Bachelor Thesis Business Economics: Finance
Roos Bernelot Moens 5749670
Supervisor: Drs. P.F.A. Tuyp Coordinator: Dr. E. Schroth
Chapter I: Introduction
In the last few decades the traditional views in the field finance have been put up for discussion. Traditionally, economic concepts dominated the field of finance. Decision makers were assumed to act completely rational and in compliance with the expected utility theory. Thaler (1999) makes two assumptions in relation to this. The first being that when it comes to rationality, decisions are made according to the principles of expected utility theory. This theory states that the decision maker chooses between risky or uncertain prospects by carefully comparing their expected utility values (Mongin, 1997) and that, in doing this, the decision maker is risk averse. The second assumption holds that unbiased forecasts are made about the future. Outcomes are predicted using models based on theory and assumptions.
In his article, Olsen (1998) discusses the evolution of the traditional and standard science of finance towards the acceptance of the concept of behavioural finance. Supporters of the field of behavioural finance do not reject the traditional assumptions, but regard them as incomplete. They want to show the added value of taking into account behavioural and psychological factors in financial research and forecasts.
One of the arguments for the introduction of behavioural finance was its additional ability to explain stock-price volatility. Olsen (1998) defines behavioural finance in his article as a science that ‘seeks to detect systematic financial market implications of psychological decision processes’. This implies that the process of decision making and the behaviour of individuals working in the financial market have an effect on the financial market. According to Shefrin (2002) the behaviour of real human beings reflects emotion and errors in judgment as opposed to the idealized human beings modelled by game theory, who are rational when it comes to their preferences, judgments, and choice of strategies. This implies that the behaviour of these individuals should be taken into account when investigating financial markets.
In this paper I will study investor behaviour in the Netherlands. In particular I will look at the effect of the mood that decision makers on financial markets are in and the implications of their mood on their psychological decision processes. Putting this in a financial perspective, I will investigate whether mood is reflected in the stock returns on the AEX index.
Much psychological research has been done in the field of mood and behaviour. Among them are several studies pointing out the relationship between people’s moods and the local weather. Barnston (1988) found that people are more stressed when the weather is unstable, cloudy, warm and humid, and least stressed during sunny, dry, cool weather with rising barometric pressure. Howarth and Hoffmann (1984) also show that good moods are positively correlated to sunny weather and higher temperatures and negatively correlated with humidity and rainfall.
There is evidence that weather affects mood, but there are also numerous papers looking at the consequences of weather-induced behaviour on the financial markets. Saunders (1993) was one of the first to conduct such a research and found that stock returns were lower on cloudy days. Hirshleifer and Shumway (2003) also found that sunshine is strongly significantly correlated with stock returns.
In my thesis I will test for a relationship between the local weather circumstances in the city of Amsterdam and the returns on the AEX index. My research will have a time span of 28 years, ranging from the beginning of the AEX in 1983 up until the end of 2010. The hypotheses are as follows:
H0: The weather in Amsterdam does not systematically affect the returns on the AEX index.
H1: The weather in Amsterdam does systematically affect the returns on the AEX index.
The rest of the paper will be constructed in the following way. Chapter II gives an overview of the literature on the themes addressed in this paper. Chapter III will discuss the data used. In Chapter IV the method used to investigate the relationship between the weather and total returns will be discussed. In Chapter V the results will be analysed, and in Chapter VI my conclusions will be presented.
Chapter II: Literature
2.1 Standard Finance versus Behavioural Finance
In his article on the ongoing discussion between the traditional science called standard finance and the later emerged science of behavioural finance Statman (1999) states that it is a misconception that standard finance does not take into account behaviour of investors. In fact, it does take behaviour into account, but in a different, more simplistic, way. In standard finance and economics mental models of choice were assumed (Olsen, 1998), which are considered as models of choice based on completely rational information processing.
The article ‘Portfolio Selection’ by Harry Markowitz, published in 1952, is generally considered to be the start of standard finance. According to Markowitz the decision maker was assumed to be a homo economicus who was ‘completely rational and focused on utility (wealth) maximization’ (Markowitz, 1952). In practice the theory of the investor being a homo economicus is found to be somewhat idealistic, and not very realistic. It is what it is: a theory. These individuals are in fact, as Thaler (1999) puts it, ‘quasi-rational investors’ or ‘people who are trying as hard as they can to make good investment decisions but make predictable mistakes’.
The emergence of the study of behavioural finance was a result of the need to base financial theories on more realistic assumptions (Doyne Farmer and Lo, 1999). While in standard finance people are supposed to behave rationally, behavioural finance argues that people actually behave as emotional human beings. Their actions are not always a result of rational information processing. There is a lot more to it, according to the behaviourists. They argue that the theory of standard finance is behaviourally incomplete (Olsen, 1998). As Statman (1999) argues: ‘good theories of investor behaviour are crucial to good theories of asset pricing’. In traditional finance some issues concerning investing behaviour remain unaddressed. Questions about why individual investors trade, how they perform, how they choose their portfolios, and why returns vary across stocks for reasons other than risk are answered in behavioural finance (Subrahmanyam, 2007).
2.2 Mood, Behaviour, and Decision Making
The key theme in behavioural finance is decision making. It all comes down to what decision an investor will make when it comes to buying or selling stock. It is
important to understand what moved an investor to make that decision. People will not always behave rationally when it comes to decision making. The behaviour of real human beings reflects emotion and errors in judgment (Shefrin, 2002).
