7) A team's main stadium is defined by the J. League at the beginning of the 2011 season. Most teams have only one main stadium with the exception of Team Yokohama FM has two (Nippa stadium and Nissan stadium). As such, our attendance data set includes 2,082 games out of a total of 3,984 J1 games during the sample period, 1993-2010. We run time-series regressions for each unique combination of a team and its main stadium, with the exception of Team Yokohama FM, for which we run regressions for two unique combinations.

We obtain corresponding precipitation data from the Japan Metrological Agency (JMA) ( In the JMA data set, precipitation is recorded hourly as "-", "0.0 (mm)," "0.5 (mm)," and hereafter increasing in 0.5-mm increments. In particular, "-" means no rainfall was observed whereas "0.0" means rainfall was observed for less than the measurable amount (i.e., at least 0.5 mm to be measurable). (8) We design a set of level measures based on the raw precipitation amounts. Specifically, we calculate the cumulative amount of precipitation around the game time for the following durations: 1, 2, 3, 4, 6, and 24 hours.

Additionally, we define a set of dummy measures from the raw precipitation amounts to account for any nonlinear effect on game attendance; that is, individuals' willingness to go to a game can be based on a certain cutoff point rather than monotonically related to the amount of rain. Lacking any previous evidence for threshold choices, we have constructed the dummy measures of precipitation in a largely exploratory manner. We set the threshold from 0 to 5 mm for all variables with 0.5 mm increments. We set additional thresholds from 5.5 to 15 mm in 1-mm increments for the 6-hour and above precipitation variables, and additional thresholds of 20, 30, 40, and 50 mm for the 24-hour precipitation variables to accommodate the higher amount of rain anticipated for longer periods. The dummy variable takes value 1 if the amount of precipitation is at least the threshold level, and 0 otherwise. As a result, we have a total of 354 precipitation variables, including both the level measures and the dummy measures. The detailed definitions of these variables are presented in the Appendix.

Empirical Calibration of Precipitation Exposures

We run a set of time-series generalized least squares (GLS) regressions to calibrate precipitation exposure for the J1 teams and to facilitate further team-based simulation analyses and evaluations. In addition to the precipitation measure, we include the following control variables from the sports risk management literature. Distance measures the distance that needs to be traveled by the supporters of the away team and is proxied by the natural log of the train ticket price in 2012. The longer the distance, the less attendance is expected (Garcia and Rodriguez, 2002). Natural log of the Population of the city where the team is located is expected to be positively correlated with game attendance (Hart, Hutton, and Sharot, 1975; Kalist, 2010). To proxy for team popularity, we count the total number of star players, stars_both, in both the home and the away teams. A player is identified as a "star" if he represented the Japanese National Soccer Team in the previous year. (9) We also include lagged average game attendance for the home team (Forrest, Simmons, and Szymanski, 2004; Kawai and Hirata, 2008), h_latt, and that for the away team (Kawai and Hirata, 2008), a_latt.

In addition, we control for a set of dummy variables that can affect game attendance. Rival equals 1 if the game is between two teams from the same prefecture, and 0 otherwise. Like football or baseball in the United States, there is likely an increased interest for games between two rival teams that are often defined by the proximate locations. Opening refers to the opening game of the season, which can generate the greatest interest and hence the highest attendance. Top3 equals 1 if either team was ranked among the top three in the previous season, and 0 otherwise, as more fans will be following the leading teams. Similarly, because fans can be especially concerned if their teams are running the risk of being downgraded to the J2 League, we also include a dummy variable, Worst3, that equals 1 if either team was ranked within the last three in the previous year, and 0 otherwise. Note that we rank as the lowest those teams just promoted to the J1 League from the J2 League this year, and thus variable Worst3 also captures any particular interest in these newly promoted teams. Weekday indicates a game on a working day since fewer spectators are likely to be at the game (Garcia and Rodriguez, 2002; Kawai and Hirata, 2008). Lastly, we also include an April dummy and an October dummy as attendance tends to be higher at the beginning and the end of the soccer season (Kawai and Hirata, 2008). Our multivariate GLS regression model is specified as


where [Y.sub.i,n,t] is the natural log of game attendance and [R.sub.i,j,n,t] is jth precipitation measure, j = 1,..., 354, for a unique combination i of a team and its home stadium's game n in soccer season t. (10) While we have run regressions using all 354 precipitation variables, we only present in Table 1 the results using the intuitive dummy measures d0_h3 and d4_h24, taking value 1 if it rains during the 3-hour period right before the game or if there is at least 4-mm precipitation on the game day, respectively. (11) Note that to conserve space, Table 1 only shows the coefficients for the precipitation dummies and the model fit statistics, while full regression results are available in the accompanying online appendix. (12)

As shown in Table 1, precipitation is a significant contributing factor to game attendance for most team/stadium combinations after controlling for other relevant factors. For example, our analysis shows that the Kyoto, Saikyogoku combination is heavily affected by adverse weather conditions. If there is at least 4 mm of rain on the game day, attendance is expected to be 20.5 percent less, other things being equal. Even for a less affected combination, such as Team Yokohama FM in the Nippa Stadium, the effect is still significant. Our results show that their attendance is 9.9 percent less when there is a minimum amount of precipitation during the 3-hour period before the game.

Table 1 also shows that the precipitation effect differs in the magnitude for different teams and stadiums. These variations may be due in part to the joint effect of popularity of the team and the capacity of the stadium. (13) Overall we find an adverse weather effect to different extents for 14 out of 15 unique combinations of teams and stadiums, illustrating that weather risk is an imperative concern for the J. League. (14)

In the next section, we use the team-by-team time-series regression results (Table 1) to further calibrate the impact of precipitation on a team's financial performance and to evaluate the proposed weather hedging mechanism for a J. League team. (15)


Design of the Simulation Analyses

Because detailed financial data are not available for the J. League teams for a sufficiently long sample period, (16) we use Monte Carlo simulations to gauge the effect of weather on teams' financial performance. More specifically, we simulate game-day precipitation using parameters estimated from historical weather data and obtain simulated attendance using the precipitation exposures measured in the regression analysis. We then use aggregate financial data to calibrate cash flow sensitivity to attendance and generate team cash flows from simulated attendance. In our simulation analysis, we focus on the intuitive dummy measure of game-day precipitation with the 4-mm threshold value (d4_h24); that is, the precipitation variable takes value 1 if the cumulative precipitation on the game day exceeds 4 mm, and 0 otherwise. This also allows us to make a comparison with the precipitation insurance product offered in Japan designed based on game-day cumulative precipitation. (17)

We illustrate our simulation analyses using a representative team, Kawasaki F, and their home stadium, Todoroki, located in Kawasaki City (population of 1.4 million) in the greater Tokyo area. Team Kawasaki F was established in 1997 and joined the J. League in 1999. The Todoroki stadium has 50 percent roof coverage and a capacity of 25,000 spectators. With an operating profit of 47 million JPY (US$573,000), Team Kawasaki F ranks in the middle among all J. League teams. (18) Additionally, all of its home games are held in the Todoroki Stadium, ensuring a more precise analysis than the teams using multiple home stadiums.

The adverse weather impact on Kawasaki F's game attendance is significant: both precipitation measures reported are highly significant at the 1 percent level. The game-day dummy measure with a 4-mm threshold value (d4_h24) contributes to a 21.4 percent reduction in game attendance in the (unreported) univariate analysis. In the multivariate specification, the adverse impact is 20.1 percent.

Game Day Precipitation Simulation and Attendance Estimation

We assume that the occurrence of a rainy game day follows a Bernoulli distribution and use Monte Carlo simulations to estimate the number of rainy day games and non-rainy day games for a soccer season. We estimate the parameter of the probability of a rainy game day from the empirical distribution of precipitation obtained from historical weather data observed at the Yokohama Observatory (the closest to the Todoroki Stadium) from 1968 to 2010. Because the soccer season spans 10 months of the year and there is significant heterogeneity in precipitation during these months, we break the months into different "seasons" before estimating the probability of precipitation. We define five seasons so that within a season there is no significant difference between the months in the likelihood of a rainy game day, whereas between seasons the differences are statistically and economically significant. The combinations are: Season 1: March, April, May, and October; Season 2: June and September; Season 3: July; Season 4: August and November; and Season 5: December. January and February are excluded because there are no games during this time. Table 2 summarizes the mean probability of a rainy game day for each season and the pairwise differences between seasons. We can see that the differences between seasons are significant both economically and statistically.

Using the probability parameters shown in Table 2, we generate for each season one Bernoulli variable for a rainy game day. According to the 2012 game schedule of Kawasaki F, eight home games are held in Season 1, four in Season 2, one in Season 3, and four in Season 4. No game id held in Season 5 in 2012. With a further assumption that precipitation measures between different games are independent, we can simulate the weather conditions for each of the 17 games in a soccer season. Ten thousand simulation paths are run to obtain the weather conditions. Table 3 presents the simulated probabilities of having n rainy game days in a soccer season. The most probable outcome is having 2 rainy days, with a probability of 0.2702. The probabilities of having 10 or more rainy days are very small and are assumed to be zero in further analysis.

We can then use the regression parameter estimates, the simulated game day precipitation data, and the actual values for the control variables from year 2012 to obtain game attendance for a soccer season. (19) We re-estimated the precipitationattendance regression for our prototype team Kawasaki F with a shorter sample period (2007-2010) to be consistent with the available financial data needed for further analysis. The regression results are qualitatively similar to those presented in Table 1. (20) In addition to using the point estimate (-0.143) for the precipitation variable (d4_h24), we also consider its 90 percent confidence interval (-0.226, -0.058) for robustness.

Team Net Cash Flows Calibration

Building on the estimated game attendance, we calculate the net cash flows (NCFs) of the team for each year. We first estimate the sensitivity of revenues and costs to annual game attendance using OLS regressions from 2007 to 2010. (21) The aggregate-level financial data are obtained from the team's financial statements available at the J. League Web site ( (22) The regression models we use are

ln(revenuet) = [[alpha].sub.revenue] + [[beta].sub.revenue][Attend.sub.t] + [[??].sub.revenue], (1)

ln([cost.sub.t]) = [[alpha].sub.cost] + [[beta].sub.cost][Attend.sub.t] + [[??].sub.cost], (2)

where [revenue.sub.t] is the annual revenue of Kawasaki F in year t, [cost.sub.t] is annual total cost of Kawasaki F in year t, [Attend.sub.t] is the total attendance for Kawasaki F in year t, and [[??].sub.revenue] and [[??].sub.cost] are residuals in each regression model. The estimation results are presented in Table 4.

