Road accidents: unexpected costs of stock market movements

Abstract

Recent interest has grown in the link between stock market returns and health conditions. We extend this literature to road accidents as changes in returns may affect anxiety and stress, leading to driver fatigue and distraction, resulting in road accidents. Using Stats19 administrative data on accidents, we investigate the relationship between FTSE100 returns and accidents in British regions from 2008 to 2019. Accidents respond positively to decreases and increases in returns with effects up to 1.2%. Daily returns can be large, having substantial effect on accidents when returns are large. Compared with US results, we find no effects for fatal accidents. This could result from the UK’s very low fraction of fatalities and differences in road infrastructure, speed, and congestion. Unanticipated changes in returns represent an exogenous shock to individuals which may causally affect driving behaviour, and this is important to road safety stakeholders and for health promotion and policy.

1. Introduction

This paper investigates the causal relationship between stock market movements and road traffic accidents. Stock market fluctuations are typically measured against financial gains and losses. However, in the wake of the Global Financial Crisis (GFC), there has been growing interest in the relationship between stock markets and wellbeing with a focus on physical and mental health conditions associated with financial losses. Most papers find health decreases as stock market performance weakens—see, for example, Chen et al. (2012) (who look at strokes), Fiuzat et al. (2010) (who consider heart attacks), and Lin et al. (2017) (who examine attempted suicide).

One aspect of health that has received little attention is the impact of the stock market on road accidents. Stock market fluctuations could affect traffic accidents through different mechanisms. It is plausible that there could be effects through health behaviours (such as alcohol consumption) and/or psychological factors (such as driver distraction). The purpose of this paper is to establish if there is a causal relationship between stock market crashes and traffic accidents. Given that stock market crashes are well publicized and have become an important financial indicator, even people without direct involvement in these markets may still be impacted through the wider economic uncertainty signals that stock market movements indicate. Directly relevant to our purpose are the two studies considering fatal car accidents and the stock market, namely Cotti et al. (2015) and Giulietti et al. (2020). However, only looking at the impact of stock markets on road traffic deaths leaves us with a partial picture as fatal accidents are typically a small proportion of total accidents. Driver behaviours and psychological distractors are known to be contributory factors for nonfatal as well as fatal accidents.

Cotti et al. (2015) examined monthly movements in stock prices and crash indicators (1987 and 2008–2009) for US states for 1984–2010 and found reductions in the Dow Jones Industrial Average (DJIA) and stock market crashes were associated with increases in fatal accidents involving alcohol. Estimates suggest the 2008–2009 stock market crash was associated with a 5.9% increase in alcohol-related fatal car accidents. Also, based on levels, a 10% reduction in the DJIA was associated with a 1.3% increase in such accidents and months with a greater than 10% reduction in returns showed a 5.3% increase in fatal accidents.

Similarly, Giulietti et al. (2020) investigated the relationship between the stock market and fatal accidents in the US for 1990–2015. They found a one standard deviation reduction in stock market returns (measured by the Standard and Poor’s 500) was associated with a 0.6% increase in the number of fatal accidents. The authors argue that their results are consistent with emotions responding to poor stock market performance. They also found the effect on accidents was asymmetric in that only the lowest tercile of returns (negative) had a significant coefficient. To investigate heterogeneity, Giulietti et al. (2020) also estimate the effect of returns on groups of drivers with differential attachment to the stock market—with low attachment proxied by drivers aged 25 years or under, living in areas with lower incomes, and driving car makes associated with lower stock market exposure (based on survey data). We would expect the effects to be weaker for those with less involvement in the stock market and this is what the authors find.

In this paper, we extend current knowledge of the impact of the stock market on road traffic accidents. Our study is important and innovative in the following ways. Firstly, a significant point of differentiation of our study is the quality of the dataset employed. Our main analysis covers some 12 years and comprises daily observations on every road accident reported to police in Britain. We use Britain’s headline stock index—the FTSE100—as the measure of the stock market given that this index is the most widely reported. Stock market data are available at high frequency (daily) and their availability over the 12 years allows us to investigate the short-run relationship with road accidents over this long time series. Secondly, we will adopt a tightly identified model that considers movements in accidents within a month and region within Britain using daily data and this allows for a more nuanced approach than the extant literature. Thirdly, using daily data represents an improvement over the monthly data used by Cotti et al. (2015). As emotional responses are reactionary, it is important for the identification to have high frequency data on both the stock market and accidents. Fourthly, we estimate a model specification that allows for asymmetric effects of (continuous) positive and negative returns, which represents an innovation over the measures of returns adopted by Giulietti et al. (2020). We then extend this concept further and allow for different effects according to the size of the stock market movement. Finally, our application to total injury road accidents creates knowledge beyond fatal accidents which have been studied to date.