In the field of psychology, many articles have been written on the subject of mood in relation to information processing and the decision making process. Previous studies pointed out that negative moods stimulate a more analytical and critical approach towards decision making and positive moods enhance a more simple
approach in analyzing situations (Schwarz, 1991). Elsbach and Barr (1999) found that temporary mood states of decision-makers may have a significant influence on how they make complex decisions. They argue that, in fact, negative moods could be beneficial in these situations, because this stimulates a more careful, analytical and critical approach. Translating these findings to a financial perspective, one might argue that negative moods are linked to risk-averse behaviour. Investments are carefully thought through and decisions only made when risk of loss is minimized. Therefore, the volatility of the returns on stocks will be relatively low. In finance theory it is suggested that low risk investments lead to lower returns than high risk investments, otherwise no one would choose to invest in those high risk investments. On the other hand, the probability of losing is higher when it comes to high risk investments. If negative moods lead to higher risk aversion, individuals will choose the low risk strategy. Following the reasoning of Elsbach and Barr (1999), negative moods would thus lead to lower volatility in returns on stocks.
Good moods and happiness, on the other hand, tend to make people more optimistic in their decision making process (Wright and Bower, 1989). In their research they found that people in happy moods assign higher probabilities to positive events and lower probabilities to negative events than people in bad moods. They will act more optimistic and confident, and dare to take more risk when it comes to investing. A positive mood thus stimulates risk taking behaviour of investors. This, following the same reasoning as earlier but now the other way around, will lead to higher volatility of the returns on stock prices.
Summarizing, moods of decision makers on the financial markets affect their risk taking behaviour. Positive moods stimulate people to assess situations in a more optimistic way and make them more inclined to take risks. Negative moods tend to make individuals more critical and analytical, and more risk averse. This behaviour
will be reflected in the returns on stocks, because high risk means higher volatility and low risk means lower volatility in the returns.
2.3 Weather and Mood
Decisions and (risk-taking) behaviour are affected by mood. When it comes to science, the concept of mood is one to handle with care. Mood is not an easily measured variable and can be affected by many factors. Much research has been done in this area and the results and opinions vary widely. An interesting outcome, however, is the effect of weather on mood.
As scientific evidence has shown, the mood of people in one area is affected by the local weather circumstances. Barnston (1988) investigated the effect of weather on mood, productivity, and frequency of emotional crisis and found that people tend to be more stressed when the weather is unstable, cloudy, warm and humid, and least stressed during sunny, dry, cool weather with rising barometric pressure. Howarth and Hoffmann (1984) also show that good moods are positively correlated to sunny weather and higher temperatures and negatively correlated with humidity. In their research ten mood variables were related to eight weather variables. They found that humidity, temperature and hours of sunshine had the greatest effect on mood.
When it comes to this weather effect, however, one should take many other factors into account. Weather does in fact have an effect on mood, but there are issues to keep in mind when addressing this relationship. As Keller et al. (2005) have pointed out in their research, the season and time spent outside also play an important role when it comes to the relationship between mood and weather. They found that the relationship between mood and weather varies among different seasons. Where Howarth and Hoffmann (1984) argued that sunny weather and higher temperatures lead to better moods, Keller has a more thorough explanation. According to his research, increasing temperatures in summer time are in fact associated with lower moods. In spring, on the other hand, higher temperatures do seem to have a positive effect on moods.
Moreover, Keller (2005) found that the effect of the weather on people who spend most of their days indoors is almost just as strong as the weather effect on people who spend their days mostly outdoors. Investors, who spend most of their days working in an office or at the stock exchange, are thus equally affected by the weather as, for instance, the gardener, who spends his days mostly working outside.
2.4 Weather, Mood, and Stock returns
Following the earlier mentioned line of reasoning, if weather affects mood, then indirectly the weather will affect the behavior of people living in that particular area. Moreover, if sunny weather and higher temperatures indeed lead to positive moods, then risk taking behavior will be stimulated. Applying this to the financial market, investors will act more optimistic and less analytic. They will be inclined to take more risk, assign higher probabilities to positive events and lower probabilities to negative events. This would mean that these weather circumstances would lead to higher volatility in the returns on stocks.
Over the last decades, several studies have been performed investigating the relationship between the local weather and stock returns. One of the first to write about this subject was Saunders (1993) in his paper on the relationship between the local weather circumstances in New York City and the daily changes in indexes of listed stocks in New York City. He paired data on daily cloud cover in the city with data on stock price indexes. He discovered that in that area, the weather was significantly correlated with the major stock indices. This supported his hypothesis that investor psychology had an effect on asset prices.
In their paper, Hirshleifer and Shumway (2003) examined the relationship between the amount of sunshine during morning hours in the city of a country’s leading stock exchange and daily market index returns in 26 countries. Like Saunders, they also found that sunshine is strongly significantly correlated with returns. They also controlled for other weather circumstances, such as rain and snow, but those turned out to be unrelated to the returns.
Like Saunders and Hirshleifer and Shumway, many other were inspired to investigate this subject, each with their own approach, different weather variables or different areas. The results of these papers are mixed. Pardo and Valor (2003), looked for evidence in the Spanish market and did not find significant influence of sunshine hours and humidity levels on stock prices there. There results were consistent with the findings of Goetzmann and Zhu (2005), who were then working on their paper using data from five major cities in the US. They found that there was no difference in trading behavior on the stock exchanges on cloudy versus sunny days.