We calculate team NCFs as

[NCF.sub.Without_hedge] = revenue - cost, (3)

and use the calibrated attendance and parameter estimates from the revenue and cost regressions (Table 4) to obtain simulated cash flows. (23) Recall that we calibrated attendance using the point estimate of the precipitation exposure from the regression analysis, along with the upper and lower bound of the 90 percent confidence interval of the estimate. Descriptive statistics for simulated NCFs are reported in Table 5.

Table 5 demonstrates the significant effect of weather conditions on the variations of NCFs of Team Kawasaki F. Kawasaki F's net income ranges from US$73,000 to US$293,000. Given a standard deviation of US$51,260 (Table 5), it is likely that Kawasaki F can suffer from negative net income taking into account losses from precipitation and possibly other losses and costs/expenses. The losses can be even more severe if the actual impact of adverse weather is higher than the estimated value, such as the estimated numbers using the lower bound of the confidence interval as Table 5 shows.


Based on the simulated NCFs, we propose a hedging mechanism and evaluate its potential contribution to the team's corporate value. We employ the Wang transform for the evaluation to mitigate problems associated with the incomplete market. An estimated range of values for the decision maker's risk aversion parameter is obtained through a survey of J. League managers.

The Wang Transform for the Evaluation

Overview of the Wang Transform. The Wang transform (Wang, 2000, 2002) was developed as a universal framework for asset pricing and can be used in an incomplete market setting pertaining to the weather risk scenarios discussed in this article. Incomplete capital markets, in which such products exist, make the pricing of such capital market products difficult and risk-neutral measures nonunique. Instead of having to estimate the risk-adjusted discount rate, a rather difficult task often required by other valuation models, the Wang transform incorporates the decision maker's risk preference into the valuation by calculating the expected value with a transformed set of probabilities. In particular, decision makers' risk preferences can still be correctly represented when the expected cash flow under the physical probability measure is zero, whereas under the classic discounted cash flow models, zero expected cash flow uniformly leads to a present value of zero, rendering the choice of the discount rate irrelevant.

The Wang transform is formulated as follows:

[F.sub.Q] (x) = [PSI]([[PSI].sup.-1](FP(x)) + [lambda]), (4)

where [F.sup.Q](x) is the cumulative distribution function (CDF) under the risk-neutral probability measure Q, and [F.sup.P](x) is the CDF under the physical probability measure P. As in other option pricing methods, such as the Black-Scholes model, the riskneutral measure stipulates valuing contingent claims as if the decisionmaker were risk neutral and thus the risk-free rate is the appropriate discount rate. [PSI](*) is the CDF of the standard Normal distribution and [[PSI].sup.-1](*) is its inverse function. They are used for transforming the physical probabilities to the risk-neutral probabilities given the risk aversion parameter [lambda].

If the parameter [lambda] is positive, the physical CDF is shifted to the left. It implies that the subjective probabilities of the good scenarios (e.g., higher cash flows) are lower, indicating that the decision maker is risk averse. Similarly, a negative [lambda] indicates that the decision maker is risk seeking and assigns a higher probability to the good scenarios. If x follows the Normal distribution, [lambda] becomes the market price of risk, or the Sharpe ratio, (E(r) - [r.sub.f])/[sigma].

The economic interpretations of the Wang transform are closely related to the Esscher transform and both approaches are widely used in the insurance and finance literature. These approaches have been applied to pricing in a variety of areas including the securitization of longevity and mortality risks (Cox, Lin, and Wang, 2006; Denuit, Devolder, and Goderniaux, 2007; Chen, Zhang, and Zhao, 2010), annuity (Lin, Tan, and Yang, 2009), mortgage (Chen, Cox, and Wang, 2010), weather derivatives (Moridaira, 2010), and options and other derivatives in general (Gerber and Shiu, 1994; Gerber and Landry, 1997; Vyncke et al., 2003; Monfort and Pegoraro, 2012). Connections between the Esscher and the Wang transforms are discussed in detail by Labuschagne and Offwood (2010). In particular, by using a negative exponential or a power utility function with an additional assumption that underlying asset returns follow a Normal distribution, we can recover the Esscher transform from the Wang transform with identical risk aversion parameters. (24)

A Hedging Mechanism for Precipitation Risk. We propose a simple hedging mechanism for precipitation risk with coverage period for a soccer season. As described previously, team cash flows are calibrated from simulated game-day weather conditions and their estimated weather exposures. We then construct the expected payoff of the hedging instrument with an estimated range of reasonable risk loadings. (25)

We continue to use Team Kawasaki F as an illustrative example. For simplicity, we construct the payoff as a complete hedge that covers 100 percent of the expected loss due to precipitation based on the point estimate of precipitation exposure. (26) This payoff structure will result in incomplete hedge if cash flow realizations are estimated based on the upper and lower bounds of the point estimate. The payoff is triggered if a game is on a rainy day.

The expected loss per game from precipitation is calibrated by an OLS regression using the simulated NCFs and the number of rainy day games per soccer season:

Simulated NCF = [[alpha].sub.rainyday] + [[beta].sub.rainyday] Simulated Number of Rainy Day Games Per Season, (5)

where [[alpha].sub.rainyday] captures expected NCF if there is no rainy-day game in a soccer season and [[beta].sub.rainyday] measures the expected cash flow differential, or expected loss, per rainy-day game in a season. The parameter estimates are [[alpha].sub.rainyday] = US$805,218 and [[beta].sub.rainyday] = -US$34,702. Under the assumption of a complete hedge, the payoff of the hedging instrument per rainy game is thus -[[beta].sub.rainyday] = US$34,702. Consequently we obtain the expected pay off from weather hedging per soccer season as

[Payoff From Hedging Per Season] = [N.summation over (n = 1)] [p.sub.n] x n x Payoff From Hedging Per Game = US$88, 783, (6)

where [p.sub.n] is the probability of having n rainy days in a soccer season. The simulated probabilities [p.sub.n] are found in Table 3.

Because precipitation-based weather derivative products are not currently traded in the Japanese market and thus price information is not available, we assume that the embedded premium is calculated as (27)

Premium of the Hedging Instrument = (1 + Safety Loading) x Expected Payoff. (7)

We investigate a range of safety loadings from 0 to 20 percent for the proposed hedging instrument. This choice of range is consistent with the literature and the market data. Previous literature assumes a loading of 0 to 20 percent in a weather derivative contract. For instance, Golden, Wang, and Yang (2007) assume that weather forward prices are actuarially fair. Vedenov and Barnett (2004) assume that a safety loading is 0, 5 percent, or 10 percent. According to information provided by MDA Weather Services (a wholly owned subsidiary of MacDonald, Dettwiler and Associates Ltd., Richmond, Canada.), a U.S. company providing comprehensive data sets and consulting services related to weather risks, the temperature option premium traded in the United States is between 10 percent and 20 percent (Climetrix, 2010). Chinacarini (2011) finds a 5 percent safety loading using CDD (i.e., cooling degree days) forward contracts for Atlanta, Chicago, Cincinnati, Dallas, and New York from 1999 to 2004. (28)

In addition, we benchmark our safety loading assumption against that of the weather insurance contracts, an alternative mechanism to manage weather risks. Weather insurance contracts are used by three J. clubs: Teams Shimizu, Hiroshima, and C Osaka. (29)

Assuming again that the occurrence of a rainy-day game follows the Bernoulli distribution, we can back out the implied safety loadings in the insurance contracts. We find that the implied safety loadings for these contracts are at least 60 percent. (30) In the subsequent evaluation analysis, we add an insurance-implied loading of 60 percent to the range of loadings from 0 to 20 percent.

We can now calculate the NCFs when a hedging instrument is adopted as

[NCF.sub.with_hedge] = [NCF.sub.without_hedge] - premium + payoff,

where the payoff and the premium of the hedging instrument are defined in Equations (6) and (7) above.

Table 6 presents descriptive statistics of simulated cash flows with hedging for team Kawasaki F, with two alternative assumptions of safety loading at 10 percent and 20 percent, respectively. We present these for each of the three cases of cash flow calibrations before hedging, based on the point estimate of cash flow sensitivity to precipitation and its lower and upper bounds, respectively. For comparison purposes, we also include again the simulated cash flows without hedging from Table 5. Intuitively, in the point estimate-based and the lower bound estimate-based cases, the standard deviation of cash flows is dramatically reduced when hedging is used and the ranges of probable cash flows are also much smaller, whereas the expected values of cash flows are reduced for the cost of hedging. Interestingly for the upper bound estimate-based case, the standard deviations are larger when hedging is used. This is because of the overhedging scenario when the cash flow sensitivity to precipitation is actually lower than expected, upon which the payoff is based.

Evaluation of the Hedging Instrument. Within the framework of the Wang transform, we gauge the value of the hedging instrument for precipitation by calculating the expected NCF differentials under the risk-neutral probability measure. We apply the Wang transform (4) to obtain q-probabilities from the p-probabilities for NCFs with and without hedging as previously described. We discuss in the next section the determination of an appropriate risk aversion parameter [lambda] for the team manager in order to obtain the risk neutral q-probabilities. The expected NCF when the soccer team does not use a hedging mechanism is:

[E.sup.Q] ([NCF.sub.without_hedge]) = [I.summation over (i = 1)][NCF.sub.without_hedge,i],[f.sup.Q]([NCF.sub.without_hedge,i])

[f.sup.Q] ([NCF.sub.without_hedge,i]) = [F.sup.Q]([NCF.sub.without_hedge,i]) - [F.sup.Q]([NCF.sub.without_hedge,i-1]),

where i indicates the ith smallest NCF among the simulation paths.

Similarly, we evaluate the expected NCF when the manager chooses to adopt such a hedging mechanism. The expected NCF with hedging is derived as

[E.sup.Q]([NCF.sub.with_edge]) = [I.summation over(i = 1)] [NCF.sub.with_hedge,i][f.sup.Q] ([NCF.sub.with_hedge,i-1]) [f.sup.Q] ([NCF.sub.with_hedge,i]) = [F.sup.Q] ([NCF.sub.with_hedge,i]) - [F.sup.Q] ([NCF.sub.with_hedge,i-1]).

Finally, we obtain the value of the hedging mechanism by calculating the firm value differential under the risk-neutral q-probability measure, that is,

V = [E.sup.Q] ([NCF.sub.with_hedge]) - [E.sup.Q] ([NCF.sub.withou_hedge]). (8)

This method of evaluation not only takes into account a decision maker's risk preference, but controls for her wealth level, a factor that tends to influence the risk preference.