2. Background

Driving is common to the majority of the adult population and road accidents represent a major public health issue (Giulietti et al., 2020). Indeed, in 2016, road accidents ranked as the top 8th cause of death globally (up from rank 10 in 2000).¹ In Britain, in 2015, the average value of road accident prevention was approximately £76,000, comprising £2 m for fatal accidents, £230,000 for serious, and £24,000 for slight accidents. These costs comprise loss of output, ambulance and hospital treatment costs, and human costs (willingness to pay) (UK Department for Transport, 2016). So, there is a large and potentially avoidable burden on the healthcare system. The relationship between road accidents and fatalities is important to individual health and hence to ‘government agencies, insurance companies, the police, and other parties seeking to understand and predict road safety outcomes’ (Burke and Teame, 2018, p. 150) and deliver better health outcomes for their citizens. It is therefore important to understand aspects of our daily lives that impact on driving behaviours. Here, we focus on the extemporaneous relationship between stock market movements and road accidents to ensure public safety and better health outcomes.

There are several rationales for why the stock market might impact on road accidents. Firstly, stock market movements are a real-time indicator (‘news’) of how the economy is performing. Such movements also impact on emotions (Ma et al., 2011; Lin et al., 2013). Changes in the stock market affect both stock holders and non-stock holders through uncertainty and wealth channels (Angrisani and Lee, 2016). Evidence suggests most individuals are aware of economic uncertainty (Kalcheva et al., 2017) and changes in the stock market can lead to short-run anxiety about future prospects (Frijters et al., 2015), even for non-stock holders. Changes in the stock market affect stock holders directly through changes in portfolio values (Engelberg and Parsons, 2016) and stock market losses may result in major losses in wealth/savings (Haw et al., 2015). Sudden wealth losses might negatively affect health due to loss aversion (Chen et al., 2016) although the size of the effect is likely to vary with the degree of exposure to the stock market.² For example, effects for stock holders might be more muted than expected if there is indirect stock holding through retirement, mutual funds, or pensions (Cotti et al., 2015) or if stocks are managed through financial advisors and evaluations of market performance are limited to reading personal financial statements (Lin et al., 2013). Inexperienced investors may be particularly vulnerable to changes in the stock market through overreaction to adverse returns in their portfolios and therefore at higher risk of negative health outcomes including road accidents (Giulietti et al., 2020). While there is some existing literature for the US on the link between stock market movements and road accidents, we might expect smaller effects for British regions as there has been a lower involvement in the stock market relative to the US. For example, based on survey data, participation in the US stock market was estimated at 48% in 1998 and 40% in 1999 whereas participation in the UK stock market was estimated at 34% in 1998 and 26% in 1999 (Banks et al., 2002; Guiso et al., 2003). More recently, a 2020 survey showed 33% of Britons owned shares (Finder.com, 2020), whereas more than 50% of US households were invested in the stock market.³

Psychological distress resulting from large changes in one’s financial position as the stock market fluctuates could distract individuals, reducing their concentration while driving and leading to erroneous and dangerous driving behaviours causing accidents (Vandoros et al., 2014; Giulietti et al., 2020). ‘Stress about one’s finances also increases the likelihood of having a serious accident’ (Vandoros et al., 2014, p. 557). McInerney et al. (2013) found increased treatment for depression among individuals with large stock holdings resulting from the 2008 stock market crash.