All these findings have inspired me, to conduct a study focused on the Amsterdam Exchange (AEX). In this paper I will use previous studies and combine the knowledge of others into a view of my own on the city of Amsterdam.
Chapter III: Data
In order to investigate the effect of the local weather on the AEX stock returns, data will have to be collected concerning these variables. Following the previous studies (Saunders, 1993; Hirshleifer and Shumway), daily data (5-day weeks) will be used. In my thesis I will use a sample period that ranges from the beginning of 1983 up until the end of 2010. The motivation for the use of this particular period is as follows. The AEX index was founded on January third of 1983 and data have been recorded ever since. I will use a sample of all data ranging from 1983 up until 2010, because this is the longest period in whole years that can be used. This way the largest amount of data will be obtained and statistically, inclusion of more data will lead to more accurate results.
As can be concluded from the literature, there is evidence that mood is affected by the weather circumstances. In turn, it was found that mood is of influence on the investing behaviour of people working in financial markets. Indirectly, the weather affects the decisions made on these stock exchanges and thus I expect to see variations in daily stock returns due to the weather circumstances. I will perform a regression analysis on the daily returns of the AEX index, using weather data as independent variables.
The data on the weather will be collected from the Koninklijk Nederlands
Meteorologisch Instituut (KNMI). This is the Royal Dutch Institute for Meteorology
and one of their tasks is to keep record of the daily weather circumstances. On several locations in the Netherlands weather stations collect data on weather on a daily basis. The data used are unprocessed data, which means that they are the raw data that are observed and recorded by the KNMI. I will carefully select the data needed in order to investigate the effect they have on the daily total return of the AEX index. The data on the total returns of the AEX will be obtained from DataStream. In this section the data used will be discussed in further detail. The descriptive statistics on the data are given in Table 1.
Table 1: Descriptive statistics of the data on the return on the AEX index (r), temperature (TG), sunshine (SP), and humidity (UG)
Return (r) on the AEX Index Temperature TG* Sunshine SP** Humidity UG*** Mean 0.000501 102.7411 35.08049 82.90130 Median 0.000540 104.0000 30.00000 84.00000 Maximum 0.118234 267.0000 102.0000 100.0000 Minimum -0.120009 -123.0000 0.000000 38.00000 Std. Dev. 0.013506 61.36047 29.64723 9.569526 Skewness -0.055969 -0.210196 0.393320 -0.752280 Kurtosis 11.76110 2.619854 1.836362 3.752417 Observations 7304 7305 7305 7305
* (in 0.1 degrees Celsius)
** (in percentage of maximum potential sunshine duration) *** (daily mean relative atmospheric humidity in percents)
Based on the descriptive statistics it can be concluded that 7305 observations were made. The return shows only 7304 observations, due to the loss of one day in calculating the returns relative to the previous day. These observations stand for daily recorded values, using 5-day weeks, over a period of 28 years (beginning of 1983 until end of 2010). There is a mean, positive return on the AEX Index of 0.0501%, with a maximum return of 11.8234%, and a minimum return of -12.0009%.
When it comes to the weather data, a mean temperature of 10.27 degrees Celsius is calculated. The highest average daily temperature recorded was 26.7 degrees Celsius, and the lowest average daily temperature was -12.3 degrees Celsius. The mean daily sunshine duration is 35.08% of the daily maximum potential sunshine duration. The maximum daily sunshine recorded is 100% of the potential sunshine duration, versus a 0% minimum on days when the sun did not come out at all. The daily mean relative atmospheric humidity level ranges from a minimum of 38% to a maximum of 100%. The mean humidity level is calculated at 82.90%. All these statistics concerning the weather do not show any unexpected values considering the Dutch climate.
3.1 Weather data
The weather data are obtained from the data centre of the Koninklijk Nederlands Meteorologisch Instituut (Royal Dutch Institute for Meteorology, KNMI)1. This institute has weather stations all over the Netherlands (see appendix figure 1), that continuously keep account of the weather. My study focuses on the returns of the Amsterdam Exchange Index, which is, as the name already suggests, located in the city of Amsterdam. In order to select the correct data, the station of choice will have to be the one that keeps track of the weather in the area of Amsterdam. The station used in my thesis is located at Schiphol airport. This station is closest to Amsterdam and therefore the best representative of the weather in that area.
As mentioned before I will use the weather conditions as defined by Howarth and Hoffmann (1984): sunshine, temperature, and humidity. From the list of data of the KNMI (see appendix figure 2), the variables that are selected will have to be the ones most compatible with those variables. All data from this database are daily values. For this thesis the data for the weekends were not included, in order to synchronize the data with the data on the AEX. The AEX data are given on a daily basis, using 5-day weeks.
When it comes to sunshine, there were two possible variables in the list of KNMI data that could be used. The first was SQ, which is defined as the (daily) sunshine duration (in 0.1 hour) calculated from global radiation (-1 for <0.05 hour). The other variable was SP, which is defined as the (daily) percentage of maximum potential sunshine duration. For my thesis, I chose to use SP as the representative of the amount of daily sunshine. This variable is clearly defined and I found it to be more easily interpreted than SQ. Moreover, SQ is calculated from global radiation, which seems irrelevant in this research, for here only the weather in Amsterdam is under consideration.
