Sunshine and Stock returns:
Weather effects in Frankfurt
V. Bouten
1Supervised by dr. J.J. Bosma
Master thesis
University of Groningen
June 2018
AbstractThis paper examines the presence of both rational and behavioral weather effects on stock market returns. The existence of weather effects has been tested by adding sunshine as a variable in the Fama-French three-factor model. It was hypothesized that
government policy on renewable energy generation would engender rational weather effects. Behavioral weather effects resulting from changes in investor optimism were
expected to be absent. This study utilizes panel data on 399 equities listed on the Frankfurt stock exchange from 1995-2017. The findings did not support the presence of
either type of weather effects.
Keywords: Stock prices, weather, government policy JEL codes: G12, G41, Q48
1Corresponding author: Valentijn Bouten, student MSc. Finance and Msc. Economics at the University of
Groningen. Postal address: University of Groningen, Faculty of Economics and Business, Nettelbosje 2, P.O.
2
Introduction
In traditional finance theory, the efficient market hypothesis (EMH) claims that market participants are rational and that the securities markets only reflect relevant economic information. These assumptions imply that all assets are correctly priced and that future stock prices follow a random walk pattern. Therefore, it is not possible to earn abnormal returns from mispricing according to the EMH. Yet, multiple researchers have
documented patterns in the market that appear to violate the assumptions of the EMH (e.g., French, 1980; Jones and Litzenberger; 1970, Thaler, 1987).
One of the alleged anomalies is the phenomenon called “weather effects”, where stock market prices are influenced by the weather conditions (Cao and Wei, 2005; Hirshleifer and Shumway, 2003; Saunders, 1993). Saunders (1993) was the first to document weather effects by investigating the relationship between New York City weather and Wall Street stock returns. The empirical results indicated a negative correlation between cloud cover and stock returns. The paper referred to psychology literature to explain the observed correlation. It was argued that favorable weather conditions (e.g., sunshine) positively impact mood states and, consequently, increases optimism among investors. Hirshleifer and Shumway (2003) extended Saunders’ work by investigating the presence of weather effects in 26 countries all over the world. Again, correlations between
weather and stock returns were observed and contributed to the same psychological link between weather and optimism. Interestingly, Hirshleifer and Shumway (2003) even claimed that weather-based trading strategies could be profitable under very low transaction costs.
Nonetheless, the claim that weather effects are indeed market anomalies remains debatable. First of all, the conclusions on weather effects are mostly based on standard regression results with little financial controls. In fact, Trombley (1997) even argued that the significance of the results on weather effects heavily rely on variable
specifications.
Furthermore, the potential of rational explanations for the observed correlations between weather and stock returns have received little attention. In traditional finance theory, markets and agents are rational; therefore, a link between weather and stock returns is only possible if weather conditions contain new relevant economic
information. Previous studies on weather effects argue that weather information is economically irrelevant; therefore, any potential weather effects reflect irrational behavior of financial agents (Saunders, 1993). Yet, a rational explanation for a
3
In this paper, the research question ‘Does the weather impact stock market returns?’ will be critically examined utilizing a panel dataset. Both the potential of behavioral and rational weather effects on the Frankfurt stock exchange returns will be investigated. The different types of weather effects enable us to examine if weather effects are strictly an anomaly in the efficient market hypothesis.
According to the rationality assumptions in the EMH, investor behavior should not be affected by the weather. Hence, it is hypothesized that there are no behavioral weather effects. For potential rational weather effects, the impact of government policy on renewable energy firms will be examined. In order to stimulate the development of the renewable energy industry, governments provide financial incentives linked to the generation of renewable energy. The financial incentives create a more favorable
investment environment, which hypothetically beneficially impacts the capital allocation towards the renewable energy sector. More specifically, feed-in tariffs on solar energy generation imply that the amount of subsidies and/or tax credits received increases on sunny days. Therefore, the attractiveness of solar energy stocks increases. Thus, the government policy on renewable energy is expected to engender (rational) weather effects; higher stock returns on sunny days. In this case, however, the explanation will not be based on psychology, but on rational economics. Hence, these weather effects resulting from government policy will not violate the EMH and traditional finance theory.