The pathway through emotional responses also suggests that individuals may be affected by both negative and positive stock movements. Positive psychology allows us a direct way to understand the impact on positive feelings of excitement, happiness, and increased life satisfaction (Seligman, 2004) which may also result in driver distraction. To understand the full impact of the stock market on accidents, we need to examine the impact on both the negative and positive psychology of drivers. Moreover, while negative and positive emotional responses to stock market movements are likely to be correlated it is not the case that there is prefect correlation, that is, it is not the case that the impact of stock market improvements on positive emotions will be quantifiably the same as an equivalent stock market depreciation on negative emotions. For example, see Zheng (2016) for a study of the correlation between subjective wellbeing and depression, and Farrell (2018) for an analysis of happiness and gambling addiction. It is also possible that while psychology tells us that positive outcomes lead to emotional responses, these might not translate into accidents as the psychological impact may be too small. Psychology recognizes negative bias which states that events of a negative nature have a greater effect on an individual’s psychological state than events of an equal positive nature (Kanouse and Hanson, 1972). Negative potency also suggests that there is a steeper negative gradient than positive (Rozin and Royzman, 2001). Indeed, consistent with the economic literature’s ideas of loss aversion (Kahneman and Tversky, 1979), we might expect that individual emotional responses will be greater for falls in returns relative to equivalent sized gains. The literature on attention also suggests that there may be bigger impacts for negative events than to equivalent sized positive events as negativity attracts greater psychological focus resulting in greater attention (Fiske, 1980). It is therefore important to allow for a quantifiable differential effect of upwards and downwards stock market movements on psychological responses and to understand the differential responses according to the size of the movement.

Stock market movements may also affect accidents in terms of behavioural responses. In the face of a stock market crash, present bias means individuals might ‘substitute towards consuming immediately pleasurable goods [such as alcohol] to alleviate declining wellbeing that arises in the face of a bleaker future’ (Cotti et al., 2015, p. 819). Emotional stress associated with adverse stock market movements may also lead to risky behaviours such as binge drinking (Yap et al., 2016). Linking with the stock market, a study by Cotti et al. (2015) showed an increase in binge drinking and alcohol-related road fatalities during the 2008–09 financial crisis.

Based on this evidence from the existing literature, we examine the hypothesis that there is a causal relationship between stock market movements and road traffic accidents (H1). We expect that poor performance of the stock market leads to changes in health conditions such as anxiety and stress (Schwartz et al., 2012), which may lead to driver fatigue and distraction (Giulietti et al., 2020), and health behaviours (such as drinking alcohol) which then lead to changes in driver behaviours that lead to more accidents. The purpose of this paper is to establish the nature of any causal relationship (H1). Individuals might respond differently to financial gains than losses so we examine both positive and negative changes in the stock market (H2). While the existing literature focuses mainly on the impact of stock market falls, psychological responses are a complex mixture of both negative and positive emotions. Finally, we hypothesize that emotions, and therefore accidents, will respond more to large changes in returns relative to small changes in returns (H3).

Given that the existing literature only considers fatal accidents, there is a potential to underreport the total effect. In our data, only about 1.4% of accidents are fatal. Moreover, fatal accidents are a non-random subset of total accidents, more likely to be caused by high speed and loss of control. For example, in 2018, about 12% of fatal accidents were caused by speeding, whereas only 5% of total accidents had this cause. Loss of control contributed to 25% of fatal accidents but only 11% of total accidents (UK Department for Transport, 2019). It is important to consider the impact of stock market movements across the full spectrum of accidents to measure total effects. Therefore, we will look at total accidents. However, in the interest of comparison to the existing literature will also report findings for the subsample of fatal accidents for our initial model specification.

Our study is also unique in that it is the first attempt to consider the relationship between movements in a stock market index and health stemming from road accidents in Britain. Britain provides a salient context for this study as it is one of the largest financial hubs globally and was hit particularly hard by the GFC (Jofre-Bonet et al., 2018).

3. Data and summary statistics

The data on accidents come from the British Stats19 administrative data on total injury accidents reported to police in Britain (UK Department for Transport, 2016, Police reported personal-injury road accident data (Stats19), available at: www.data.gov.uk/dataset/road-accidents-safety-data). Stats19 includes information about every accident involving personal injury and at least one vehicle that has been reported to police within 30 days of occurrence. The data are collected at the scene of the accident or reported later by those involved in the accident or other members of the public. Fatal accidents involve death of at least one person within 30 days of the accident. Nearly all accidents involving fatalities are reported to police. However, there may be some underreporting of very slight accidents (UK Department for Transport, 2016).