For temperature, there were several different data available, but only one that was useful. I chose TG, defined as average daily temperature in 0.1 degrees Celsius, as a variable. In this case, that is exactly the information that is needed. One might argue that if mood is influenced by temperature, than variations in temperature within one day will cause an effect mood within that one day. In this paper, however, these
variations are assumed to be too small to worry about and therefore considered negligible.
Finally, for humidity the variable UG is used, which is the daily mean relative atmospheric humidity (in percents). From the data collected on humidity, UG was the only variable that was representative of the one needed for my regression analysis.
3.2 AEX data
The independent variable in my thesis is the return on the AEX index. This index is a stock market index, composed of the 25 Dutch securities that trade most actively on the Euronext Amsterdam. The AEX index gives an indication of the developments in the whole stock market. It is thus a good representative of any effects caused by decisions made by investors operating on the Amsterdam Stock Exchange.
The AEX data on returns are obtained from DataStream. The variable that I will use in my thesis is the total return index, which includes dividends. The return on the AEX will be calculated as equation (1) shows:
rt = ( Pt - Pt-1 ) / Pt-1 = ( Pt / Pt-1 ) – 1, (1)
where rt is the return on the AEX index, Pt is the return index on the AEX as obtained from DataStream and Pt-1 is the return index on the AEX with a one-period lag.
As mentioned earlier, the weekends are removed from the sample period. These days are regarded as non-trading days and therefore not used in the regression analysis.
An assumption that has to be made in this thesis is that the investors that trade on the Amsterdam Exchange, actually live in the area. In their article on weather, stock returns and localized trading behaviour Loughran and Schultz (2004) demonstrate that trading has a strong local component. They found evidence that trading is concentrated among investors living near the company’s headquarters of the sample stocks used in their research. This supports the above assumption.
Focusing on the companies of the AEX index (see appendix figure 3), I found that 12 companies have their headquarters in the city of Amsterdam, 10 companies are within one hour of travel distance from Amsterdam, and 3 companies are located at a travel distance of over 2,5 hours. This means that, except for 3 companies, all companies are within a radius of approximately 60 kilometres of the Schiphol weather
station. It is helpful that the Netherlands is a small country and that there are no dramatic differences in climate within the country. Saunders (1993) also argues that those people who physically trade securities may affect security prices to a measurable degree. In my paper I make the assumption that those investors who trade AEX index stocks, live physically close to Amsterdam.
Chapter IV: Method
In order to find a relationship between the return on the AEX and the weather in Amsterdam, I will perform a regression analysis using the ordinary least squares regression method (OLS). In this regression I will pair the data on AEX returns with the data on the local weather circumstances. The total return on the AEX index will be the dependent variable. The explanatory variables will be the daily weather data that were found to be of greatest influence, as mentioned by Howarth and Hoffmann (1984) and as selected from the database of the KNMI: sunshine hours SP, temperature TG, and humidity UG. The regression performed on the AEX will be as follows:
rAEX,t = β0 + β1SPt + β2TGt + β3UGt + εt (2)
By performing this regression, I will test the following hypothesis:
H0: The weather in Amsterdam does not systematically affect the returns on the AEX index.
H1: The weather in Amsterdam does systematically affect the returns on the AEX index
Following the outcomes of previous performed studies on this subject, my expectations are that the returns on the AEX will be higher when the weather conditions are good, that is, when the amount of sunshine and the temperature are higher. On the other hand, returns will be lower when humidity is higher.
As for the sample period, first I will look at the outcomes for the entire sample, ranging from January 1st 1983 up until December 31st 2010. The data cover 5-day weeks, leaving the non-trading weekend days out of the regression.
Another issue that will be addressed will be the important role internet has started to play in stock trade. Over the last two decades internet has penetrated our lives and has made markets increasingly more global. Ever since 1994 at least 4 new equity trading systems (TSA, SWX, SETS and XETRA) have been introduced in Europe, which are all electronic order matching systems (Demarchi and Foucault, 2000). This development might have an influence on the effect of the weather on the stock market, for it means that more trade is performed via these electronic systems. I will test whether these developments have any implications on the weather effect on
stock returns. After testing for an overall relationship between the weather and stock returns over the entire period ranging from 1983 up until 2010, I will compare two sub samples. Due to the introduction of the electronic order matching systems around 1994 (Demarchi and Foucault, 2000), I expect that the relationship between weather and stock returns will be less strong after 1995. The first sub sample will therefore range from 1983 up until the end of 1995, and the second will range from 1996 up until the end of 2010. The outcomes of these two sub samples will be compared.
A third regression will be performed to test for the findings of Keller et al. (2005). Their conclusion that in spring people react in a more positive way to better weather circumstances is very interesting. In my research I will investigate whether the weather effect is stronger in spring, as opposed to the summer season. I expect the relationship to be stronger in spring.
When performing a regression analysis several requirements have to be met. In the following sections, these requirements will be explained en tested.
5.1 Multicollinearity
Because the weather variables are closely related, I will test for multicollinearity. If the variables are highly correlated with one another, multicollinearity exists. Movements in one variable will be extremely similar to movements in another explanatory variable, and therefore it will be hard to distinguish the one variable from the other. A solution to this problem would be to remove one of the variables from the regression.