The data analysis focuses on the relationship between sunshine and equity returns on the Frankfurt stock exchange from 1995 to 2017. The Fama-French three-factor model will be used as the main benchmark to model variation in stock price returns. The Fama-French three-factor model contains a size factor, a value factor, and a market risk factor. These three factors jointly explain approximately 90 percent of the variation in expected returns of a standardized set of portfolios (Fama, French, 1993). Weather will be added as a control variable in the Fama-French model to assess the impact of weather on asset pricing. Furthermore, a policy variable will be included to examine the impact of rational weather effects resulting from government policy on renewable energy. The results do not provide strong evidence for the presence of either type of weather effects. Multiple model specifications do indicate significant rational weather effects, but these findings are very small and reliant on model specification. Therefore, the real impact of this observed relationship is likely to be negligible.
4
The results from the empirical analysis will be presented in the results section along with multiple robustness checks. Finally, a conclusion section with the most important findings in this paper will follow.
Literature review
Several studies have documented weather effects on stock markets (Cao and Wei, 2005; Hirshleifer and Shumway, 2003; Saunders, 1993). One of the first researches on weather effects was performed by Saunders (1993) on the effects of the New York weather on stock exchange returns. The research focused on four different time samples ranging from 1927 to 1989. Daily percentage of cloud cover was utilized as the weather variable, because of the highly correlations with sunshine, humidity, and precipitation. The
percentages of cloud cover were grouped into three categories: 0-20%, 30-90%, and 100%, reflecting clear days, partially clouded days, and clouded days, respectively. A simple regression with monthly controls indicated a significant relationship between cloud cover and stock returns for two of the four time samples.
To explain the relationship between weather and stock returns, Saunders (1993) refers to literature in psychology. Multiple studies have indicated effects of weather conditions on individual mood states (e.g., Bell, 1981; Cunningham, 1979; Howarth and Hoffman, 1984). Furthermore, sunlight has been linked to increased optimism (Howarth and Hoffman, 1984). The increased optimism amongst investors due to sunlight will explain the observed correlation between weather and stock returns (Saunders, 1993). To support the behavioral explanation for weather effects, Saunders (1993) argues that weather is economically irrelevant in rational markets due to diversification.
Furthermore, the author refers to a paper by Richard Roll (1984), who found only a small impact of weather on orange juice futures pricing in Florida. According to
Saunders (1993), this directly illustrates the economic irrelevance of weather; therefore, any significant relationship between the weather and stock prices illustrates an
irrational behavior by financial agents.
5
on each individual country, the data indicated significant weather effects in for 7 out of the 26 countries investigated. Additionally, Hirshleifer and Shumway (2003) also provide an analysis on weather-based trading strategies, where the mean-variance efficiency of a global equity portfolio is tested. The authors conclude from this analysis that weather-based trading could be profitable for agents facing very low transaction costs (below 5 basis points).
Data
The analysis is performed in a panel setting with data extracted from multiple different data sources. The data comprises of three segments: Firm data, weather data, and Fama-French benchmark factors.
Firm data
The stock market data comprises of the daily closing prices of virtually all equities traded on the Frankfurt stock exchange. All equities listed on the cDAX in the time period ranging from 1 January 1995 to 1 January 2017 have been included. Firms that have been delisted in the relevant time span are kept in the dataset to avoid survivor bias. The cDAX comprises of all equities traded on the Frankfurt stock exchange that are listed in the General standard or Prime standard segment. These segments contain the large majority of all stocks traded in Frankfurt and is therefore assessed as a proper representation of the Frankfurt stock market in total. Additionally, all other Frankfurt-traded equities of firms that actively generate solar energy have been added to the dataset. Since this paper focuses on the effects of German weather, companies that are listed in Frankfurt but that are (primarily) active outside of Germany have been
excluded from the dataset.
Table 1: Equity types
Below, the categories of equities utilized in analysis are displayed. Solar energy equities include all equities of all firms that actively generate solar energy as one of their core business activities.
Total number of equities - Financial equities
- Renewable energy equities o Solar energy equities
399 48 31 16
6
included renewable energy firms is displayed in the appendices section. Furthermore, the distance from each firm’s corporate headquarters to Frankfurt has been calculated in kilometers. This distance variable will be utilized to account for regional differences in weather.
Weather data
All weather data was retrieved from the Deutscher Wetter Dienst. The weather data consists of two elements; daily sunshine duration in hours and cloud cover in okta. The relevant time span, again, ranges from 1 January 1995 to 1 January 2017. The
descriptive statistics of both weather variables are depicted in table 2 below.
Table 2: Descriptive statistics of weather variables
This table displays the descriptive statistics of the weather variables SDt, representing daily
sunshine duration in hours, and CCt, representing daily cloud cover in okta. Weather variables
are measured at Frankfurt airport.