As the Stats19 data are, in effect, a census, we can construct a daily series that extends back from 2019 to 1996, providing data on accidents over 8,766 days.⁴ Daily accidents are further split by British Government Office Regions (GORs).⁵ GORs are used as a large proportion of journeys are likely to cross boundaries for smaller areas, potentially confounding regional effects. We use the pre GFC data to test for the impact of the GFC on the relationship between the stock market and accidents. But for the main analysis in the paper, the data begin in 2008 at the start of the GFC in line with the literature on the health impacts of the GFC and run to 2019 a period significantly after the crisis but before the effects of the global COVID-19 pandemic. The GFC is known to have increased the awareness in the population of the importance of financial markets on wealth and economic stability. It is therefore important to include this period in our data and also to look at a period post GFC to ensure that our findings are not specific to the GFC period. Including the GFC also allows for more variation in returns than is found in usual trading times. This additional variation is important to the identification of the effects.⁶

In assessing the impact of stock market movements on accidents, we refer to the domestic stock market as it is likely to be uppermost in the minds of individuals. Stock market data comprise the UK Financial Times Stock Exchange 100 Index (FTSE100). The FTSE100 is a share index of the top 100 companies listed on the London Stock Exchange. It represents some 80% of market capitalization.

While accidents happen on all days of the week, stock market movements only occur on trading days, that is, weekdays. The precision of our estimates relies on the careful matching of the accident data with the stock market returns data. Given we know the time when each accident occurred, we can match each accident to the most recent known returns at the time the accident took place. In doing so, we assume that drivers respond to the latest information available to them.

As the stock market closes at 5.00 pm, accidents are aggregated into 7 time periods across the week and each period begins at 5 pm after the close of market and finishes at 4.59 pm the next day (see Fig. 1). For each period, accidents are related to the most recent known returns (R). So, for example, accidents occurring from 5 pm Tuesday to 4:59 pm Wednesday (period 2) are related to Tuesday’s return (which is the log difference between Monday and Tuesday’s closing prices). ‘Weekend’ returns carry over from the close of the market on Fridays, so all accidents from 5 pm Friday until 4:59 pm Monday (periods 5–7) are related to Friday’s return. It is important to note that we cannot use same period returns as the return for any period is not known until the market closes.

Table 2 shows the distribution of accidents for each period. We assume that people respond to the most recent information and, in the absence of any new information, retain the last piece of information that they had. This is consistent with the knowledge that humans have memory and so will retain the last piece of information received and update only when new information is available. Ideas of the brain as an information processing system have been influential in the psychology of the brain (see Baddeley, 2007). This approach is important as accidents occur 24 h a day 7 days a week. Previous studies often limit their analysis to weekdays when the stock market is open but human responses to information do not stop when the market closes. For these reasons, it is important to model accidents across all 7 periods. However, as a robustness check, we will estimate the effects separately for ‘weekdays’ (periods 1–5) and ‘weekends’ (periods 6 and 7).

3.1 Descriptive analysis

Our descriptive analysis is conducted for 2008–2019 as this is the key timeframe for most of our results, although differences do occur due to model specifications and robustness tests. We will begin by looking at the accidents data (our dependent variable) before considering the stock market data (our key explanatory variable). Finally, we consider the correlations between accidents and the stock market.

Descriptive statistics for total accidents and for fatal accidents are given in Table 1. Overall, there are about 36 accidents per region per period, although there is substantial variation with a standard deviation of 19. Fatal accidents show a similar pattern to total accidents although from a much lower base. The distribution of total and fatal accidents is shown in Fig. 2 and indicates the positive skewness in the data.

There is a pattern in total and fatal accidents, with peaks in total accidents in period 4 and in fatal accidents in periods 5 and 6 (Table 2).

Given that the literature (noted above) suggests accidents with certain characteristics might be more impacted by the stock market, for example, accidents involving alcohol, Table 3 presents descriptive statistics for accident subsamples defined by accident and driver characteristics. This allows us to consider the potential mechanisms driving any causal effects. However, our data are limited. Firstly, we consider accidents that are recorded as involving alcohol and drugs. Secondly, we consider gender as males and females may have different financial investments that mean they respond differently to stock market shocks. Males and females may also have differing emotional responses (Croson and Gneezy, 2009; Eriksson and Simpson, 2010). Finally, we consider driver age as wealth is known to be a function of age (Pissarides, 1980). Our data suggest that alcohol is a factor in more accidents than drugs. Most accidents involved male drivers. Having a driver aged 25–49 years is most common.