Estimation of the Risk Aversion Parameter [lambda]

To obtain a range of risk aversion coefficient [lambda] appropriate for the J. League team managers, we follow the economics literature to obtain a direct measure by creating and distributing four sets of questionnaires that ask interviewees to choose from paired hypothetical lotteries. We adopt the methodology proposed in the recent literature (Pennings and Smidts, 2000; Andersen et al., 2008) to design our questionnaires. This approach is easier to implement than estimating the risk preference in a natural experiment setting (Kahneman and Tversky, 1979).

Table 7 summarizes the questionnaires distributed to 10 current J. League team managers. (31) The questionnaires are designed based on data from an average J1 club, Iwata. Iwata's revenue in 2010 ranked 10th among 18 J1 clubs and is the closest to the average revenue of all J1 clubs in the 2010 fiscal year. Panels A and B reflect the manager's preference between two risky scenarios where one is relatively risker than the other. Panel A includes questions regarding attendance while Panel B includes questions regarding dollar amounts. Panels C and D reflect the manager's preference between a risk-free scenario and a risky scenario. Again, Panel C includes attendance-related questions and Panel D includes dollar-amount-related questions.

In Panels A and B, Scenario A is relatively risker than Scenario B. For Panel A (B), the higher value in Scenario A is calculated as the mean attendance (revenue) of Iwata in 1994-2010 plus two standard deviations. Similarly, the lower value in Scenario A equals the mean attendance (revenue) of Iwata in 1994-2010 minus two standard deviations of attendance (revenue). The higher value in Scenario B is calculated as the mean attendance (revenue) of Iwata in 1994-2010 plus one standard deviation in Panel A (B). Similarly, the lower value in Scenario B equals the mean attendance (revenue) of Iwata in 1994-2010 minus one standard deviation. In Panels C and D, we set Scenario B's attendance so that the implied interval of [lambda] is the same as in Panels A and B for each question. The actual distributed questionnaires do not include expected values, their differences, or the implied intervals of [lambda] (all in italics in Table 7).

We asked the J. League managers to evaluate the pair of scenarios in each question given the probabilities and the outcomes (in the number of attendance or monetary values), and indicate their preferred scenario for each pair. By noting the question at which their choice of preferred scenario switches from Scenario B (less risky) to Scenario A (more risky), we can obtain the implied interval of her risk aversion parameter l. For example, if the manager indicates that she prefers Scenario B to A for Q1-Q5 and then switches to preferring A to B for Q6-Q9, her implied [lambda] ranges from 0 to 0.25. We collected 7 out of 10 sets of questionnaires that were distributed and the results are summarized in Table 8.

We calculate the most plausible range of parameter [lambda] for each manager as follows. First, we obtain the range of [lambda] implied from each set of questions in the questionnaires as presented in Panels A-D of Table 7. Second, for each team manager, we determine the most frequently represented range of [lambda] in the four sets. For example, if the implied interval of estimated [lambda] is 0.25-0.52 in Set 1 (Panel A), 0.52-0.84 in Set 2 (Panel B), 0.25-0.52 in Set 3 (Panel C), and 0-0.25 in Set 4 (Panel D), the most frequently represented range of implied [lambda] is 0.25-0.52. In Table 8, we note that Manager 4 has inconsistent risk preferences because instead of switching from Scenario B to A, his preferred scenario changes from A to B in all panels. Lastly, we find the most plausible range of risk aversion parameter [lambda] for an average J. club manager to be 0.25-0.52 since this particular range appears the most times (four out of seven) among all managers' estimated ranges.

Contribution of the Weather Hedging Instrument to Team Value

Building on the empirically calibrated precipitation exposure, simulated NCFs, estimated safety loading, and estimated risk aversion parameter, we evaluate the hedging instrument as a risk management option for our prototype Team Kawasaki F according to (8). To calculate the expected values under the Q-measure, we use the simulated cash flows in Table 6 for the three cases presented (the point estimate, the upper bound estimate, and the lower bond estimate), and transform the physical P-measure into the risk-neutral Q-measure with a set of risk aversion coefficient [lambda] ranging from 0.25 to 0.52. Note that using a Monte Carlo simulation with 10,000 iterations, the physical CDF is obtained by simply assuming equal probabilities of each simulation path, that is, 1/10,000. Results are shown in Table 9.

Table 9 shows the extent to which the team's corporate value increases with a hedging instrument as the manager's risk aversion increases. Panel A presents our baseline case built on the point estimate of precipitation exposure. When the safety loading for the hedging instrument is 10 percent, the manager will find risk management valuable as long as her risk aversion coefficient [lambda] is at least 0.2, thus including our entire estimated range of risk aversion for J. League managers. Even when hedging is more costly with a 20 percent safety loading, it will be valuable to managers with a [lambda] of 0.4 or higher, still within the range of our estimated most plausible risk aversion parameters. (32) Intuitively, when the impact of precipitation is higher than expected, proxied by the lower bound of the 90 percent confidence interval of the point estimate, the value of the hedging instrument and the acceptable safety loadings are even higher. The results are presented in Panels C and D. Interestingly, in the case when cash flow sensitivity is much lower than expected, proxied by the upper bound case (Panels B and D), managers have to incur an unnecessarily high cost to overhedge, or end up speculating rather than hedging. Therefore, the value of weather derivatives is negative for the risk-averse managers. The more risk-seeking managers will also find the hedging instrument valuable because it provides them with induced uncertainties due to overhedging. (33)

Our results show that the empirically calibrated risk management instrument, combined with a reasonable risk premium in the marketplace, can be employed to help J. League clubs improve their corporate value. When the safety loading becomes prohibitively large at 60 percent as is estimated from the precipitation insurance product in Japan, the J. League managers cannot easily opt for managing this significant risk exposure. This is perhaps why only three teams in the J. League have purchased weather insurance despite the wide-spread concern about this risk exposure. Our results also suggest the need for establishing a weather derivatives market in Japan (and other countries) to facilitate effective and feasible weather risk management for the sports and many other industries.


Weather risks are suspected to have significant impact on the sports industry. They not only entail lower attendance and hence reduce team profits, but also threaten the retention of corporate sponsorships that are critical to the teams' survival and success. In the recent economic recession, many sports club sponsors have pulled their support due to the uncertain and sometimes unfavorable financial outcomes of these sports teams. Our empirical analysis documents significant weather exposures for sports teams. Although many other factors can contribute to the fluctuations in revenues, weather risk, specifically precipitation, is one that can be affordably managed with a hedging instrument as illustrated in this article. Our analysis and results provide insights for the international markets in general in understanding the impact of precipitation risk, assessing its financial consequences, and evaluating weather derivatives as a risk management option from a managerial perspective. Our results also point to the need for introducing a weather derivatives market in Japan and other countries to allow the use of such risk management options.

To the best of our knowledge, this is among the first papers to calibrate the weather exposure of the sports industry and analyze the value of a hedging instrument with risk preference estimates obtained from surveying actual managers while addressing the incompleteness of the market. With detailed attendance data from Japan's premier soccer league, J. League, and matching weather data, we calibrate the precipitation exposure of J. League teams through a comprehensive set of multivariate GLS regressions. We then use simulation analysis to obtain the cash flows for a prototype J. club and propose a simple hedging instrument. Based on survey results, we explore different managerial risk preferences and show that team value can be significantly increased by hedging. In fact, with a reasonable range of safety loadings (less than 10 percent), hedging always contributes positively to team value for the estimated risk preferences of J. League managers.

In this article, we have focused on precipitation risk for Japanese soccer teams to take advantage of our detailed data sets. Future research can look into these issues in other business applications and other international markets. It will be interesting to examine the joint impact of multiple weather effects (such as temperature and precipitation) and design effective risk management strategies taking into account their dependence. We can also relax the assumption of a complete hedge to determine the optimal hedge ratio given the premium and the risk aversion. Overall, sports teams have many other risks affecting the entertainment value of the team, such as the risk of player injuries and delegation risks. These, along with the weather risk, can all be incorporated into an enterprise risk management framework in future studies.


We define a set of level and dummy measures of precipitation for the regression analysis to examine its impact on game attendance. The level measures are defined based on the raw cumulative precipitation amount and the dummy measures are defined using different threshold values. For the level measures, we examine six different durations: 1, 2, 3, 4, 6, and 24 hours. The 1-hour variables comprise each of the 9 hours before the game, each of the 2 hours during the game, and the 1 hour after the game, resulting in a total of 12 different measures. The 2-hour precipitation variable refers only to the cumulative amount of precipitation during the game period. The 3-hour precipitation variables comprise the 2 hours during the game plus the 1 hour before or after the game, the 3 hours immediately before the game, the 3-hour period beginning 6 hours before the game, and the 3-hour period beginning 9 hours before the game. The 4-hour and the 6-hour precipitation variable refer only to the 4 and 6 hours prior to the game period, respectively. Finally, the 24-hour precipitation variables include the cumulative precipitation on the game day, on the day before the game day, on the day 2 days before the game day, and on the day 3 days before the game day.

Lacking any previous evidence for threshold choices, we have constructed the dummy measures of precipitation in a largely exploratory manner. We set the threshold values from 0 to 5 mm for all variables with 0.5-mm increments. We set additional thresholds from 5.5-mm to 15 mm in 1-mm increments for the 6- and 24hour precipitation variables, and additional thresholds of 20 mm, 30 mm, 40 mm, and 50 mm for 24-hour precipitation variables to address the higher amount of rain anticipated for longer periods. The dummy variable takes value 1 if the amount of precipitation is at least the threshold amount, and 0 otherwise. In total, we have 354 level and dummy measures of precipitation. The definitions are summarized in Table A1.
Table A1
Definitions of Level and Dummy Measures of Precipitation
                                                            Number of
Length of                     Level     Threshold for       Variables
Period      Timing            Measure   Dummy Measures      Defined

1 hour      Each of the 9     Yes       0.0-5.0 mm in       144
            hours before                0.5 mm increments
            the game, Each
            of the 2 hours
            during the
            game, the 1
            hour after the
2 hours     The 2-hour game   Yes       0.0-5.0 mm in       12
            period                      0.5-mmin crements
3 hours     The 1 hour        Yes       0.0-5.0 mm in       60
            before the game             0.5-mm increments
            plus 2-hour
            game period
            The 1 hour        Yes       0.0-5.0 mm in
            after the game              0.5-mm increments
            plus 2-hour
            game period
            The 3 hours       Yes       0.0-5.0 mm in
            before the game             0.5-mm increments
            From 6 hours      Yes       0.0-5.0 mm in
            before the game             0.5-mm increments
            to 3 hours
            before the game
            From 9 hours      Yes       0.0-5.0 mm in
            before the game             0.5-mm increments
            to 6 hours
            before the game
4 hours     The 4 hours       Yes       0.0-5.0 mm in       12
            before the game             0.5-mm increments
6 hours     The 6 hours       Yes       0.0-5.0 mm in       22
            before the game             0.5-mm increments
                                        6.0-15.0 mm in
                                        1.0-mm increments
24 hours    On the game day   Yes       0.0-5.0 mm in       104
            1, 2, and 3                 0.5-mm increments
            days before the
            game day                    5.0-15.0 mm in
                                        1-mm increments
                                        20.0-50.0 mm in
                                        10-mm increments
                                        Total:              354

DOI: 10.1111/jori.12071


Andersen, S., G. W. Harrison, M. I. Lau, and E. E. Rutstrom, 2008, Eliciting Risk and Time Preferences, Econometrica, 76(3): 583-618.