SDt CCt Average 4.6 5.4 Std.dev. 4.4 2.0 Median 3.4 5.9 Mode 0 4 Maximum 15.6 8 Minimum 0 0
The correlation between the SDt and CCt is -08171
Fama-French benchmark factors
The European benchmark factors of the Fama-French three factor model were retrieved trough Kenneth R. French’s website (“Fama/French factors”, 2018). Originally, Fama and French (1993) used monthly factor returns for the three-factor model. In this study, daily factor returns are used to match the other data. The relevant time span for the factors utilized, again, ranges from 1 January 1995 to 1 January 2017.
Methodology
7
variable CCSt indicates the categories full cloud cover, no cloud cover, and modest cloud
cover by the values -1, 1, and 0 respectively.
Table 3: Mean returns per cloud cover categories
The three cloud cover categories are full cloud cover (okta value 8), no cloud cover (okta values 0 to 1.6), and modest cloud cover (all other okta values).
Cloud cover category Mean return in %
Full cloud cover 0,140***
(0,021)
No cloud cover 0,117***
(0,016)
Modest cloud cover 0,099***
(0,006)
Notes: Pairwise mean comparison t-tests have been performed to test the differences in mean
returns per cloud cover category. The mean comparison tests indicate that the mean return percentages are not significantly different for all three possible pairwise combinations of cloud cover categories. different across the three cloud cover categories. *, **, ***, indicates statistically
different from 0 for 10%, 5%, 1% level, respectively.
After the Saunders (1993) model, the main model of this paper will be examined. In this study, the impact of weather will be tested in a traditional finance framework. Panel data is utilized containing daily data on 399 Frankfurt-traded equities from 1995 to 2017. The Fama-French three-factor model functions as the fundament in the analysis. The daily amount of sunlight will be added to the three-factor model to capture any potential weather effects. The main research model that will be investigated is:
𝑅𝑖𝑡 = 𝛼 + 𝐹𝐹𝑡+ 𝛽𝑆𝐷𝑡+ 𝛾(𝑆𝐷𝑡∗ 𝐺𝑖𝑡) + µ𝐺𝑖𝑡+ 𝑋𝑖𝑡+ 𝜖𝑖𝑡 (E1)
Rit represents the daily return in percentages of equity i at time t. 𝛼 is a constant and FF
represents the three-factors from the Fama-French three-factor model2. The behavioral
weather effects will be captured by the coefficient 𝛽 for variable SDt (daily sunshine
duration in hours at time t) and the rational weather effect will be captured by the coefficient 𝛾 for the interaction term SDt*Git. According to the traditional finance theory,
the coefficient measuring behavioral weather effects (β) should be insignificant. The impact of the rational weather effects (coefficient γ) is expected to be positive and significant. Additionally, the models will be estimated with the weather variables CCSt
and CCt, instead of the variable SDt, in order to assess the impact of alternating the
sunshine measures.
2 Corresponding equation is:
8
The variable Git is an indicator variable for German government policy on renewable
energy generation from sunlight. The indicator variable will have a value of 1 only for the renewable energy firms during the time period that the government policy is active and 0 otherwise. The main government policy under investigation will be the
Erneuerbare-Energien-Gesetz (EEG), which is a series of government actions that aim at
stimulating the generation of renewable energy. The EEG was first implemented at 1 April 2004 and since then it has been amended multiple times. First, the variable Git will
correspond to the implementation date of the EEG. The main analysis will focus on the original EEG implementation date. In the robustness tests, the variable Git will
correspond to different amendment dates (See appendix 1).
In the main model, the Xit represents a matrix of control variables. Control variables
included are Disti and the cross-products SDt*Disti and SDt*Disti*Git, where Disti
represents the distance from each firm’s corporate headquarters to Frankfurt. These cross-products are included to control for in-country weather differences. The daily three-factors by Fama-French are expected to control for the impact of seasonality. Additionally, a model with monthly dummies will be estimated to decrease the potential impact of seasonality. Multiple different estimation methods will be utilized with
different specifications to be able to test the robustness of the obtained findings. Expectations
Additionally, the analysis will contain a section that focuses on the impact of
expectations. It seems reasonable to assume that individuals act upon expectations about future weather instead of the actual weather conditions. Consequently, investor decisions might be affected relatively more on days that the actual weather deviates significantly from expected weather conditions. Therefore, a regression analysis that focuses on the relationship between stock returns and deviations from expected weather conditions will be included. Two different approaches will be utilized to
9
Results
First, a similar model to the one utilized by Saunders (1993) will be constructed and performed on the Frankfurt stock exchange data. The daily returns are regressed on a constant, the categorial cloud cover variable, monthly controls (June omitted), and a one-day lagged return variable. The lagged variable was included in the model to control for nonsynchronous trading effects (Saunders, 1993). Unlike the original model, this model does not contain controls for days of the week. Another important difference to note is the fact that instead of index returns, longitudinal data containing individual stock returns is utilized.