Looking at the stock market data (the key explanatory variable), we see the distribution of returns is symmetric about zero with a mean of 0.008, standard deviation of 1.16, and skewness of −0.13. The distribution has ‘fat tails’, with kurtosis of 11.61. There are 3,131 daily observations on returns from Monday to Friday over the sample and the range is from −9.27% to 9.38%. The distribution is shown in Fig. 3 and a summary is shown in Table 4.

While Fig. 3 shows us the stock market movements in terms of the direction of the movements we are also interested in the relative sizes of negative and positive returns. Table 4 shows the returns data categorized by quintiles. We use these quintiles to distinguish between large and small positive and negative returns.

Bivariate correlations can give some indication of what the relationship between accidents and stock market returns might look like. Looking at continuous returns we see the expected negative correlation given the existing literature. In terms of the direction of the stock market movement, we find that numbers of total accidents are weakly positively correlated with both positive and negative stock market returns, implying a V-shaped response (see Table 5 and Fig. 4). Although these correlations look small, given the sample size they are statistically significant at the 1% level.

4. Empirical specification

For all three specifications, $lnArt$ is the natural logarithm of the number of accidents in region r in period t. $Rt-1$ (Equation 2) is the lagged stock market return (from the end of the previous period) with $Rt-1+$ a continuous measure of positive returns and $Rt-1-$ a continuous measure of negative returns (in absolute values) (Equation 3).⁸$Rk,t-1$ (Equation 4) represents the categorical returns in terms of their size quintile (with Q3 as the reference category). Finally, $periodit$ is a dummy for each period and captures period effects, $GORmonthjt$ is a dummy for each region and month in the observation window and acts like a region-time-fixed effect, and $εrt$ is the normal error term for each region. We use logarithms for the dependent variable as we believe changes in returns have a proportional effect on accidents. As noted, returns are based on closing prices and are therefore lagged to ensure they occur before the accidents in each period. This ensures our models do not contain any simultaneity issues. The first model (Equation 2) is estimated for total accidents and then for fatal accidents. The remaining two models (Equations 3 and 4) are only estimated for total accidents. We cluster standard errors by region-month to remove any bias introduced due to the stock market covariate being more highly aggregated (at t) than the dependent variable (rt).

Figure 4 illustrates our modelling approach and the a priori expected findings. The negatively sloped line shows the stylized regression line that would be expected if we estimate using continuous returns. The slope of the continuous returns regression line will be function of the relative slopes of the positive and negative returns regression line. If the line is steeper for negative than positive returns, the line for continuous returns will slope downwards (assuming the range for negative and positive returns is approximately the same). For a variable that exhibits both negative and positive movements, important information may be lost if the variable is modelled continuously and predictions from such models may be misleading. The directional returns V-shape captures this information. Categorical returns essentially discretize directional returns and the size of the steps may be different for negative and positive changes and also may be larger for bigger movements relative to smaller movements. The categorical returns model therefore captures the possible nuances of the complex relationship between accidents and the stock market more completely than the other two models.

The coefficient interpretation is the same as for a logarithmic model.

We adopt a fixed effects panel regression technique which means the effects of the stock market are identified by variation within regions within months in the number of accidents, according to the GOR-month fixed effects. The GOR-month fixed effects capture aspects such as emergency service expenditure, which might impact on injury severity for example. Period fixed effects are likely to capture traffic volumes and congestion which obviously impact on the number of accidents.

In understanding the relationship between stock market returns and accidents, there are many confounding factors that are hard to measure or obtain data on. In our analysis, we difference these factors away using a fixed effects model with tight identification: one month within region variation. Our approach means we might be losing some valid variation, but the tighter identification provides confidence that we are capturing the effects of changes in returns on numbers of accidents as precisely as possible.

Given that the dependant variable is a count of the number of accidents, a Poisson or Negative Binomial model might be a more appropriate estimation strategy. The highly restrictive requirements of the Poisson model mean that overdispersion is typically a problem and so we present the ordinary least squares fixed effects (OLS FE) linear models and Negative Binomial models. However, there is a literature which suggests the fixed effects in Negative Binomial models are incorrectly specified—see Allison and Waterman (2002) and Guimarães (2008) among others—and so in line with the existing literature on road traffic accidents and stock markets (Cotti et al., 2015; Giulietti et al., 2020), we choose to present count data results as robustness checks and focus our discussion on the linear results.