Bird, P. J. W., 1982, The Demand for League Football, Applied Economics, 14(6): 637-649.

Borland, J., and R. MacDonald, 2003, Demand for Sport, Oxford Review of Economic Policy, 19(4): 478-502.

Brockett, P. L., L. L. Golden, M.-M. Wen, and C. C. Yang, 2009, Pricing Weather Derivatives Using the Indifference Pricing Approach, North American Actuarial Journal, 13(3): 303-315.

Brockett, P. L., M. Wang, and C. Yang, 2005, Weather Derivatives and Weather Risk Management, Risk Management and Insurance Review, 8(1): 127-140.

Bruggink, T. H., and J. W. Eaton, 1996, Rebuilding Attendance in Major League Baseball: The Demand for Individual Games, in: J. Fizel, E. Gustafson, and L. Hadley, eds., Baseball Economics: Current Research (Westport, CT: Praeger), pp. 9-31.

Butler, M. R., 2002, Interleague Play and Baseball Attendance, Journal of Sports Economics, 3(4): 320-334.

Carmichael, F., J. Millington, and R. Simmons, 1999, Elasticity of Demand for Rugby League Attendance and the Impact of BskyB, Applied Economics Letters, 6(12): 797800.

Chen, B., L. Zhang, and L. Zhao, 2010, On the Robustness of Longevity Risk Pricing, Insurance: Mathematics and Economics, 47(3): 358-373.

Chen, G., M. C. Roberts, and C. S. Thraen, 2006, Managing Dairy Profit Risk Using Weather Derivatives, Journal of Agricultural and Resource Economics, 31(3): 653-666.

Chen, H., and S. H. Cox, 2009, Modeling Mortality With Jumps: Applications to Mortality Securitization, Journal of Risk and Insurance, 76(3): 727-751.

Chen, H., S. H. Cox, and S. S. Wang, 2010, Is the Home Equity Conversion Mortgage in the United States Sustainable? Evidence From Pricing Mortgage Insurance Premiums and Non-Recourse Provisions Using the Conditional Esscher Transform, Insurance: Mathematics and Economics, 46(2): 371-384.

Chinacarini, L., 2011, No Chills or Burns From Temperature Surprises: An Empirical Analysis of the Weather Derivatives Market, Journal of Futures Markets, 31(1): 1-33.

Climetrix, 2010,

Coates, D., and B. R. Humphreys, 2010, Week to Week Attendance and Competitive Balance in the National Football League, International Journal of Sport Finance, 5(4): 239-252.

Cox, S. H., Y. Lin, and S. Wang, 2006, Multivariate Exponential Tilting and Pricing Implications for Mortality Securitization, Journal of Risk Insurance, 73(4): 719-736.

Davis, M. C., 2008, The Interaction Between Baseball Attendance and Winning Percentage: A VAR Analysis, International Journal of Sport Finance, 3(1): 58-73.

Denuit, M., P. Devolder, and A.-C. Goderniaux, 2007, Securitization of Longevity Risk: Pricing Survivor Bonds With Wang Transform in the Lee-Carter Framework. Journal of Risk and Insurance, 74(1): 87-113.

DeSchriver, T. D., 2007, Much Adieu About Freddy: Freddy Adu and Attendance in Major League Soccer, Journal of Sport Management, 21(3): 438-451.

Donkers, B., B. Melenberg, and A. V. Soest, 2001, Estimating Risk Attitudes Using Lotteries: A Large Sample Approach, Journal of Risk and Uncertainty, 22(2): 165-195.

Forrest, D., R. Simmons, and S. Szymanski, 2004, Broadcasting, Attendance and the Inefficiency of Cartels, Review of Industrial Organization, 24: 243-265.

Fuller, C., and S. Drawer, 2004, The Application of Risk Management in Sport, Sports Med, 34(6): 349-356.

Garcia, J., and P. Rodriguez, 2002, The Determinants of Football Match Attendance Revisited: Empirical Evidence From the Spanish Football League, Journal of Sports Economics, 3(1): 18-38.

Gerber, H. U., and B. Landry, 1997, Skewness and Stock Option Prices, North American Actuarial Journal, 1(3): 50-58.

Gerber, H. U., and E. S. W. Shiu, 1994, Option Pricing by Esscher Transforms, Transactions of the Society of Actuaries, 46: 99-191.

Gerber, H. U., and E. S. W. Shiu, 1995, An Actuarial Bridge to Option Pricing, Securitization of Insurance Risk Risk: The 1995 Bowles Symposium, M-FI97-1: 45-62.

Godin, F., S. Mayoral, and M. Morales, 2009, Contingent Claim Pricing Using a Normal Inverse Gaussian Probability Distortion Operator, Journal of Risk Insurance, 79(3): 841-866.

Goldberger, A. S., 1968, The Interpretation and Estimation of Cobb-Douglas Functions, Econometrica, 36(3/4): 464-472.

Golden, L. L., C. C. Yang, and H. Zou, 2010, The Effective ness of Using Basis Hedging Strategy to Mitigate the Financial Consequences of Weather-Related Risks, North American Actuarial Journal, 14(2): 157-175.

Golden, L. L., M. Wang, and C. Yang, 2007, Handling Weather Related Risks Through the Financial Markets: Considerations of Credit Risk, Basis Risk and Hedging, Journal of Risk and Insurance, 74(2): 319-346.

Hart, R. A., J. Hutton, and T. Sharot, 1975, A Statistical Analysis of Association Football Attendances, Applied Statistics, 24(1): 17-27.

Holt, C. A., and S. K. Laury, 2002, Risk Aversion and Incentive Effects, American Economic Review, 92(5): 1644-1655.

Hurlimann, W., 2004, Distortion Risk Measure and Economic Capital, North American Actuarial Journal, 8(1): 86-95.

Hynds, M., and I. Smith, 1994, The Demand for Test Match Cricket, Applied Economics Letters, 1(7): 103-106.

IPCC, 2012, Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation, in: C. B. Field, V. Barros, T. F. Stocker, D. Qin, D. J. Dokken, K. L. Ebi, M. D. Mastrandrea, K. J. Mach, G.-K. Plattner, S. K. Allen, M. Tignor, P. M. Midgley, eds., A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change (Cambridge, UK: Cambridge University Press), p. 582. World Wide Web: (accessed March 21, 2014).

Jewson, S., 2002, Weather Option Pricing with Transaction Costs, Energy Risk (December 2002), pp. 50-53. World Wide Web: pdf/december/technical.pdf (accessed March 21, 2014).

Jones, B. L., and R. Zitikis, 2007, Risk Measures, Distortion Parameters, and Their Empirical Estimation, Insurance: Mathematics and Economics, 41(2): 279-297.

Kahneman, D., and A. Tversky, 1979, Prospect Theory: An Analysis of Decision Under Risk, Econometrica, 47(2): 263-291.

Kalist, D. E., 2010, Terror Alert Levels and Major League Baseball Attendance, International Journal of Sport Finance, 5(3): 181-192.

Kawai, S., and T. Hirata, 2008, J. League Kansen Juyo ni Kansuru Kenkyu [Attendance Demand for J. League], Sport Sangyo Kenkyu [Industrial Sport Research], 18(2): 11-19.

Kijima, M., and Y. Muromachi, 2008, An Extension of the Wang Transform Derived From Buhlmann's Economic Premium Principle for Insurance Risk, Insurance: Mathematics and Economics, 42(3): 887-896.

Labuschagne, C. C. A., and T. M. Offwood, 2010, A Note on the Connection Between the Esscher-Girsanov Transform and the Wang Transform, Insurance: Mathematics and Economics, 47(3): 385-390.

Lee, S., L. A. Farley, and O. Kwon, 2010, The Effectiveness of Risk Management Plans in Recreational Sport Programs of Division I-A Universities, Recreational Sports Journal, 34(1): 58-68.

Leggio, K. B., 2007, Using Weather Derivatives to Hedge Precipitation Exposure, Managerial Finance, 33(4): 246-252.

Lin, X. S., K. S. Tan, and H. Yang, 2009, Pricing Annuity Guarantees Under a RegimeSwitching Model, North American Actuarial Journal, 13(3): 316-332.

Martin, S. W., B. J. Barnett, and K. H. Coble, 2001, Developing and Pricing Precipitation Insurance, Journal of Agricultural and Resource Economics, 26(1): 261-274.

McLeish, D. L., and M. R. Reesor, 2003, Risk, Entropy, and the Transformation of Distributions, North American Actuarial Journal, 7(2): 128-144.

Monfort, A., and F. Pegoraro, 2012, Asset Pricing With Second-Order Esscher Transforms, Journal of Banking and Finance, 36(6): 1678-1687.

Moridaira, S., 2010, Day-Counting Weather Derivative Pricing Models: An Application of Event Risk Insurance and Derivatives Modeling, Working Paper, The Geneva Association.

Morley, B., and D. Thomas, 2007, Attendance Demand and Core Support: Evidence From Limited-Overs Cricket, Applied Economics, 39(16): 2085-2097.

Myers, R., 2008, What Every CFO Needs to Know About Weather Risk Management, Storm Exchange, Inc. & CME Group (New York, NY). World Wide Web: (accessed September 21, 2012).

Nihon Keizai Shimbun, 2002a, S-Pulse Introduced Precipitation Related Insurance Coordinated by Shizuoka Bank, Nihom Keizai Shimbun Sha (Shizuoka version), May 2, 6 (in Japanese).