Table 4: Simple regression output (total observations = 1.694.160).
Below, the output is displayed of the regression model similar to Saunders (1993). The dependent variable is the daily return in percentages. The main variable CCSt indicates the
categories full cloud cover, no cloud cover, and modest cloud cover by the values -1, 1, and 0 respectively. The model contains monthly controls (June omitted) and a one-day lagged return
variable. Furthermore, cross-sectional fixed effects are applied. In total, 399 equities were included in the regression. The relevant time frame ranges from 1995 to 2017.
Variables Coefficients Constant (α) -0.037* (0.020) Cloud cover (CCSt) 0.017 (0.019) Returnsi, t-1 (in %) -0.066*** (0.001)
Monthly controls Yes
Fixed effects Yes
Notes: R-squared = 0.004, model’s F-statistic = 592.02 (p = 0.000). *, **, ***, statistically different
from 0 for 10%, 5%, 1% level, respectively. All coefficients for the monthly controls were found statistically significant at the 5% level.
10
Next, the focus will be on our main model, as presented in the methodology section. This model allows for potential behavioral weather effects trough investor optimism and potential rational weather effects trough government policy on solar energy generation. The fundamental elements of the model consist of the three factors of the Fama-French three-factor model. Applying solely the three-factor model to the data yields an R-squared value of 0.007; i.e. the model is able to explain approximately 0.7% if the variation in stock prices. This value is significantly lower than the R-squared values in the original model by Fama and French (1993). There are two big differences in the application of the three-factor model compared to Fama and French (1993) that may explain the observed reduction in the R-squared value. First, Fama and French (1993) focus on monthly returns with monthly factor return portfolios, whereas this study utilizes daily factor data. Daily stock data is likely to contain relatively more randomness unrelated to the included factors. The daily randomness is diversified away in monthly return data, which may explain why the model would perform better in that case. Furthermore, Fama and French (1993) focus on time-series data instead of panel data. This study utilizes panel data on 399 equities from 1995-2017. The panel setting allows for an analysis on the variation between cross-sections trough different firm-specific characteristics. The Fama-French three-factor model, however, is not able to explain variation across panel units. Therefore, the R-squared values are likely to be lower in a panel setting compared to an analysis on only time-series data. Nevertheless, all factors from the three-factor model did significantly impact stock returns in our model. Hence, the factors do explain parts of the variation in stock prices. The three-factor model output is presented in appendix 2.
Next, the three-factor model will be extended by including factors for weather effects to the model. A simple sunshine variable and a factor that measures the impact of
government policy on solar energy generation will be included. In the first two models, weather is measured as the hours of sunshine per day (SDt). The first model contains
equity-clustered robust standard errors and fixed effects, but no monthly control variables. The second model does include monthly dummies. In the next two models, cloud cover is used to examine the impact of sunshine. The third model utilizes the amount of cloud cover in okta and the fourth model utilizes the categorial cloud cover variable as described previously.
The outcomes of the panel regressions are displayed in table 5. The coefficients for the Fama-French factors show similar values compared to the three-factor model
regression. Again, all factors are statistically significant for the one percent level. The addition of weather controls does not significantly increase the R2 values compared to
11
control for regional weather differences were all found to be insignificant. Furthermore, the addition of monthly controls did not appear to alter the model outcomes.
In all four models, the impact of the sunshine variable on stock returns is found to be insignificant. The findings from our earlier model, the model that replicates the one used by Saunders (1993), suggested the absence of (behavioral) weather effects on the
Frankfurt stock market. Also in this model, the outcomes do not support a behavioral link between weather and stock returns. Interestingly, the coefficients for the rational weather effects (denoted by Sunshine * Government Policy in table 5) indicate a
statistically significant relationship with stock returns for three out of the four models. The significance is observed for the ten percent significance level in two models and for the five percent level in the other model. This suggests that the firms that receive government (financial) support linked to the amount of solar energy generated experience increased stock returns on relatively sunny days.