Notwithstanding seasonal effects, under the efficient markets hypothesis, stock prices incorporate all currently available information about firms and their performance, and hence are highly variable and largely unpredictable (Ratcliffe and Taylor, 2015). Importantly, this implies that our models should be estimating causal effects as large changes in the stock market represent an exogenous shock to individuals which may causally affect their driving.

5. Results

We are interested in the relationship between stock market movements and accidents, but the GFC was hugely influential in terms of movements in returns and in terms of raising awareness of the importance of the stock market as an economic indicator (Deaton, 2012; Jareño and Negrut, 2015). We therefore begin our analysis by testing the conjecture that the relationship between accidents and the stock market is different in the pre-GFC sample.⁹

Table 6 shows that the pre-GFC sample (1996–2007) is very different to the sample once the GFC began and afterwards (2008–2019). This suggests that estimating effects over the entire timeframe is not sensible. This modelling supports the notion that the GFC increased awareness of the stock market as a real time economic indicator and so lifted its relevance to citizens and its reporting across various forms of media. Given the evidence above we will conduct the rest of our analysis on data from 2008 to 2019.

5.1 Continuous returns

The continuous returns results (Equation 2) for total accidents and for fatal accidents are presented in Table 7. For the total accidents sample, we see a negative, but quantitatively small, statistically significant coefficient. This tells us that as returns decrease accidents increase and is consistent with the existing literature that suggests stock market falls lead to more accidents. The relatively small effect is consistent with low involvement in the stock market of the British population and might also result from most changes in stock market returns being small enough that they go unnoticed (Frijters et al., 2015). In contrast to the existing literature though, we find this negative relationship only for the total accidents sample and find no significant effect on fatal accidents. The existing literature for the US finds effects for fatal accidents but there is no existing analysis for total accidents. However, it is worth noting that the number of road fatalities in Britain is much smaller than in the US. Per 100,000 people, the road traffic fatality rate in the US was 12.4 deaths but only 2.9 deaths for the UK in 2016 (World Health Organization, 2018). The very low fraction of road traffic deaths in the UK means that it is more likely that we might see the impact of stock market movements in our total accidents sample rather than the fatalities sample and this is precisely what we find. There are also differences in road infrastructure (and speed and congestion) that may be influencing this cross-country comparison. These results support the hypothesis (H1) that there is a causal relationship between stock market movements and road traffic accidents. As we see no effect on fatalities for Britain, we continue our empirical investigation looking at the total accidents sample.

5.2 Directional returns

We now look at the findings when returns are continuous but divided into two variables according to whether they represent falls in the market (negative returns) or gains in the market (positive returns) (Equation 3). Focusing on the OLS FE results in Table 8, we see a 1% increase in positive returns leads to a 0.7% increase in accidents and a 1% decrease in negative returns leads to a 1.2% increase in accidents. That is, accidents respond to the absolute value of the percentage change in returns and the function follows an asymmetric V-shape. These results are consistent with the hypothesis that any change in returns leads to cognitive distraction and distraction is associated with accidents (Young and Regan, 2007; Galéra et al., 2012). These results show that positive psychological effects are important, as well as negative psychological effects. This is consistent with our hypothesis (H2) that drivers have emotional responses to both good and bad news and that both types of news can lead to accidents. It is important to note that the effects are quantitatively larger for negative returns than for positive returns and this is consistent with the notion of negative bias, negative potency, and loss aversion.

5.3 Categorical returns

To investigate how the size of the stock market movement impacts on accidents we create a categorical variable, which models returns using dummy variable indicators for quintiles of returns with the middle quintile as the reference category (shown in Equation 4).

Again, we see in Table 9 an asymmetric V-shape in returns consistent with the directional returns model. Furthermore, relative to small returns, large negative returns (Q1) are associated with a 2.3% increase in accidents and large positive returns (Q5) are associated with a 2.1% increase in accidents. The pattern of coefficients shows that larger movements in returns have larger impacts on accidents, for both negative and positive returns. This is consistent with preferential attention being given to large changes relative to smaller changes. These results support hypothesis (H3) that the impact of the stock market on accidents is a function of the relative size of the stock market movements.

As a final point, we note that all the results are robust to the choice of the estimation strategy using linear or count data models. The coefficients only vary slightly so there is nothing to suggest that our focus on the linear models is misleading or biased in any way.