The correlation matrix of the weather variables is given in table 2.
Table 2: Correlations between the weather variables and their probabilities
Correlation SP TG UG SP 1.000000 (---) TG 0.197728 (0.0000) 1.000000 (---) UG -0.626631 (0.0000) -0.346432 (0.0000) 1.000000 (---)
Problems concerning multicollinearity would arise when the correlations would be close to 1 or -1. In this case, no problematic correlation between the explanatory variables is found in the regression. The probabilities of the correlations are given between parentheses, and show that all values of the correlations are significant.
5.2 Heteroskedasticity
In the regression analysis the error term is required to have a constant variance. This can be tested using the White test for heteroskedasticity. In this test under the H0 homoskedasticity of the error term is tested against the alternative hypothesis (H1) of heteroskedasticity. If heteroskedasticity exists, the OLS-estimator is no longer efficient. In that case, White standard errors have to be used.
The test statistic of the White test is equal to the R2 multiplied by the number of observations in the sample and amounts to 10.72320. This test follows a Chi-squared distribution and the p-value of the test statistic is equal to 0.2952. Comparing this value to a significance level of 5% (0.05) the H0 of homoskedasticity is not rejected. It would not even be rejected at a significance level of 10% (0.10). I can assume that there is no significant heteroskedasticity of the error term.
5.3 Autocorrelation
For the OLS estimator to be consistent there should be no residual autocorrelation. This is tested using the Breusch-Godfrey test for autocorrelation. In the Breusch-Godfrey test the H0, where no autocorrelation is hypothesized, is tested against the H1 of autocorrelation. When autocorrelation is detected, the OLS-estimator will no longer be efficient and consistent. In that case, corrected Newey-West standard errors have to be used.
The test statistic of the Breusch-Godfrey test is equal to the R2 multiplied by the number of observations in the sample and amounts to 2.330673. This test follows a Chi-squared distribution and the p-value of the test statistic is equal to 0.3118. Comparing this value to the significance level of 5% (0.05) the H0 is not rejected. Again, even when a 10% significance level is used, the H0 will not be rejected. I can assume that there is no significant residual autocorrelation.
As the results of the previous tests point out, the requirements are met by the model. No problems were encountered in the model specification procedure.
V. Results
In this study, I will first test for a relationship between the weather and the stock returns over the total sample, from 1983 to 2010. Next I will do a second regression, dividing my sample into two sub samples, to look for any differences in the effect due to the introduction of electronic trading systems in the nineties. Finally, I will compare the weather effect on the AEX returns in spring with the effect it has in summer.
5.1 Total Sample (1983 – 2010) The model used here is as follows:
rAEX,t = β0 + β1SPt + β2TGt + β3UGt + εt. (3)
As mentioned earlier, my expectations are that increasing temperature and sunshine hours are positively related to the stock returns. Increasing humidity, on the other hand, I expect to be negatively related to stock returns. These expectations follow the findings of Howarth and Hoffmann (1984), who have provided evidence that good moods are positively correlated to sunny weather and higher temperatures and negatively correlated with humidity. The results of the regression analysis are represented in table 3.
When it comes to sunshine (SP), the results are in line with my expectations and the coefficient significantly differs from zero. An increase in the amount of daily sunshine causes an increase in the returns on the AEX. The temperature estimate, however, does not follow my expectations. The coefficient differs significantly from zero (at a 5 percent level), but according to the coefficient sign, higher temperatures lead to lower stock returns. Humidity does not show any significant correlation with the AEX returns.
Using the general-to-specific modelling method the above model is simplified by leaving the humidity variable out of the regression. The results are shown in column (3) in table 3. There are no dramatic differences in the effects of sunshine and temperature in this model in comparison with the more general model.
Table 3: Regression output total sample
The dependent variable is the return on the AEX Index. Standard errors are reported in parentheses. General-to-specific modelling is applied in column (3).
Variable (2) (3) C -0.001492 0.000613 (0.002090) (0.000337) SP 2.14E-05* 1.72E-05* (6.84E-06) (5.43E-06) TG -6.13E-06** -6.95E-06* (2.74E-06) (2.63E-06) UG 2.26E-05 - (2.21E-05) - Number of observations 7305 7305 Adjusted R-squared 0.001678 0.001673
* significant at the 1 percent level ** significant at the 5 percent level *** significant at the 10 percent level
5.2 Sub samples
In this section the regressions performed have the same form as regression (2). Now the sample has been divided into two smaller sub samples to control for any differences between the two periods. The breaking point between these two sub samples will be the period when electronic trading systems were introduced which is halfway through the nineties.
My expectations are that, due to the introduction of the electronic trading systems, the effects of the weather on investing behaviour will be less strong. Trading will be less centralized and more dispersed and therefore less dependent of the weather. Furthermore, my expectations of the effects and signs of the variables will be similar to the expectations as described for the entire sample period.
5.2.1 Sub sample 1 (1983 – 1995)
In this section, a regression is performed on the first sub sample. The results are given in table 4.
No significant relationship is found between any of the weather variables and the AEX returns. This is not in line with my expectations that a (stronger) relationship would exist before the introduction of electronic trading systems.