The observed rational weather effect is modest, however; the estimates suggest that one additional hour of sunshine duration on a day increases the stock return by 1 basis point. The average daily sunshine duration over the sample period was 4.6 hours with a standard deviation of 4.4. Therefore, it seems unlikely that investors are able to make profits after transaction costs with trading strategies based on this particular rational weather effect. Also, the regression results indicate a significant, negative relationship for the government policy variable. The government policy variable has a value of one for the period that the EEG policy was active, but only for the firms that actively generated solar energy. The coefficient of this variable is found to be significantly negative in three of the four models. Hence, the results suggest that the returns of firms receiving government support under the EEG policy had relatively lower returns during that specific time period than the other firms. The estimated negative effect varies from 11 basis points to slightly under 15 basis points. Again, the real relevance of this
12
Table 5: Main panel regression output.
Below, the panel regression output for our main model is presented. The dependent variable is the daily return in percentages. All four models contain equity-clustered robust standard errors
and cross-sectional fixed effects. In model (1) and (2), Sunshine is measured as hours of daily sunshine. Model (2) also contains monthly controls (June omitted). Cloud cover in okta (higher
value implies less sunshine) is utilized as the sunshine control in model (3). Categorial cloud cover is the weather measure in model (4). The categorial cloud cover variable denotes the categories full cloud cover, no cloud cover, and modest cloud cover by the values -1, 1, and 0
respectively. The analysis utilizes data on 395 equities from 1995-2017.
Variables Coefficients (1) (2) (3) (4) Constant (α) 0.152*** (0.015) 0.040** (0.017) 0.122*** (0.015) 0.139*** (0.010) Risk-free rate (rFt) -6.515*** (1.313) -6.546*** (1.330) -6.501*** (1.313) -6.476*** (1.307) Market risk factor (rMt – rFt) 0.562***
(0.015) 0.559*** (0.016) 0.562*** (0.015) 0.562*** (0.015) Size factor (smbt) 0.111*** (0.022) 0.104*** (0.022) 0.111*** (0.022) 0.111*** (0.022) Value factor (hmlt) -0.169*** (0.013) -0.170*** (0.026) -0.169*** (0.026) -0.169*** (0.026) Sunshine -0.004 (0.003) (0.004) -0.000 (0.008) -0.009 (0.030) -0.027 Sunshine * Government Policy 0.010*
(0.006) (0.006) 0.011* -0.034** (0.015) (0.079) -0.038 Government Policy -0.149**
(0.066) -0.145** (0.065) (0.064) -0.006 (0.061) -0.112*
Monthly controls No Yes No No
Fixed Effects Yes Yes Yes Yes
Distance controls Yes Yes Yes Yes
Observations included 1,678,938 1,678,938 1,678,938 1,678,938
R2 0.007 0.007 0.007 0.007
13
Multiple differentiations with respect to the estimated models have been examined. The exclusion of the financial crisis between 2007 and 2009 does not result in significantly different estimation results. Also, eliminating all financial firms from the dataset does not impact the model outcomes. The inclusion of only DAX-listed firms, only renewable energy firms, or a combination of both resulted in insignificant coefficients for both types of weather effects. Furthermore, a specification that only included firms with a headquarters within 50 kilometers of Frankfurt has been tested. A group of 40 equities fulfilled this distance requirement. In this model, the estimates suggested that the constant, the risk-free rate and the value factor had a significant impact on the stock returns. Additionally, the rational weather effect variable reported an estimated coefficient of 0.016 with a standard error of 0.008, implying statistical significance for the ten percent level. Since all firms are arguably in the same weather region, it is reasonable to expect that the likelihood of observing weather effects is the highest for those equities.
Next, the impact of multiple different amendment dates to the German EEG policy have been investigated. These different amendment dates impact the government policy variable Git and the relevant cross-products. The output for the models with different
policy dates is presented in table 6 below.
Model (1) in table 6 corresponds to the output from model (1) in table 5. The relevant policy date for model (1) is 01-04-2000, the implementation date of the EEG. The other models in the table represent the models corresponding to the policy dates 01-08-2004, 01-01-2009, and 01-07-2000, respectively. The outcomes of the four different models is very similar. The sunshine variable is insignificant for all four models, while the
14
Table 6: Output different policy dates.
Below, the panel regression output for our main model with multiple different policy dates is presented. The dependent variable is the daily return in percentages. The variable Git is adjusted
to match different amendment dates of the EEG policy. The four policy dates under examination are 01-04-2000 (main model), 01-08-2004, 01-01-2009, and 01-07-2000. The numbers below represent the models corresponding to those dates in that respective order. The content of each
amendment is presented in appendix 1. All four models contain equity-clustered robust standard errors and cross-sectional fixed effects. The weather variable utilized is the daily
sunshine duration in hours. The analysis utilizes data on 395 equities from 1995-2017.