5.4 Explorative pathways analysis

To understand the pathways through which emotional responses result in greater accidents, we partition the total accidents sample by characteristics of accidents and the drivers involved (Table 10). These characteristics of the accidents/drivers define a set of subsamples on which we run the directional returns specification (Equation 3). This allows us to observe the differential effects of the stock market in relation to driver behaviours and driver characteristics. For example, male driver involvement in Table 10 indicates that the stock market effects have been estimated for the sample partitioned according to there being at least one male driver involved in the accident. For driver characteristics, it is not known from our data if the specified driver is ‘at fault’. Each ‘line’ in Table 10 is a separate regression estimate based on the subset of accidents defined by the chosen characteristic of the accident or driver. In each of these regressions, the dependent variable uses the inverse hyperbolic sine transformation to include periods/regions where there are zero accidents with that characteristic.¹⁰ Given that these regressions partition the sample by accident characteristics, we see more occasions where no accidents with this characteristic occur. This analysis is exploratory as our data are at the accident level and the characteristics of the driver are those that the police record so may be subjective and/or incomplete.

There is no significant response of total accidents involving alcohol or drugs to stock market returns. This suggests that the pathway from stock market shocks to accidents is not through behavioural factors relating to the consumption of alcohol or drugs. Our proxies for psychological factors are gender and age. Both male and female drivers’ accident involvement is increased by positive stock market returns, with larger effects for negative returns. Females are known to be more emotionally responsive than men. Consistent with this, we find the effects of positive stock market movements increase accidents more when the drivers are female than male (p = 0.08). We see the effects according to the age of the driver are positive for negative and positive returns. However, for young drivers, the effect is only significant for negative returns. There are significant differences between driver age groups from 25 years and up. For example, effects of positive and negative returns are higher for 50–64 year olds than 25–49 year olds (p = 0.09, 0.06, respectively) and lower for drivers aged 65 or over compared with drivers aged 50–64 years (p = 0.05, 0.02 respectively), and for those over the age of 65 years there is no effect for either positive or negative returns. Again, quantitatively larger effects are seen for negative returns. It is important to note that data limitations prevent a detailed investigation of the mechanisms that drive the causal results established in this paper and this is an avenue for future research.

In general, our results show that both positive and negative returns increase accidents. Moreover, this result is robust to variations in the sample definition (by characteristics of the accident and drivers); therefore, it is not the result of an aggregation bias. This finding is consistent with the hypothesis that stock market news—be it good or bad—induces emotional responses which ‘load up’ an individual’s cognitive resources and decrease capacity to attend to tasks (Mitchell and Phillips, 2007). This may lead to driver distraction and such driver distraction is a known contributor to accidents (Young and Regan, 2007).

5.5 Robustness check: exclusion of the GFC period

During the GFC, people became more aware of the stock market (Deaton, 2012). Stock market volatility also increased during the GFC and hence there might have been a change in the relationship with road accidents after the GFC. It is important to ensure that while the increased stock market volatility during the GFC is useful from an identification perspective the findings are not driven by this unusual period. Hence, we re-ran the directional returns model on a subset of observations comparing 2008–2019 with 2010–2019 (post-GFC). The results show that positive and negative returns increase road accidents in both samples (Table 11). Therefore, we can conclude that the results are not being driven by the GFC.

5.6 Robustness check: Modelling periods 1–5 and periods 6 and 7 effects separately

One of the important aspects of our study is the careful matching of each accident to the most recent returns information available at the time of the accident. However, given that markets are closed on weekends and we use lagged returns, we split our sample according to accidents that occur in periods 1–5 versus periods 6 and 7. In Table 12, we see that the statistical significance is being driven by periods 1–5. This analysis suggests that drivers respond to recent news, and as news becomes old (such as Friday’s returns for drivers on Sundays), it ceases to cause accidents.¹¹

5.7 Robustness check: Modelling standard deviations

As a final robustness check, we model the standard deviation of returns as this captures stock market volatility and is a measure of risk for investors. Table 13 shows consistent results. We see positive standard deviations increase accidents and negative standard deviations also increase accidents. This suggests the causal effect of returns and the volatility of returns work in the same way to cause accidents. Our premise is that returns are widely quoted and easily understood by observers and so we prefer to model using returns. However, these results suggest an interesting avenue for further research. For example, the information in returns might impact the population but the information regarding risk might only impact investors. Unfortunately, we do not have this information in these data.