Nihon Keizai Shimbun, 2002b, Sun Freche Hiroshima Have Contracted Weather Related Insurance Following to Shimizu S-pulse, Nihon Keizai Shimbun Sha (Hiroshima version), September 12, 23 (Japanese).

Noll, R. G., 1974, Attendance and Price Setting, in: R. Noll, ed., Government and the Sports Business (Washington, DC: Brookings Institute).

Pelsser, A., 2008, On the Applicability of the Wang Transform for Pricing Financial Risks, Astin Bulletin, 38(1): 171-181.

Pennings, J. M. E., and A. Smidts, 2000, Assessing the Construct Validity of Risk Attitude, Management Science, 46(10): 1337-1348.

Perez-Gonzalez, F., and H. Yun, 2013, Risk Management and Firm Value: Evidence From Weather Derivatives, Journal of Finance, 68(5): 2134-2176.

Schofield, J. A., 1983, The Demand for Cricket: The Case of the John Player League, Applied Economics, 15(3): 283-296.

Skees, J., S. Gober, P. Varangis, R. Lester, and V. Kalavakonda, 2001, Developing Rainfall-Based Index Insurance in Morocco, The World Bank Policy Research Working Paper Series 2577.

Smith, K., 1993, The Influence of Weather Climate on Recreation and Tourism, Weather, 48(12): 398-404.

Toohey, K., and T. Taylor, 2008, Mega Events, Fear, and Risk: Terrorism at the Olympic Games, Journal of Sport Management, 22: 451-469.

Vedenov, D. V., and B. J. Barnett, 2004, Efficiency of Weather Derivatives as Primary Crop Insurance Instruments, Journal of Agricultural and Resource Economics, 29(3): 387-403.

Vyncke, J. D., M. J. Goovaerts, A. De Schepper, R. Kaas, and J. Dhaene, 2003, On the Distribution of Cash Flows Using Esscher Transforms, Journal of Risk and Insurance, 70(3): 563-575.

Wang, S. S., 2000, A Class of Distortion Operators for Pricing Financial and Insurance Risks, Journal of Risk and Insurance, 67(1): 15-36.

Wang, S. S., 2002, A Universal Framework for Pricing Financial and Insurance Risks, ASTIN Bulletin, 32(2): 213-234.

Wang, S. S., 2007, Normalized Exponential Tilting: Pricing and Measuring Multivariate Risks, North American Actuarial Journal, 11(3): 89-99.

Weather Risk Management Association, 2011, 2011 Weather Risk Derivative Survey. World Wide Web: (accessed March 15, 2014).

Welki, A. M., and T. J. Zlatoper, 1999, U.S. Professional Football Game-Day Attendance, Atlantic Economic Journal, 27(3): 285-2981.

Yang, C. C., L. S. Li, and M.-M. Wen, 2011, Weather Risk Hedging in the European Markets and International Investment Diversification, Geneva Risk and Insurance Review, 36: 74-94.

Yokoyama, T., 2014, On the Risk Management for Agriculture, Sonpo Japan Nipponkoa RM Report 105 (Japanese). World Wide Web: jp/publications/pdf/r105.pdf (accessed March 21, 2014).

Haruyoshi Ito is Assistant Professor of Finance, at the Graduate School of International Management, International University of Japan, 777 Kokusai-cho, Minami Uonuma-shi, Niigata 979-7277, Japan. Ito can be contacted via e-mail: Jing Ai is Associate Professor of Risk Management and Insurance, at the Department of Financial Economics and Institutions, Shidler College of Business, The University of Hawaii at Manoa, Honolulu, HI 96822. Ai can be contacted via e-mail: Akihiko Ozawa is Auditor, Hitachi Capital NBL Corporation, Nishi Shimbashi Square 9F, 1-3-1, Nishi Shimbashi, Minato-Ku, Tokyo 105-0003, Japan. Ozawa can be contacted via e-mail: aozawaletter@yahoo. We would like to thank participants at the 2013 Asian Pacific Risk and Insurance Association Annual Meeting, three anonymous referees, the editor, and Patrick Brockett for their valuable comments. All errors are ours.

(1) Available at September 20, 2012).

(2) In untabulated analysis, we find that the correlations among temperature (HDD and CDD) and precipitation derivatives are very low most times so the more popular temperature derivatives cannot be used to hedge precipitation risks.

(3) Only typhoon, thunderstorm, eruption of volcano, and catastrophic earthquake have caused suspension of a game.

(4) The cost to build a new stadium with a roof is approximately US$244 million, and adding a roof to an existing stadium costs approximately US$73 million or US$730,000 assuming a durable period of 100 years and 0 real interest rate. Assuming an average attendance of 20,000 per game, the average profit per attendance of US$37, an average of 3 rainy day games, and assuming a 60 percent insurance loading (implied in the currently available insurance contracts), the annual premium of weather insurance would be US$186,146. As we will show later, the estimated annual cost of the weather derivative (even at the high 20 percent loading) is much lower ($88,783 from equation (6)*1.2 = $106,540). Annual revenues of J1 teams range from US$20 million to US$85 million.

(5) Most alternative evaluation approaches proposed for the incomplete market require an assumption on the utility function. For example, the Esscher transform requires an exponential or power utility function. Brockett et al. (2009) and Golden, Wang, and Yang (2007) propose the indifference pricing method for weather derivatives assuming a mean-variance utility function.

(6) The other sources of revenue include other advertisement income and dividend distributions from the J. League association. However, data are not available at the game-by-game level that we need for our analysis.

(7) Roof coverage can include none, 25 percent, 50 percent, 75 percent, 100 percent of the spectator area, or domes that cover 100 percent of the spectator area and the entire soccer field. We define a nonroofed stadium as one with a significantly large ([greater than or equal to] 50 percent) exposure in the spectator area. Because we do not study the roofed stadiums in this article, the conclusions drawn and the insights derived are not directly applicable to these teams.

(8) We use data from the manned weather station closest to the stadium even though there might be an unmanned station closer to the stadium. This is because a manned station provides more detailed precipitation information, including a distinction between "--" (no rain) and "0.0" (minimum rain), and a more refined unit of increments at 0.5 mm instead of 1mm as for unmanned stations.

(9) One limitation of this definition is that we are potentially excluding foreign star players' impact on game attendance. Because of the recent decline in the percentage of foreign players on J. League teams and the difficulty in comparably defining foreign star players as well as data availability, we do not attempt to include foreign star players here in the analysis. However, in our experiments (not reported), the inclusion of these players yields qualitatively similar results.

(10) While we have searched for and included as controls most variables motivated by the literature or practice, there are possibly other variables not included due to data availability or difficulty in measurement, for example, pollution, market sentiment, and so on.

(11) All univariate and multivariate regression results are available upon request. Among all 354 precipitation measures, 177 are significant at the 1 percent or 5 percent level for Team Shimizu, which is mostly affected by adverse weather condition, while only 6 measures are significant at the 1 percent or 5 percent level for Yokohama FM, Nippa, which is least affected. On average, 89 variables are significant at the 1 percent or 5 percent level.

(12) Most control variables have signs that are consistent with our expectation and the previous literature, as discussed in the article. Surprisingly, Population is negatively significant or nonsignificant for most teams. This is perhaps due to the negative growth of Japanese population in some of the cities where the teams locate, although the popularity of soccer games still prevails.

(13) In fact (untabulated) analysis shows moderate correlation between the significance of the precipitation effects and the game attendance scaled by capacity of the stadiums. For instance, in the case of Team Yokohama FM, the effect of weather is stronger when they play at the Nissan stadium than at the Nippa stadium, although Nippa has no roof while Nissan has a roof covering half of the spectators. Nippa stadium's capacity is only 15,454 compared to a capacity of 73,327 in the Nissan stadium. The average attendance for all Yokohama FM matches was 25,686 in the 2010 season. Therefore, it is plausible that tickets for Nippa are more desirable than for Nissan, making it difficult for fans to give up tickets for a game in the Nippa stadium even on a rainy day.

(14) The combination Kofu, Kose does not have significant results under any of our precipitation definitions for unique reasons. The team focuses mainly on the regional market and there is no other professional sports team close to the Kofu area, resulting in more robust attendance. We did not control for this factor in the regressions because it is unique to this one team.

(15) We have also run panel regressions with team and year fixed effects, using the same set of precipitation measures and control variables. Instead of estimating team-specific precipitation exposures, the panel regressions estimate the average precipitation exposure for a J1 team. Consistent with the team-specific regressions, we find that the average impact of precipitation is highly significant. The control variables also have signs and significance levels similar to the time-series regressions. The panel regression results using the two precipitation measures d0_h3 and d4_h24 are presented in the accompanying online appendix.

(16) Financial information is only available annually for years 2005-2010 and is aggregated at the team level.

(17) Note that in simulating the impact of precipitation on teams' financial outcomes, we focus on attendance as this is the main channel on which precipitation takes effect. J. League teams do have other sources of income that are not directly related to game attendance. Notably, a significant percentage of the ticket sales are in the form of season tickets (40 percent according to the J. League Fun Survey in 2009, a summary of which is available from the J. League Web site at as of March 18, 2014). In addition, advertising income contributes greatly to teams' revenues. However, attendance (and hence precipitation) can still impact these financial sources in other indirect ways, such as concessions and other sales in the stadium, advertisement and commercial revenues based on popularity, and so on. Because we lack the data and precise measures to account for the indirect impact, our analysis mainly explains the income variations due to reduced gate receipt sales and related financial losses.

(18) In fiscal year 2010, J1 teams had an average operating income of -41million JPY (-US$0.5 million), and those that had positive operating income had an average of 133 million JPY (US$1.63 million).

(19) Available at the J. League Web site (, last accessed in March 2012.

(20) We use the same multivariate regression model to calibrate the impact of precipitation on game attendance, but rerun it with a shorter sample period for two reasons. First, only 6 years of financial data (2005-2010) are available for use as compared to the 18 years of attendance data used for the weather exposure estimation. Second, a structural change between 2006 and 2007 was identified using the Chow test in the relationship between game attendance and adverse weather conditions. OLS regression is used since no autocorrelation is identified in the residuals. The coefficient for precipitation variable (d4_24) is -0.143 with a p-value of 0.008 and standard deviation of 0.052. Full estimation results are shown in the accompanying online appendix.