Variables Coefficients (1) (2) (3) (4) Constant (α) 0.152*** (0.015) 0.148** (0.015) 0.153*** (0.015) 0.151*** (0.015) Risk-free rate (rFt) -6.515*** (1.313) -6.547*** (1.312) -6.745*** (1.326) -6.661*** (1.319) Market risk factor (rMt – rFt) 0.562***
(0.015) 0.562*** (0.015) 0.562*** (0.015) 0.562*** (0.015) Size factor (smbt) 0.111*** (0.022) 0.111*** (0.022) 0.112*** (0.022) 0.111*** (0.022) Value factor (hmlt) -0.169*** (0.013) -0.170*** (0.026) -0.170*** (0.026) -0.170*** (0.026) Sunshine -0.004 (0.003) (0.003) -0.003 (0.003) -0.003 (0.003) -0.003 Sunshine * Government Policy 0.010*
(0.006) (0.007) 0.003 (0.007) 0.007 (0.007) 0.007 Government Policy -0.149**
(0.066) (0.059) -0.048 -0.224*** (0.044) (0.043) -0.198*
Monthly controls No No No No
Fixed Effects Yes Yes Yes Yes
Distance controls Yes Yes Yes Yes
Observations included 1,678,938 1,678,938 1,678,938 1,678,938
R2 0.007 0.007 0.007 0.007
15
In addition to the previously performed models, a weighted least squares model will be estimated with the inversed geographical distance between each firm’s headquarters and Frankfurt as the weight variable. This method will account for intra-country
weather differences by attributing lower weights to equities that are further away from the weather measurement location Frankfurt. The weighted model has been estimated with the three weather variables; sunshine duration, cloud cover, and categorial cloud cover. The estimated model output is shown in table 7. For all three models, a similar R² value as compared to our previously estimated models has been obtained. The Fama-French factors were found to be significant at the one percent level, except for the size factor. The results suggest that the size factor is not significantly related to the stock returns in this model. This also holds for the government policy variable.
16
Table 7: Weighted Least Squares output.
Below, the weighted least squares output for our main model is presented. The dependent variable is the daily return in percentages. The inversed geographical distance between each
firm’s headquarters and Frankfurt is used as the weight variable. In model (1), Sunshine is measured as hours of daily sunshine. Cloud cover in okta (higher value implies less sunshine) is
utilized as the sunshine variable in model (3). Categorial cloud cover is the weather measure in model (4). The categorial cloud cover variable denotes the categories full cloud cover, no cloud cover, and modest cloud cover by the values -1, 1, and 0 respectively. The analysis utilizes data
on 395 equities from 1995-2017. Variables Coefficients (1) (2) (3) Constant (α) 0.092*** (0.008) 0.147** (0.013) 0.105*** (0.006) Risk-free rate (rFt) -4.771*** (0.513) -4.774*** (0.513) -4.796*** (0.513) Market risk factor (rMt – rFt) 0.543***
(0.005) 0.543*** (0.005) 0.543*** (0.005) Size factor (smbt) -0.013 (0.009) (0.009) -0.013 (0.009) -0.013 Value factor (hmlt) 0.052*** (0.009) 0.052*** (0.009) 0.052*** (0.009) Sunshine 0.003*** (0.001) -0.008*** (0.002) (0.014) 0.010 Sunshine * Government Policy 0.003
(0.004) (0.009) -0.013 (0.062) 0.050
Government Policy -0.021
(0.044) (0.063) -0.060 (0.039) -0.011
Monthly controls No No No
Fixed Effects Yes Yes Yes
Observations included 1,678,938 1,678,938 1,678,938
R2 0.007 0.007 0.007
17
Weather expectations
In this section, the impact of expectations on weather effects will be tested and discussed. If weather information is indeed relevant for investors, then it seems
reasonable to assume that investors rely on expected weather conditions when making financial decisions. The expected weather information will then be reflected in the stock prices. Therefore, the examined weather effects on stock returns are likely to be more prevalent on days that the actual weather conditions deviate significantly from the expected weather conditions. On those days, the weather contains a relatively large amount of new information. A similar theory can be applied to the presence of
behavioral weather effects. The impact of favorable weather conditions on mood is likely to be enhanced if the expectations on the weather conditions were less favorable.