6. Conclusions

This paper contributes to our knowledge of the wider impacts of stock market crashes on health. We examine the impact of stock market movements on road traffic accidents. The existent literature has only examined the impact on fatal accidents. There is no reason to assume that only fatal accidents will be impacted (as fatal accidents are a small proportion of total accidents).

We apply a behavioural and economic psychology lens to motivate how stock market movements might elicit behaviours and emotional responses that can lead to accidents. Importantly, we use daily data on accidents and FTSE100 returns which is a significant improvement on the existing literature which typically looks at monthly data and might contain aggregation bias. Given that we are interested in responses that may be short term, it is important that we have precisely timed data that match the time of the accident to the most recent known stock market returns. Our data are timed to accurately capture the last stock market information in the period before the accident occurred. This also allows us to model accidents across the full week at times when the stock market is open and when it is closed. This is important as accidents still occur when the stock market is closed.

Employing a tightly specified model, we examine stock market returns in several ways to estimate this causal relationship. Firstly, we consider continuous returns and effects consistent with the existing literature that stock market falls causally increase accidents. Our results show this finding for total accidents and no effects for fatal accidents. However, the existing literature is for the US while our study is for Britain and the level of fatal accidents in Britain is much smaller.

Secondly, we consider the direction of the stock market movements in terms of continuous gains and losses. Our results for total accidents show that both positive and negative continuous returns can lead to accidents. We find an asymmetric V-shaped response which reflects negative bias, negative potency, and loss aversion. Importantly, our findings are consistent with the positive psychology paradigm that shows emotional responses to positive outcomes are not necessarily equivalent (in size) to the emotional responses of equivalent negative outcomes, leading to asymmetric responses. The impact of positive returns on accidents is not visible in the continuous returns specification.

Next, we consider if the impacts differ by the relative size of the stock market movement. We discover that large movements have bigger impacts on accidents than small movements (for both loses and gains). Again, this is consistent with a psychological pathway. Large stock market movements will have bigger emotional responses and so are more likely to cause accidents to occur through driver distraction. Consistent with the ideas of preferential attention to negative events, we see that the responses to stock market falls are larger than the responses to market gains. Again, these results are not visible in the continuous returns specification. Our results on directional and categorical returns are important as important information may be lost unless we allow for directional/categorical returns in our modelling. These key findings are confirmed by multiple robustness tests and estimation methodologies presented in the paper.

This paper suggests drivers are responding to stock market conditions, leading to accidents. Exploratory mechanisms analysis ruled out behavioural responses through alcohol and drug use and lean towards psychological responses as the most likely pathway for the causal findings reported in this paper. The theoretical mechanisms that can explain this finding are through the psychological impact on emotions, leading to driver distraction resulting in accidents. Negative returns lead to worry and anxiety, whereas positive returns lead to an abundance of happiness/euphoria about the individual’s own financial position. Emotional responses can lead to driver distraction which in turn is a known cause of accidents. While more research is needed to empirically prove this pathway, the results suggest that awareness campaigns focusing on the impacts of psychological factors and their associated emotional responses on driving abilities, particularly driver distraction and inattention, could mitigate the effects of more extreme positive and negative stock market movements. To date, governments have heavily focused on the impact of behavioural responses such as alcohol and drugs on driving outcomes. More recently some focus has been on physical distractors such as the use of mobile phones while driving. To date, we know of no campaigns that have focused on psychological distractors that can cause accidents. This suggests that such campaigns could protect against road accidents and be influential in protecting population health outcomes.

Supplementary data

Supplementary material is available on the OUP website. These are the data and replication files. Data on involvement of alcohol and drugs are available from the UK Department for Transport on application (e-mail: [email protected]).

Funding

This work was supported by an Australian Government Research Training Program Scholarship.

Footnotes

Acknowledgements

The authors would like to thank Michael Shields, David Johnston, and seminar participants at the School of Economics, Finance and Marketing seminar, RMIT University, and at the Department of Management, Marketing and Entrepreneurship and Department of Economics and Finance seminar, University of Canterbury for helpful comments. This work has also benefitted from comments provided by two anonymous referees and from feedback received at the Australian Health Economics Society Conference and the Monash Business School Doctoral Colloquium. They would also like to acknowledge Angel Zhong for assistance collecting the stock market data. All remaining errors and omissions are solely the responsibility of the authors.

References