(21) We cannot run regressions of financial information on the weather conditions directly due to data availability. We only have a very short period of financial information for each team and if we run a pooled regression, the effect of weather on the financials are contaminated because the effect of weather on attendance is varied among teams as we have previously shown. Attendance is closely related to financial outcomes for the pooled data as the significance levels of the coefficients for attendance in the pooled revenue regression and cost regression are 0.04 and 0.02, respectively. Adjusted [R.sup.2]s of the revenue and the cost regression are 86 percent and 94 percent, respectively.

(22) Summary statistics for game attendance and key financial data for team Kawasaki F are provided in the accompanying online appendix.

(23) In calculating the estimated revenues and costs, we add an extra term, the exponential of the squared standard deviations of residuals from the revenue and cost regressions divided by 2, to adjust for the bias resulting from the logarithm transformation. More specifically E[cost] = exp{[[alpha].sub.cost] + [[beta].sub.cost] simulated attendance}exp([[phi].sup.2.sub.[epsilon]]/2), and E[revenue] = exp{[[alpha].sub.revenue] + [[beta].sub.revenue] simulated attendance}exp([[sigma].sup.2.sub.[epsilon]]/Z). This is because E[cost] [not equal to] exp{[[alpha].sub.cost] + [[beta].sub.cost] simulated attendance}. In fact, cost = exp{[[alpha].sub.cost] + [[beta].sub.cost] simulated attendance + [[??].sub.cost]} = exp{[[alpha].sub.cost] + [[beta].sub.cost] simulated attendance}exp([[??].sub.cost]). Thus, E[cost] = E[exp{[[alpha].sub.cost] + [[beta].sub.cost] simulated attendance}]E[exp([[??].sub.cost])]. By assuming error term [[??].sub.cost] follows normal distribution, we have E[cost] = exp{[[alpha].sub.cost] + [[beta].sub.cost] simulated attendance}exp([[sigma].sup.2.sub.[epsilon]]/2). See Goldberger (1968).

(24) Pelsser (2008) argues that the Wang transform has limited applicability in financial pricing because it is not consistent with arbitrage free pricing in the general context. Since the weather derivative market we are working with is typically not an arbitrage-free economy and there is not a uniform method for pricing in this case, the Wang transform seems to be an appropriate approach for us to adopt here.

(25) In order to focus on the effect of the weather-related exposure and due to data limitations, we do not take into consideration interest rate risk, credit risk, basis risk, and others.

(26) Although we are considering a complete hedge ex ante, the linear payoff structure of the derivative is based on the expected loss from the precipitation exposure instead of the actual loss determined after the games. Therefore, the weather derivative product we consider is not to be confused with a weather insurance policy. In addition, we perform most analyses based on the complete-hedge contract design only for simplicity. Similar analysis can be easily implemented for contracts that provide only a partial hedge ex ante, such as those developed from the lower bounds of precipitation exposure estimates discussed in this article.

(27) Here we design a weather hedging instrument that pays off even when there is only one rainy day per season. In some real-life scenarios, the firm might want to use a similar future contract or option contract whose payoff is positive only if the number of rainy days is above the average or a set threshold value. However, we do not have sufficient information on the prices of these precipitation contracts because they are not traded for the Japanese market. We attempted to estimate the bounds using limited information from the U.S. market as a proxy but were not able to obtain meaningful results. In addition, we do not have the risk preference of market participants to calibrate the premium under the risk neutral measure. This is why we resort to the current design of the hedging instrument. The insights, however, should be largely similar.

(28) Chinacarini (2011) estimates the median temperature forward premium to be 5.76 percent. The forward premium is defined as E([F.sub.t] - [S.sub.t+1])/[F.sub.t], where [F.sub.t] is forward price at time t and [S.sub.t+1] is underlying weather index at time t + 1. If we calculate the safety loading by [F.sub.t],/E([S.sub.t+1]) - 1, the 5.76 percent premium translates to a safety loading of 5 percent.

(29) We summarize the policies for Teams Shimizu and Hiroshima in the accompanying online appendix (details only available for 2001). For example, Team Shimizu's insurance policy stipulates that the insurance company pays the team US$2,957 (242,500 JPY) if there is at least 10 mm of precipitation on the day of the insured game. The details of the contract for Team C Osaka are not available.

(30) We describe our estimation for the implied safety loading in detail in the accompanying online appendix.

(31) The teams are Kobe, Tokyo V, Yokohama FM, C Osaka, Nigata, Okayama, Sapporo, Ehime, Tochigi, and Tokushima.

(32) There are possibly other types of transaction costs in addition to the premium loading considered in the article. Our results on the evaluation of weather hedging instrument is fairly robust to reasonable transaction costs. For example, Jewson (2002) and Chen, Roberts, and Thraen (2006) cite a 5 percent or 6 percent transaction cost for weather derivatives. Even for a high premium loading of 20 percent, the hedging instrument is still valuable to managers whose risk preference falls within the estimated bounds in our article.

(33) Overhedging will be a rare case in practice because a complete hedge is unlikely due to the cost of hedging.
Table 1
Precipitation Effects on Game Attendance

         Shimizu Nihon Daira     G Osaka Banpaku

d0_h3    -0.103 ***              -0.127 ***
         (0.027)                 (0.043)
d4_h24               -0.108 ***              -0.084 *
                     (0.028)                 (0.049)
AIC      -75.337     -75.315     124.558     130.587
BIC      -19.713     -19.692     181.343     187.372
N        239         239         257         257

         Kofu Kose               Iwata Yamaha

d0_h3    0.013                   -0.075 ***
         (0.086)                 (0.027)
d4_h24               -0.045                  -0.064 ***
                     (0.091)                 (0.028)
AIC      -8.764      -9.114      -129.601    -126.824
BIC      11.756      11.406      -75.086     -72.309
N        32          32          223         223

         Kyoto Saikyogoku        C Osaka Nagai

d0_h3    -0.168 ***              -0.062
         (0.049)                 (0.073)
d4_h24               -0.229 ***              -0.097
                     (0.053)                 (0.085)
AIC      63.548      56.645      135.925     135.354
BIC      108.095     101.192     183.218     182.647
N        144         144         142         142

         Kashiwa Kashiwa         Omiya NACK

d0_h3    -0.074 **                -0.141 *
         (0.029)                 (0.073)
d4_h24               -0.105 ***              -0.085
                     (0.031)                 (0.069)
AIC      -109.325    -114.518    -17.931     -15.205
BIC      -60.734     -65.926     12.978      15.704
N        154         154         51          51

         Nagoya Mizuho Rikujo    Hiroshima Hiroshima

d0_h3    -0.085 **               -0.176 ***
         (0.039)                 (0.048)
d4_h24               -0.169 ***              -0.125 **
                     (0.041)                 (0.057)
AIC      -2.421      -14.637     126.912     135.238
BIC      45.965      33.749      176.084     184.41
N        186         186         196         196

         Kawasaki F Todoroki     Yokohama FM Nissan

d0_h3    -0.181 ***               -0.114 **
         (0.050)                 (0.047)
d4_h24               -0.224 ***              -0.156 ***
                     (0.051)                 (0.055)
AIC      8.892       2.396       42.366      40.408
BIC      52.95       46.454      91.866      89.908
N        116         116         163         163

         Yokohama EM Nippa

d0_h3    -0.104 **
d4_h24              -0.068
AIC      -22.109    -18.135
BIC      16.398     20.373
N        82         82

         Shonan Hiratsuka

d0_h3    -0.098 *
d4_h24              -0.133 **
AIC      1.745      0.117
BIC      41.382     39.754
N        88         88




Notes: This table presents parameter estimates from GLS regressions
examining the impact of precipitation on game attendance for each
unique team/home stadium combination. (The natural log of) game
attendance is regressed on precipitation, as measure by a dummy
variable d0_h3 or d4_h24, respectively, taking value 1 if there is any
precipitation in the 3 hours before the game or at least 4-mm
precipitation during the game day, and a set of control variables. AIC
(Akaike information criteria) and BIC (Bayesian information criteria)
measures are presented for model fit. N is the total number of
observations for each regression: the number of games held in each J1
team's home stadium during the sample period.***, **, and * indicate 1
percent, 5 percent, and 10 percent significance level, respectively.
Full regression results are available in the accompanying online

Table 2

Mean Probability of a Rainy Day Game in Each Season and Pairwise
Differences Between Seasons

Season   Mean          1           2           3          4           5

1        0.157         0
2        0.193         0.036 ***   0
3        0.137         0.020 *     0.056 ***   0
4        0.103         0.054 ***   0.090 ***   0.034 ***  0
5        0.064         0.093 ***   0.129 ***   0.073 ***  0.039 ***   0

Notes: This table presents mean probabilities of rainy days for each
season and pairwise differences between seasons. Season 1 includes
March, April, May, and October; Season 2 includes June and September;
Season 3 includes July; Season 4 includes August and November; and
Season 5 includes December. January and February are excluded since no
game is held. A rainy day is defined as a day with at least 4-mm
precipitation as in the variable d4_h24. These probabilities are
calculated from the weather data observed at the Yokohama Observatory
from 1968 to 2010. *** and * indicate 1 percent and 10 percent
significance levels, respectively.

Table 3
Simulated Probability of Having n Rainy Game Days in a Soccer Season

              Number of Rainy Game Days (n)

              0        1        2        3        4        5

Probability   0.0596   0.1897   0.2702   0.2390   0.1411   0.0657

              Number of Rainy Game Days (n)

              6        7        8        9

Probability   0.0261   0.0071   0.0014   0.0001

Notes: Probabilities of having n rainy days in a soccer season are
simulated using the mean probabilities of rainy days in each of the
five seasons as shown in Table 2. The probabilities of having 10 or
more rainy days are ignorable. Ten thousand simulation paths are used.

Parameter Estimates for Financial Performance of Team Kawasaki F

Dependent    ln(revenue)              ln(cost)

Intercept    6.540 *** (-0.339)        6.598 *** (-0.221)
Attendance   5.167 x [10.sup.-6] ***   4.932 x [10.sup.-6] ***
             (1.101 x [10.sup.-6])     (7.175 x [10.sup.-7])
sd([??])     0.020                    0.013

Notes: This table presents the parameter estimates and standard errors
(in parentheses) from regressions of ln(revenue) and ln(cost) on
attendance, respectively, for team Kawasaki F. Also presented are the
standard deviations of the residuals in the regression
models. *** indicates 1 percent significance level.