The general-to-specific modelling method is applied in columns (5) and (6) to see if there is any effect when the least significant variable is left out of the regression. In this case the least significant variable is UG and the most significant variable is TG. Subsequently UG and SP were left out of the regression. No significant correlation was found after applying the general-to-specific modelling method.
Table 4: Regression output of the sub sample 1
The dependent variable is the return on the AEX Index. Standard errors are reported in parentheses. General-to-specific modelling is applied in columns (5) and (6).
Variable (4) (5) (6)
C -0.000236 0.000870 0.001057
(0.002474) (0.000392) (0.000363)
SP 1.04E-05 8.16E-06 -
(8.09E-06) (6.46E-06) -
TG -4.48E-06 -4.78E-06 -3.91E-06
(3.27E-06) (3.20E-06) (3.13E-06)
UG 1.19E-05 - -
(2.62E-05) - -
Number of observations 3390 3390 3390
Adjusted R-squared 0.000107 0.000341 0.000165
* significant at the 1 percent level ** significant at the 5 percent level *** significant at the 10 percent level
5.2.2 Sub sample 2 (1996 – 2010)
In this section, a regression is performed on the second sub sample. The results are given in table 5.
In the regression on the second sub sample a significant relationship is found between sunshine and the AEX returns, but not for the other weather variables. This is not in line with my expectations. I expected to find a stronger relationship between the weather and the AEX returns in the first sub sample, as opposed to the second sub sample. The reverse is true, however. The only significant effect found is between the sunshine variable and the AEX returns in the second sub sample.
The second sub sample is modelled more specifically by applying the general-to-specific modelling method. Again, the humidity variable UG is left out of the regression. The results are given in column (8) in table 5. In this case, even the temperature variable is significant (at a 5 percent level), as well as the sunshine
variable. This is a remarkable outcome and not in line with my expectations whatsoever. These results suggest that there exists a significant relationship between the weather variables sunshine and temperature after the introduction of electronic trading systems. No significant relationship exists, however, before the introduction of electronic trading systems.
Table 5: Regression output of the sub sample 2
The dependent variable is the return on the AEX Index. Standard errors are reported in parentheses. General-to-specific modelling is applied in column (8).
Variable (7) (8) C -0.002169 0.000315 (0.003344) (0.000538) SP 3.05E-05* 2.55E-05* (1.08E-05) (8.50E-06) TG -7.22E-06 -8.38E-06** (4.30E-06) (4.02E-06) UG 2.67E-05 - (3.55E-05) - Number of observations 3915 3915 Adjusted R-squared 0.002306 0.002417
* significant at the 1 percent level ** significant at the 5 percent level *** significant at the 10 percent level
5.3 Seasonal differences
In this section a regression is performed to compare the weather effect on the AEX returns between spring with the effect in summer. According to Keller et al. (2005), increasing temperatures in summer time are found to have a negative effect on moods. In spring, on the other hand, higher temperatures do seem to have a positive effect on moods.
My expectations are that, due to this effect, higher temperatures will lead to higher stock returns in spring. In summer, however, higher temperatures will lead to lower stock returns. In order to investigate this relationship the following model is used:
rAEX,t = β0 + β1TGt*SPRDUM + β2TGt *SUMDUM + εt, (4) where SPRDUM is a dummy variable for the spring season, which is 1 for the months of April, May and June and 0 otherwise. SUMDUM is a dummy variable for the
summer season which is 1 for the months of July, August and September and 0 otherwise. The results are given in table 6.
No significant effects are found in the regression. After applying general-to-specific modelling, however, a significant correlation (at a 10 percent level) is found between temperature in summer and the stock returns. Following my expectations, higher temperatures in summer lead to lower returns on the stocks of the AEX Index. No relationship was found, however, between spring temperature and returns on the AEX.
Table 6: Regression output with dummies for the spring and summer season The dependent variable is the return on the AEX Index. Standard errors are reported
in parentheses. General-to-specific modelling is applied in column (10).
Variable (9) (10)
C 0.000624 0.000655
(0.000217) 0.000182
TG*SPRDUM 7.54E-07 -
(2.86E-06) -
TG*SUMDUM -3.49E-06 -3.67E-06***
(2.25E-06) (2.14E-06)
Number of observations 7305 7305
Adjusted R-squared 0.000137 0.000264
* significant at the 1 percent level ** significant at the 5 percent level *** significant at the 10 percent level
Chapter VI: Conclusion
Psychological science has provided evidence that there exists a relationship between the condition of the weather and the moods of individuals in one area. The objective of this paper was to investigate whether there was a relationship between the local weather circumstances in the City of Amsterdam and the behaviour of investors trading on the AEX. If such a relationship were to exist it was hypothesized that this would be reflected in the returns on the AEX index. A regression analysis was performed on the return on the AEX index and three weather variables, of which there was scientific evidence to be of strong influence on moods. The variables chosen were sunshine (SP), temperature (TG) and humidity (UG).
First I performed a regression on the total sample, including all weather variables. A significant relationship was found between sunshine and temperature and the returns on the AEX returns, but not for humidity. As opposed to my expectations, however, higher temperatures do not lead to higher stock returns, but in fact to lower stock returns. An explanation for this fact could be found in Keller’s (2005) research. In this paper it was argued that in summer higher temperatures lead to lower stock returns. The regressions done in table 6 support these findings. It does not explain, however, why this effect would also count for other seasons. The results of my research suggest that temperature is mostly of influence on stock returns in summer, but this remains to be investigated.