The impact of expectations will be tested by utilizing a similar panel regression model as described earlier (see table 5). In the expectations model, the weather variable will consist of the deviations from the expected weather values instead of the actual weather values. Two different proxies for weather expectations will be utilized. First, expected weather values are predicted by means of month and year variables. Thus, this method results in a proxy that is based upon monthly weather average while taking yearly differences into account. The used deviation variable will then consist of the difference between the actual and the predicted weather values. This first expectations proxy is primarily based on seasonal weather averages. Therefore, this proxy is referred to as the ‘average proxy’. The second proxy is constructed by utilizing the one-year lagged
18
Table 8: Expectation models output.
Below, the panel regression output for our expectation model with two is presented. The dependent variable is the daily return in percentages. Both models contain equity-clustered robust standard errors and cross-sectional fixed effects. The weather control variable utilized is
the daily sunshine duration in hours. The first model (1) utilizes the average proxy for expectations, based on monthly average weather values. The second model (2) utilizes the lagged proxy for expectations, based on weather values on the same day in the previous year.
Variable Sunshine expectation deviation denotes the actual weather values – the expectation proxy values. The analysis utilizes data on 395 equities from 1995-2017.
Variables Coefficients (1) (2) Constant (α) 0.140*** (0.010) 0.139*** (0.010) Risk-free rate (rFt) -6.475*** (1.318) -6.219*** (1.289) Market risk factor (rMt – rFt) 0.562***
(0.015) 0.566*** (0.016) Size factor (smbt) 0.111*** (0.022) 0.117*** (0.022) Value factor (hmlt) -0.169*** (0.026) -0.171*** (0.026) Sunshine expectation deviation -0.001
(0.004) (0.002) 0.000 Sunshine expectation deviation *
Government Policy 0.007 (0.006) 0.008** (0.004) Government Policy -0.113* (0.061) -0.123* (0.068) Monthly controls No No
Fixed Effects Yes Yes
Distance controls Yes Yes
Observations included 1,678,938 1,646,296
R2 0.007 0.007
19
Conclusion
Previous literature on weather effects have referred to a psychological link between favorable weather conditions and investor optimism to explain correlations between stock returns and weather. It was argued that weather does not contain any
economically relevant information; therefore, weather effects reflect irrationality in financial markets (Cao and Wei, 2005; Hirshleifer and Shumway, 2003; Saunders, 1993). In this paper, the presence of both rational and behavioral weather effects has been examined utilizing panel data on 399 Frankfurt-listed equities. Fama-French three-factor model was utilized as a fundament for the research. By adding weather as a control variable to this model, we would be able to examine the significance of weather effects in asset pricing. The purely irrational aspect of weather effects has been tested by including a rational weather effect variable into the model. The rational weather effect variable captures the impact of government policy on renewable energy generation on weather effects. The model thus allowed for both behavioral weather effects and rational weather effects. The government policy on renewable energy under
consideration was the EEG. It was hypothesized that the behavioral weather effects would not significantly impact the stock returns alongside the Fama-French model. Additionally, it was expected that the feed-in-tariffs resulting from government policy on renewable energy generation did engender weather effects. Overall, it was hypothesized that weather effects would only be present in a rational form, implying that weather effects are not necessarily anomalies to traditional finance theory.
Panel data on 399 equities traded on the Frankfurt stock exchange from 1995-2017 was utilized to investigate the presence of weather effects. A variety of models have been estimated to examine the impact of weather effects. The estimation results reported noticeably low R-squared values for all models, where the Fama-French factors were the prime explanatory variables. The addition of the behavioral and rational weather
variables did not significantly enhance the three-factor model’s performance.
Admittedly, multiple estimated models indicate a significant rational weather effect and some effects also suggested the presence of behavioral weather effects. Yet, the size of the observed effects was relatively small (<5 basis points) and the significance of the effects varied with different model specifications.
20
that the weather conditions deviated significantly from expectations are likely to result in adjustments by the financial agents, altering the stock returns. The findings of the expectation models indicated similar results to the other estimated models; ample evidence of a real impact of weather on stock returns.
Overall, there is no real evidence found for the presence of any weather effects impacting stock returns on the Frankfurt stock exchange. Yet, for certain model
specifications, the rational weather effects appeared to have a significant impact on the stock returns. Potentially, this finding suggests that the purely irrationality assumption behind weather effects is not necessarily valid. Either way, no conclusions can be drawn on this matter until further research.
References
Bell, P.A. (1981), “Physiological comfort, performance and social effects of heat stress”,
Journal of Social Issues, 37, 71–94.
Cao, M. and Wei, J. (2005), “Stock market returns: A note on temperature anomaly”, Journal of Banking and Finance, 29(6), 1559-1573.