Table 5
Descriptive Statistics for Simulated Net Cash Flows for Team Kawasaki
F (in Thousands of U.S. Dollars; Number of Simulation Paths: 10,000)

Statistic            Lower Bound   Point Estimate   Upper Bound

Expected value       670.93        716.44           768.38
Median               686.65        727.45           773.39
Standard deviation   75.86         51.26            22.47
Maximum              807.72        807.56           807.72
Minimum              373.12        508.69           674.14
Skewness             -4.51          -5.09            -5.73

Notes: This table presents descriptive statistics of simulated net
cash flows for team Kawasaki F. We first run simulations to generate
game-day precipitation as presented in Table 3 and estimate the
attendance using regression coefficients as described in footnote 20.
In addition to using the point estimate for the precipitation exposure
(-0.143), we also use its 90 percent confidence interval (-0.226,
--0.058), that is, the "lower bound" and "upper bound" estimates. We
estimate the cost and revenue based on the calibrated attendance and
parameters presented in Table 4. NCF is then calculated as

Table 6
Simulated Net Cash Flows With Hedging for Team Kawasaki F in
(Thousands of U.S. Dollars)

                     Lower Bound

                     No        Hedging   Hedging
                     Hedging   (10%)     (20%)

Expected value       670.93    662.05    653.17
Median               686.65    662.47    653.59
Standard deviation   75.86     25.60     25.60
Maximum              807.72    710.06    701.18
Minimum              373.12    575.78    566.90
Skewness             -4.51     -2.31     -2.31

                     Point Estimate

                     No        Hedging   Hedging
                     Hedging   (10%)     (20%)

Expected value       716.44    707.56    698.68
Median               727.45    708.09    699.22
Standard deviation   51.26     4.95      4.95
Maximum              807.56    731.60    722.72
Minimum              508.69    688.34    679.46
Skewness             -5.09     -4.96     -4.96

                     Upper Bound

                     No        Hedging   Hedging
                     Hedging   (10%)     (20%)

Expected value       768.38    759.50    750.62
Median               773.39    751.10    742.22
Standard deviation   22.47     28.73     28.73
Maximum              807.72    888.80    879.92
Minimum              674.14    710.06    701.18
Skewness             -5.73     6.39      6.39

Notes: This table presents summary statistics for simulated net cash
flows with hedging for the three estimates of precipitation exposure
(the point estimate, the lower bound of the point estimate, and the
upper bound). Net cash flows with hedging are calculated as
[NCF.sub.with_hedge] = [NCF.sub.without hedge] - premium + payoff,
where payoff is the expected loss based on the point estimate and
premium is expected payoff plus loadings. Net cash flows without
hedging are from Table 5.

Table 7
Summarized Questionnaires for Risk Aversion Estimation of J. League

            Scenario A

      P     Attendance   P     Attendance

Panel A: Greater Difference in Outcomes
(Scenario A) Versus Lesser Difference (Scenario
B), Unit: One Person

01    10%   19,007       90%   7,449
02    20%   19,007       80%   7,449
03    30%   19,007       70%   7,449
04    40%   19,007       60%   7,449
05    50%   19,007       50%   7,449
06    60%   19,007       40%   7,449
07    70%   19,007       30%   7,449
08    80%   19,007       20%   7,449
Q9    90%   19,007       10%   7,449

            Scenario A

      P     Revenue      P     Revenue

Panel B: Greater Difference in Outcomes
(Scenario A) Versus Lesser Difference (Scenario
B), Unit: Thousand JPY

Q10   10%   37,577       90%   14,727
Qll   20%   37,577       80%   14,727
Q12   30%   37,577       70%   14,727
Q13   40%   37,577       60%   14,727
Q14   50%   37,577       50%   14,727
Q15   60%   37,577       40%   14,727
Q16   70%   37,577       30%   14,727
Q17   80%   37,577       20%   14,727
Q18   90%   37,577       10%   14,727

      Scenario B                             E(A)     E(B)

      P      Attendance   P     Attendance

Panel A: Greater Difference in Outcomes
(Scenario A) Versus Lesser Difference (Scenario
B), Unit: One Person

01    10%    16,118       90%   10,338       8,605    10,916
02    20%    16,118       80%   10,338       9,761    11,494
03    30%    16,118       70%   10,338       10,916   12,072
04    40%    16,118       60%   10,338       12,072   12,650
05    50%    16,118       50%   10,338       13,228   13,228
06    60%    16,118       40%   10,338       14,384   13,806
07    70%    16,118       30%   10,338       15,540   14,384
08    80%    16,118       20%   10,338       16,695   14,962
Q9    90%    16,118       10%   10,338       17,851   15,540

             Scenario B                      E(A)     E(B)

      P      Revenue      P     Revenue

Panel B: Greater Difference in Outcomes
(Scenario A) Versus Lesser Difference (Scenario
B), Unit: Thousand JPY

Q10   10%    31,865       90%   20,438       17,012   21,581
Qll   20%    31,865       80%   20,438       19,297   22,723
Q12   30%    31,865       70%   20,438       21,582   23,866
Q13   40%    31,865       60%   20,438       23,867   25,009
Q14   50%    31,865       50%   20,438       26,152   26,152
Q15   60%    31,865       40%   20,438       28,437   27,294
Q16   70%    31,865       30%   20,438       30,722   28,437
Q17   80%    31,865       20%   20,438       33,007   29,580
Q18   90%    31,865       10%   20,438       35,292   30,722

      E(A)--E(B)   Implied Interval of [lambda]

Panel A: Greater Difference in Outcomes
(Scenario A) Versus Lesser Difference (Scenario
B), Unit: One Person

01    -2,311       -[infinity]   -1.28
02    -1,733       -1.28         -0.84
03    -1,156       -0.84         -0.52
04    -578         -0.52         -0.25
05    0            -0.25         0
06    578          0             0.25
07    1,156        0.25          0.52
08    1,733        0.52          0.84
Q9    2,311        0.84          1.28

      E(A)--E(B)   Implied Interval of [lambda]

Panel B: Greater Difference in Outcomes
(Scenario A) Versus Lesser Difference (Scenario
B), Unit: Thousand JPY

Q10   -4,569       -[infinity]   -1.28
Qll   -3,426       -1.28         -0.84
Q12   -2,284       -0.84         -0.52
Q13   -1,142       -0.52         -0.25
Q14   0            -0.25         0
Q15   1,143        0             0.25
Q16   2,285        0.25          0.52
Q17   3,427        0.52          0.84
Q18   4,570        0.84,         1.28

            Scenario A               Scenario B

      P     Atten-   P     Atten-    P      Atten-    E(A)
            dance          dance            dance

Panel C: Risky Scenario (Scenario A) Versus
Deterministic Scenario (Scenario B), Unit:
One Person

Q19   50%   19,007   50%   7,449     100%   17,851    13,228
Q20   50%   19,007   50%   7,449     100%   16,695    13,228
Q21   50%   19,007   50%   7,449     100%   15,540    13,228
Q22   50%   19,007   50%   7,449     100%   14,384    13,228
Q23   50%   19,007   50%   7,449     100%   13,228    13,228
Q24   50%   19,007   50%   7,449     100%   12,072    13,228
Q25   50%   19,007   50%   7,449     100%   10,916    13,228
Q26   50%   19,007   50%   7,449     100%   9,761     13,228
Q27   50%   19,007   50%   7,449     100%   8,605     13,228
Q28   50%   19,007   50%   7,449     100%   7,449     13,228

            Scenario A               Scenario B
      P     Revenue   P     Revenue   P      Revenue   E(A)

Panel D: Risky Scenario (Scenario A) Versus
Deterministic Scenario (Scenario B), Unit:
Thousand JPY

Q29   50%   37,577    50%   14,727    100%   35,292    26,152
Q30   50%   37,577    50%   14,727    100%   33,007    26,152
Q31   50%   37,577    50%   14,727    100%   30,722    26,152
Q32   50%   37,577    50%   14,727    100%   28,437    26,152
Q33   50%   37,577    50%   14,727    100%   26,152    26,152
Q34   50%   37,577    50%   14,727    100%   23,867    26,152
Q35   50%   37,577    50%   14,727    100%   21,582    26,152
Q36   50%   37,577    50%   14,727    100%   19,297    26,152
Q37   50%   37,577    50%   14,727    100%   17,012    26,152
Q38   50%   37,577    50%   14,727    100%   14,727    26,152

      Risk Free   Implied Interval of [lambda]

Panel C: Risky Scenario (Scenario A) Versus
Deterministic Scenario (Scenario B), Unit:
One Person

Q19   4,623       -[infinity]   -1.28
Q20   3,467       -1.28         -0.84
Q21   2,312       -0.84         -0.52
Q22   1156        -0.52         -0.25
Q23   0           -0.25         0
Q24   -1156       0             0.25
Q25   -2,312      0.25          0.52
Q26   -3,467      0.52          0.84
Q27   -4,623      0.84          1.28
Q28   -5,779      1.28          [infinity]

      Risk Free   Implied Interval of [lambda]

Panel D: Risky Scenario (Scenario A) Versus
Deterministic Scenario (Scenario B), Unit:
Thousand JPY

Q29   9,140       -[infinity]   -1.28
Q30   6,855       -1.28         -0.84
Q31   4,570       -0.84         -0.52
Q32   2,285       -0.52         -0.25
Q33   0           -0.25         0
Q34   -2,285      0             0.25
Q35   -4,570      0.25          0.52
Q36   -6,855      0.52          0.84
Q37   -9,140      0.84,         1.28
Q38   -11,425     1.28          [infinity]

Notes: This table summarizes the four sets of questionnaires
distributed to J. League managers in which they select the preferred
scenario in each question. The questionnaires are designed based on
data from the average J1 club, Iwata. In Panels A and B, Scenario A is
relatively risker than Scenario B. The higher value in Scenario A is
calculated as the mean attendance (revenue) of Iwata from 1994 to 2010
plus two standard deviations in Panel A (B). The lower value in
Scenario A is similar but with two standard deviations below the mean
instead. The higher and lower value in Scenario B is similarly
calculated as in Scenario A but with one standard deviation above or
below the mean instead of two standard deviations. In Panels C and D,
we set Scenario B's attendance so that the implied interval of A is
the same as in Panels A and B for each question. The implied interval
of A for each respondent corresponds to a switch of preferred choice
from Scenario B to A in all panels. The columns E(A), E(B), Risk
Free--E(A), and Implied Interval of A were not included in the actual
distributed questionnaires. Revenues in the questionnaire were denoted
in lapanese yen (JPY). In Panel B, JPY 37,577 thousand, 14,727
thousand, 31,865 thousand, and 20,438 thousand are equivalent to
US$458,000, US$180,000, US$389,000, and US$249,000, respectively (as
of March, 2012 when we distributed the surveys).