After performing this regression, two additional regressions were performed on two sub samples. The motivation for this was the introduction of electronic trading systems in the nineties. It was expected that the relationship between the weather and the stock returns would be less strong after this event. Electronic trading systems have made the stock exchange less location specific and therefore the weather effect was expected to diminish. After comparing the outcomes of these regressions however, no significant relationship was found in the first sub sample at all. The second sub sample, on the other hand, did show a significant correlation between sunshine and the returns on the AEX. After applying the general-to specific modelling method, a significant relationship was also found for temperature. Here, again, an increasing amount of sunshine has a positive effect on the stock returns and increasing temperatures have a negative effect on the stock returns. As mentioned earlier, this is not in line with my expectations concerning the temperature, but is an interesting outcome that remains to be further looked in to. As these outcomes suggest, the fact
that electronic trading systems were introduced led to an increase of the relationship between the weather variables of sunshine and temperature and the returns on the AEX. Following this outcome, one might argue that while investors are stuck to their desks and computers, they might have become more aware of the daily weather circumstances and their valuations of these weather circumstances might have changed accordingly. This is an interesting matter and in order to fully comprehend the situation, further research is required.
Finally, the findings of Keller et al. (2005) were incorporated in my studies. A regression was performed to see whether there was a difference in the effect of temperature on stock returns between the spring season and the summer season. In this regression dummy variables for the spring season and the summer season were created. No significant relationship was found for both seasons, but after applying the general-to-specific modelling method and eliminating the spring dummy from the equation, a significant relationship was found between temperature in summer time and the returns on the AEX returns. As I expected, in this case, higher temperatures in summer lead to lower stock returns. One might expect that the effect of temperature on stock returns only is significant in the summer season and has no effect during other seasons. Further research will have to be done for this assumption to hold.
My conclusion from this research is that investor behaviour, and thus the returns on the AEX Index stocks, are affected by the weather, albeit not by all weather elements. Where humidity has no effect at all, sunshine has a positive effect on the stock returns, and temperature a negative effect. The latter two effects seem contradictory, because sunshine is often linked to higher temperatures. But it does make sense that too high temperatures could have a negative effect on mood. This could be explained via the seasonal differences in the temperature effect on the returns on the AEX Index.
The results imply that there is evidence that investors do not behave completely rational when it comes to decision making in the investing process. Their moods are affected by the weather, which in turn affects their investing behaviour. It has to be said, though, that the weather only explains a small part of the behaviour of individuals. Many factors are of influence when it comes to behaviour and future research is the key to the mysteries of the human mind.
Chapter VII: Bibliography
Barberis, N. and Richard Thaler (2002). A Survey of Behavioural Finance. National
Bureau of Economic Research, Working paper, University of Chicago.
Barnston, A.G. (1988). The effect of weather on mood, productivity, and frequency of emotional crisis in a temperate continental climate. International Journal of
Biometeorology, vol 32, pp.134-143.
Demarchi, Marianne and Thierry Foucault (2000). Equity Trading Systems in Europe: A Survey of Recent Changes. Annals of Economics and Statistics, No. 60, pp. 73-115.
Doyne Farmer, J. and Andrew W. Lo (1999). Frontiers of Finance: Evolution and Efficient Markets. Proc. Natl. Acad. Sci, Vol. 96, pp. 9991-9992, (From the Academy).
Goetzmann, William N. and Ning Zhu (2005). Rain or Shine: Where is the Weather Effect? European Financial Management, Vol. 11, No. 5, pp. 559-578.
Hirshleifer, D. and Tyler Shumway (2003). Good Day Sunshine: Stock Returns and the Weather. Journal of Finance, vol. 58, no. 3, pp. 1009-1032.
Howarth, E. And M.S. Hoffmann (1984). A Multidimensional Approach to the Relationship between Mood and Weather. British Journal of Psychology, 75, pp. 15-23.
Jacobsen, B. and Wessel Marquering (2008). Is it the Weather? Journal of Banking
and Finance, vol. 32, no. 4, pp. 526-540.
Keller, Matthew C. et al. (2005). A Warm Hart and a Clear Head: The Contingent Effects of Weather on Mood and Cognition. Psychological Science, Vol. 16, No. 9, pp. 724-731.
Kotharia, S.P., Jonathan Lewellen and Jerold B. Warner (2006). Stock Returns, Aggregate Earnings Surprises, and Behavioural Finance. Journal of Financial
Economics, vol. 78, pp. 537-568.
Mongin, Phillipe (1997). Expected Utility Theory. Handbook of Economic
Methodology, pp. 342-350.
Olsen, Robert A. (1998). Behavioral Finance and Its Implications for Stock-Price Volatility. Financial Analysts Journal, Vol. 54, No. 2, pp. 10-18.
Pardo, A. and Enric Valor (2003). Spanish Stock Returns: Where is the Weather Effect? European Financial Management, vol. 9, no. 1, pp. 117-126.
Saunders Jr., Edward M. (1993). Stock Prices and Wall Street Weather. The American
Economic Review, vol. 83, no. 5, pp. 1337-1345.
Schwarz, Norbert and H. Bless (1991). Happy and Mindless, But Sad and Smart? The Impact of Affective States on Analytic Reasoning. J. Forgas, ed. Emotion and
Social Judgments. Pergammon Press, Oxford, England, pp. 55-71.
Shefrin, Hersh (2002). Behavioural Decision Making, Forecasting, Game Theory, and role-play. International Journal of Forecasting, 18, pp. 375–382.
Statman, Meir (1999). Behavioral Finance: Past Battles and Future Engagements.
Financial Analysts Journal, Vol. 55, No. 6, pp.18-27
Subrahmanyam, Avanidhar (2007). Behavioural Finance: A Review and Synthesis.
European Financial Management, vol. 14, no. 1, pp. 12-29.
Thaler, Richard H. (1999). The End of Behavioural Finance. Financial Analysts
Journal, vol. 55, no. 6, pp. 12-17.
Wright, William F. and Gordon H. Bower (1989). Mood effects on subjective
probability assessment. Organizational Behavior and Human Decision
Appendix
Figure 1: Overview of all weather stations in the Netherlands (knmi.nl)
Number Location Number Location
210 Valkenburg 280 Eelde
225 IJmuiden 283 Hupsel
235 De Kooy 286 Nieuw Beertha
240 Schiphol 290 Twenthe 242 Vlieland 310 Vlissingen 249 Berkhout 319 Westdorpe 251 Hoorn (Terschelling) 323 Wilhelminadorp
257 Wijk aan Zee 330 Hoek van Holland
260 De Bilt 340 Woensdrecht 265 Soesterberg 344 Rotterdam 267 Stavoren 348 Cabauw 269 Lelystad 350 Gilze-Rijen 270 Leeuwarden 356 Herwijnen 273 Marknesse 370 Eindhoven 275 Deelen 375 Volkel 277 Lauwersoog 377 Ell 278 Heino 380 Maastricht 279 Hoogeveen 391 Arcen
Figure 2: List of description of weather elements recorded by KNMI (knmi.nl) Element Description Element Description
DDVEC Vector mean wind direction in degrees (360=north, 90=east, 180=south, 270=west,
0=calm/variable)
DR Duur van de neerslag (in 0.1 uur)
FHVEC Vectorgemiddelde windsnelheid (in 0.1 m/s)
RH Etmaalsom van de neerslag (in 0.1 mm) (-1 voor <0.05 mm)
FG Etmaalgemiddelde windsnelheid (in 0.1 m/s)
RHX Hoogste uursom van de neerslag (in 0.1 mm) (-1 voor <0.05 mm) FHX Hoogste uurgemiddelde windsnelheid (in 0.1 m/s) RHXH Uurvak waarin RHX is gemeten FHXH Uurvak waarin FHX is gemeten PG Etmaalgemiddelde luchtdruk herleid tot zeeniveau (in 0.1
hPa) berekend uit 24 uurwaarden
FHN Laagste uurgemiddelde windsnelheid (in 0.1 m/s)
PX Hoogste uurwaarde van de luchtdruk herleid tot zeeniveau (in 0.1 hPa) FHNH Uurvak waarin FHN is
gemeten
PXH Uurvak waarin PX is gemeten
FXX Hoogste windstoot (in 0.1 m/s)
PN Laagste uurwaarde van de luchtdruk herleid tot zeeniveau (in 0.1 hPa) FXXH Uurvak waarin FXX is
gemeten
PNH Uurvak waarin PN is gemeten
TG Etmaalgemiddelde temperatuur (in 0.1 graden Celsius)
VVN Minimum opgetreden zicht
graden Celsius) gemeten
TNH Uurvak waarin TN is gemeten VVX Maximum opgetreden zicht TX Maximum temperatuur (in 0.1
graden Celsius)
VVXH Uurvak waarin VVX is gemeten
TXH Uurvak waarin TX is gemeten NG Etmaalgemiddelde bewolking (bedekkingsgraad van de bovenlucht in achtsten, 9=bovenlucht onzichtbaar) T10N Minimum temperatuur op 10
cm hoogte (in 0.1 graden Celsius)
UG Etmaalgemiddelde relatieve vochtigheid (in procenten)
T10NH 6-uurs tijdvak waarin T10N is gemeten
UX Maximale relatieve vochtigheid (in procenten) SQ Zonneschijnduur (in 0.1 uur)
berekend uit de globale straling (-1 voor <0.05 uur)
UXH Uurvak waarin UX is gemeten
SP Percentage van de langst mogelijke zonneschijnduur
UN Minimale relatieve vochtigheid (in procenten) Q Globale straling (in J/cm2) UNH Uurvak waarin UN is gemeten
EV24 Referentiegewasverdamping
Fig. 3 AEX index: companies
Stock’s name Market
AEGON AMS
AHOLD KON AMS
AIR FRANCE – KLM PAR,AMS
AKZO NOBEL AMS
APERAM AMS,PAR ARCELORMITTAL AMS,BRU,PAR
ASML HOLDING AMS
BOSKALIS WESTMIN AMS
CORIO AMS
DSM KON AMS
FUGRO AMS HEINEKEN AMS
ING GROEP AMS,BRU
KPN KON AMS
PHILIPS KON AMS
POSTNL AMS RANDSTAD AMS
REED ELSEVIER AMS
ROYAL DUTCH SHELLA AMS
SBM OFFSHORE AMS
TNT EXPRESS AMS
TOMTOM AMS UNIBAIL-RODAMCO PAR,AMS
UNILEVER AMS