Cunningham, M.R. (1979), “Weather, mood and helping behavior: Quasi-experiment with the sunshine Samaritan”, Journal of Personality and Social Psychology, 37, 1947– 1956.
Fama, E. F. and French, K. R. (1993), “Common risk factors in the returns on stocks and bonds”, Journal of Financial Economics, 33(1), 3-56.
French, K. R. (1980), “Stock returns and the weekend effect.”, Journal of financial
economics, 8(1), 55-69.
Fama/French factors (2018), retrieved from
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html#HistBench marks
Hirshleifer, D. and Shumway, T. (2003), “Good Day Sunshine: Stock Returns and the Weather”, Journal of Finance, 58(3), 1009-1032.
Howarth, E. and Hoffman, M.S. (1984), “A multidimensional approach to the relationship between mood and weather”, British Journal of Psychology, 75, 15–23.
Jones, C. P., and Litzenberger, R. H. (1970), ”Quarterly earnings reports and intermediate stock price trends.”, Journal of Finance, 25(1), 143-148.
Policies and Measures: Germany (2010), retrieved from https://www.iea.org/policiesandmeasures/pams/germany/
21
Saunders, E. (1993), “Stock Prices and Wall Street Weather”, American Economic
Review, 83(5), 1337-1345.
Thaler, R. H. (1987), "Anomalies: The January Effect.", Journal of Economic
Perspectives, 1(1), 197-201.
Trombley, M. A. (1997), “Stock prices and Wall Street weather: Additional evidence”,
Quarterly Journal of Business and Economics, 11-21.
Appendices
Appendix 1: Important Policy Dates
Some important dates with respect to the German policy on renewable energy generation in the period 1995-2017 are displayed below. The main policy program is the
Erneuerbare-Energien-Gesetz, which was first implemented at 01-04-2000. The most important policy implications for
photovoltaic energy (PVE) generation are depicted in the implications column. The different dates will be utilized in our analysis on weather effects resulting from government incentives.
All data is collected from the International Energy Agency website (“Policies and Measures: Germany”, 2018).
Date Policy Implications
01-04-2000 Erneuerbare-Energien-Gesetz (EEG) Implementation of large scheme of feed-in-tariffs for all renewable energy generation. Remuneration of PVE plants capped at 350MW.
01-08-2004 EEG (2004) Large amendment of EEG. Increase in PVE tariffs and PVE plant cap.
01-01-2009 EEG (2009)* Tariff increases for all renewable energy sources. Only PVE tariffs slightly decrease.
01-07-2010 PV(2010) Large decrease in PVE tariffs.
22
Appendix 2: Fama-French three-factor model (total observations = 1.694.559).
Below, the output of the Fama-French three-factor model is displayed. The dependent variable is the daily return in percentages. In total, 399 equities were included in the regression.
Variables Coefficients
Constant (α) 0.136***
(0.008) Risk-free rate (rFt) -6.035***
(0.679) Market risk factor (rMt – rFt) 0.559***
(0.006) Size factor (smbt) 0.111***
(0.012) Value factor (hmlt) -0.170***
(0.013)
Notes: R-squared = 0.007, model’s F-statistic = 3065.16 (p =0.000). *, **, ***, statistically
23
Appendix 3: Renewable Energy Firms Included
In this table, all equities that are included in the analysis as renewable energy firms are shown. Also, the main source of renewable energy generation is noted in the second column. The term
mixed indicates that it actively generates energy from the sun, but also from at least one more
renewable source.
Stock Main Source for Energy Generation
7C SOLARPARKEN K Sun
BAYWA Biofuel
BAYWA REGISTERED Sun
BOSCH SOLAR ENERGY Mixed
CENTROTEC SUSTAINABLE Sun
CENTROTHERM PHTO. Biofuel
CROPENERGIES Wind
E ON N Mixed
ENBW ENGE.BADEN-WURTG. Mixed
ENCAVIS Mixed
ENERGIEKONTOR Water
GELSENWASSER Mixed
HELIOCENTRIS ENERGY SOLUTIONS Mixed
INNOGY Mixed LECHWERKE Mixed MAINOVA Sun MANZ Wind MVV ENERGIE Wind NORDEX Biofuel PETROTEC Sun
PHOENIX SOLAR Wind
PNE WIND Wind
RWE Biofuel
RWE PREF. Wind
S&O AGRAR Biofuel
SENVION Sun
SFC ENERGY Sun
SMA SOLAR TECHNOLOGY Sun
SOLAR FABRIK Biofuel
SOLARWORLD K Sun