Table 8
Summary of the Most Plausible Ranges for the Implied Risk Aversion
Parameter [lambda]

Manager   Min [lambda]   Max [lambda]

1         0.25           0.52
2         0.25           0.52
3         0              0.52
4         Inconsistent   Inconsistent
5         0.84           Infinity
6         0              0
7         0.25           0.52
All       0.25           0.52

Notes: This table presents the most plausible ranges for J. League
managers' risk aversion parameter [lambda]. Min and Max [lambda]
indicate the lower and upper bounds of the most plausible range of l
for each manager. We calculate the most plausible range of parameter
[lambda] for each manager as follows. First, we obtain the range of
[lambda] implied from each set of questions in the questionnaires as
presented in Table 7, Panels A-D. Second, for each team manager, we
determine the most frequently represented range of [lambda] in the
four sets. We note that Manager 4 has inconsistent risk preferences
because his preferred scenario changes from A to B in all panels.
Lastly, we find the most plausible range for an average J. club
manager as the range that appears the most times among all managers.

Table 9
The Wang-Transform-Based Evaluation of the Weather Hedging Instrument

                 Safety Loading

                                   With Hedge

[lam-   No       0%       5%       10%      15%      20%      60%
bda]    Hedge

Panel A: Expected Net Cash Flows Under Q-Measure and Value of Hedging,
Based on Point Estimate (Thousands of U.S. Dollars)

0       716.44   716.44   712.00   707.56   703.12   698.68   663.17
0.1     711.34   715.94   711.50   707.06   702.62   698.18   662.67
0.2     706.17   715.44   711.00   706.56   702.12   697.68   662.17
0.25    703.56   715.19   710.75   706.31   701.87   697.43   661.92
0.3     700.92   714.93   710.49   706.05   701.61   697.17   661.66
0.4     695.61   714.42   709.98   705.54   701.10   696.66   661.15
0.5     690.23   713.90   709.46   705.02   700.58   696.14   660.63
0.52    689.14   713.79   709.35   704.91   700.47   696.03   660.52
0.6     684.79   713.37   708.93   704.49   700.05   695.61   660.10
0.7     679.30   712.84   708.40   703.96   699.52   695.08   659.57
0.8     673.76   712.30   707.86   703.42   698.98   694.55   659.03
0.9     668.18   711.76   707.32   702.88   698.44   694.01   658.49
1       662.56   711.22   706.78   702.34   697.90   693.46   657.95

Panel B: Expected Net Cash Flows Under Q-Measure and Value of Hedging,
Based on the Upper Bound (Thousands of U.S. Dollars)

0       768.38   768.38   763.94   759.50   755.06   750.62   715.11
0.1     766.15   765.59   761.15   756.71   752.27   747.83   712.32
0.2     763.87   762.86   758.42   753.98   749.54   745.10   709.59
0.25    762.72   761.52   757.08   752.64   748.20   743.76   708.25
0.3     761.57   760.19   755.75   751.31   746.87   742.44   706.92
0.4     759.22   757.59   753.15   748.71   744.27   739.83   704.32
0.5     756.85   755.05   750.62   746.18   741.74   737.30   701.78
0.52    756.37   754.56   750.12   745.68   741.24   736.80   701.29
0.6     754.44   752.59   748.15   743.71   739.27   734.83   699.32
0.7     752.01   750.20   745.76   741.32   736.88   732.44   696.93
0.8     749.55   747.89   743.45   739.01   734.57   730.13   694.62
0.9     747.07   745.65   741.21   736.77   732.34   727.90   692.38
1       744.56   743.50   739.06   734.63   730.19   725.75   690.23

Panel C: Expected Net Cash Flows Under Q-Measure and Value of Hedging,
Based on the Lower Bound (Thousands of U.S. Dollars)

0       670.93   670.93   666.49   662.05   657.61   653.17   617.66
0.1     663.39   668.38   663.94   659.50   655.06   650.62   615.11
0.2     655.75   665.80   661.37   656.93   652.49   648.05   612.54
0.25    651.89   664.51   660.07   655.63   651.20   646.76   611.24
0.3     648.01   663.22   658.78   654.34   649.90   645.46   609.95
0.4     640.18   660.62   656.18   651.74   647.30   642.87   607.35
0.5     632.28   658.02   653.58   649.14   644.70   640.26   604.75
0.52    630.68   657.50   653.06   648.62   644.18   639.74   604.23
0.6     624.29   655.41   650.97   646.53   642.10   637.66   602.14
0.7     616.25   652.81   648.37   643.93   639.49   635.05   599.54
0.8     608.15   650.21   645.77   641.33   636.89   632.45   596.94
0.9     600.00   647.61   643.17   638.74   634.30   629.86   594.34
1       591.82   645.03   640.59   636.15   631.71   627.28   591.76

Panel D: Maximally Acceptable Safety Loadings

0       0.00%    0.00%    0.00%
0.1     5.18%    -0.63%   5.61%
0.2     10.44%   -1.14%   11.32%
0.25    13.10%   -1.36%   14.21%
0.3     15.78%   -1.55%   17.13%
0.4     21.18%   -1.84%   23.02%
0.5     26.66%   -2.02%   29.00%
0.52    27.76%   -2.04%   30.20%
0.6     32.19%   -2.09%   35.05%
0.7     37.78%   -2.04%   41.18%
0.8     43.41%   -1.87%   47.38%
0.9     49.09%   -1.59%   53.63%
1       54.81%   -1.19%   59.94%

                 Safety Loading

                 Value of Hedging Instrument

[lam-   No       0%      5%      10%      15%      20%      60%
bda]    Hedge

Panel A: Expected Net Cash Flows Under Q-Measure and Value of Hedging,
Based on Point Estimate (Thousands of U.S. Dollars)

0       716.44   0.00    -4.44   -8.88    -13.32   -17.76   -53.27
0.1     711.34   4.60    0.16    -4.28    -8.72    -13.16   -48.67
0.2     706.17   9.27    4.83    0.39     -4.05    -8.49    -44.00
0.25    703.56   11.63   7.19    2.75     1.69     6.13     -41.64
0.3     700.92   14.01   9.57    5.13     0.69     -3.75    -39.26
0.4     695.61   18.81   14.37   9.93     5.49     1.05     -34.46
0.5     690.23   23.67   19.23   14.79    10.35    5.91     29.60
0.52    689.14   24.65   20.21   15.77    11.33    6.89     -28.62
0.6     684.79   28.58   24.14   19.70    15.26    10.82    -24.69
0.7     679.30   33.54   29.10   24.66    20.22    15.78    -19.73
0.8     673.76   38.54   34.11   29.67    25.23    20.79    -14.72
0.9     668.18   43.59   39.15   34.71    30.27    25.83    -9.68
1       662.56   48.66   44.22   39.78    35.34    30.90    -4.61

Panel B: Expected Net Cash Flows Under Q-Measure and Value of Hedging,
Based on the Upper Bound (Thousands of U.S. Dollars)

0       768.38   0.00    -4.44   -8.88    -13.32   -17.76   -53.27
0.1     766.15   -0.56   -5.00   -9.44    -13.87   -18.31   -53.83
0.2     763.87   -1.02   -5.45   -9.89    -14.33   -18.77   -54.29
0.25    762.72   1.21    -5.65   10.09    -14.52   18.96    -54.48
0.3     761.57   1.37    -5.81   10.25    14.69    19.13    -54.64
0.4     759.22   1.63    -6.07   10.51    -14.95   19.39    -54.90
0.5     756.85   1.79    -6.23   10.67    15.11    19.55    -55.06
0.52    756.37   1.81    -6.25   10.69    15.13    -19.57   -55.08
0.6     754.44   -1.85   -6.29   -10.73   -15.17   -19.61   -55.12
0.7     752.01   -1.81   -6.25   -10.69   -15.13   -19.57   -55.08
0.8     749.55   -1.66   -6.10   -10.54   -14.98   -19.42   -54.93
0.9     747.07   -1.41   -5.85   -10.29   -14.73   -19.17   -54.68
1       744.56   -1.06   -5.50   -9.94    -14.38   -18.82   -54.33

Panel C: Expected Net Cash Flows Under Q-Measure and Value of Hedging,
Based on the Lower Bound (Thousands of U.S. Dollars)

0       670.93   0.00    -4.44   -8.88    -13.32   -17.76   -53.27
0.1     663.39   4.98    0.54    -3.90    -8.34    -12.77   -48.29
0.2     655.75   10.05   5.61    1.17     -3.27    -7.70    -43.22
0.25    651.89   12.62   8.18    3.74     0.70     -5.14    -40.65
0.3     648.01   15.21   10.77   6.33     1.89     -2.55    38.06
0.4     640.18   20.44   16.00   11.56    7.12     2.68     -32.83
0.5     632.28   25.74   21.30   16.87    12.43    7.99     -27.53
0.52    630.68   26.81   22.37   17.94    13.50    9.06     -26.46
0.6     624.29   31.12   26.68   22.24    17.80    13.36    -22.15
0.7     616.25   36.56   32.12   27.68    23.24    18.80    -16.71
0.8     608.15   42.06   37.62   33.18    28.74    24.30    -11.21
0.9     600.00   47.61   43.18   38.74    34.30    29.86    -5.65
1       591.82   53.22   48.78   44.34    39.90    35.46    -0.05

Notes: Panels A, B, and C show the expected net cash flows with and
without hedging and the value of the hedging instrument in the three
cases when cash flow sensitivity to precipitation is calibrated using
the point estimate, the upper bound of the point estimate, and the
lower bound, respectively. The expected net cash flows are calculated
using the Wang Transform given the risk aversion [lambda] and the
safety loading. The value of the hedging instrument is calculated as
the difference between the expected net cash flows with and without
hedging. Panel D presents the maximally acceptable safety loading,
that is, zero expected value for the hedging instrument. The estimated
range of [lambda] is 0.25-0.52. These results are shown in bold.
COPYRIGHT 2016 American Risk and Insurance Association, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Ito, Haruyoshi; Ai, Jing; Ozawa, Akihiko
Publication:Journal of Risk and Insurance
Article Type:Report
Geographic Code:9JAPA
Date:Dec 1, 2016
Previous Article:Net contribution, liquidity, and optimal pension management.
Next Article:On the propensity to surrender a variable annuity contract: an empirical analysis of dynamic policyholder behavior.

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters