National differences in gambling-driven stock trading behavior: evidence from a simulated trading game

The accumulation of wealth is a crucial objective for private households, as it enables them to make profitable investments, such as purchasing a car, pursuing educational opportunities, or smoothing their consumption over economic downturns and retirement. In many economies where house prices are soaring, the equity market remains the only accessible avenue for long-term investment, particularly for those with limited savings. However, numerous studies show that private households often make mistakes when trading on equity markets, resulting in suboptimal or even negative returns (see Barber et al. 2013, forareview).

In this study, we focus on one trading anomaly commonly known as ’excessive trading,’ as identified by Barber and Odean (2000). Private households have a tendency to excessively trade by frequently selling shares and buying into new ones. Since they often lack the ability to distinguish winning from losing stocks, the trading activity does not lead to changes in the gross returns of their portfolio. However, fees associated with these trades (e.g., commission fees or bid-ask spreads) can significantly reduce their net returns, leading to suboptimal performance.¹ A number of studies (Dorn and Sengmueller 2009; Grinblatt and Keloharju 2009) identify an individual’s tendency to gamble as an important factor contributing to excessive trading. Furthermore, even when controlling for other influencing factors, Liu et al. (2022) find that this relationship remains relevant

Drawing on previous research, we aim to conduct a national comparison to investigate the relationship between gambling behavior and the tendency to trade excessively in different European countries. There is substantial variation in gambling activity across European countries, and Italy has a much more pronounced gambling sector than Germany, as evidenced by average citizen spending of over 2000€ per year in Italy, compared to just over 600€ in Germany (Agenzia delle Dogane 2019, 2020; Statista 2021). We aim to examine whether these national differences in gambling behavior translate into differences in the tendency to engage in excessive trading across these two countries.

We conducted a survey to quantitatively answer our research question. Participants play a non-incentivized trading game following the method of Mosenhauer (2021), which features a stylized and simplified hold-or-trade scenario. For this task, participants are instructed to aim for the highest possible returns. Based on real-world historical stock prices, participants are randomly assigned a stock and can choose to keep it to earn its returns or exchange it for another one at a known commission fee. The participants’ behavior yielded metrics of their trading activity across trading rounds and profitability (see Sect. 3 for a detailed description of the design). We correlate this behavior with personal characteristics obtained from a follow-up questionnaire, which includes two established constructs of gambling activity: the 'Consumption Screen for Problem Gambling' (Rockloff 2012) and the 'Problem Gambling Severity Index' (Miller et al. 2013; Loo et al. 2011). We supplement these data with demographics, most notably the participants’ nationality grouped into German-speaking (Germany, Austria and German-speaking South Tyrol) and Italian-speaking (Italy except South Tyrol) participants.

We obtain 55 responses from participants belonging to the German language group (mainly Austria and Germany) and 36 from Italy. Regarding both our survey metrics on gambling activity, we confirm large differences in gambling behavior between the national groups, with Italians reporting a much higher affinity to gambling. As predicted by our theory, we also find that Italians show a significantly higher turnover in the trading game. Interestingly, while both the 'Consumption Screen for Problem Gambling' and the 'Problem Gambling Severity Index' predict trading activity across national groups, we find that the 'Problem Gambling Severity Index' dominates the 'Consumption Screen for Problem Gambling' when considered as determinants simultaneously. This suggests that gambling on stock markets should rather be seen as a harmful addiction than a utility-generating entertainment. Lastly, we confirm that the tendency to gamble explains most of the observed national differences in trading activity to the point where the participant’s nation loses significance as an explanatory variable at the 5% level when controlling for gambling preferences. Our results thus suggest that cultural differences between Italians and German-speaking individuals, apart from gambling habits, can be disregarded when predicting the discussed trading behavior.

We believe that our results have important implications, particularly for the policymaking of the Italian government. Grinblatt and Keloharju (2009) suggest that both traditional forms of gambling and highly active stock trading are motivated by a drive for 'sensation seeking.' Gao and Lin (2015) further suggest that individuals with such a drive will attempt to satisfy their 'gambling fix' but are willing to substitute among different forms of gambling, i.e., they may engage in less day trading when lottery jackpots are high. Similarly, Kumar et al. (2021) found that gambling in stock markets increased strongly when casinos were forced to close in response to the contact restrictions brought on by the Covid-19 epidemic. All this implies that if the Italian government aims to curtail the harmful consequences of gambling activity, policies should target the root of the problem, namely an individual’s drive to gamble, rather than its symptoms. Moreover, the results suggest that an important avenue of wealth creation for Italians is severely undermined as long as national preferences toward gambling remain unchanged.

The central hypothesis of this paper is that there are differences in trading behavior between German and Italian language groups, which are related to their respective gambling behavior. To outline possible correlations, we distinguish two strands of literature. In past studies, evidence for positive correlations (Dorn and Sengmueller 2009; Grinblatt and Keloharju 2009; Mosenhauer et al. 2021) between gambling and trading can be found, describing how a person’s preferences for sensation-seeking affect both their gambling and trading behavior. A second strand documents negative correlations (Dorn et al. 2015; Gao and Lin 2015), describing the substitutability of gambling and some forms of stock trading. After a brief description of the state of gambling in the discussed nations, we discuss the related literature.

This study uses a survey to investigate how national differences in gambling behavior drive differences in trading behavior. It employs a convenience sample of self-selected participants. Using a link, anyone could participate in a survey created in German, Italian, and English, from which potential participants could choose as per their preference. The link was shared via email among students of the authors’ institution and via social media. We expect that some of the participants invited others to participate, too.

Following Mosenhauer (2021), the survey experiment simulates trading decisions based on past returns in an abstracted hold-or-trade scenario. In ten different rounds of trading, participants are assigned random stocks from the 'Standard & Poor’s 500' index in a randomly determined month within the time period from January 1960 to July 2017. The only information available to the participants was the stock’s past performance in the last 10 months. In each round, the participant has to make a binary decision to either hold the allotted stock or trade it. If the participant decides to hold the assigned stock, they would simply earn the real-world returns of that stock during the allotted month. On the other hand, if the participant decides to trade their stock, they are assigned another random stock in a random month out of the basket. Additionally, a commission fee of 5% (percentage points) is deducted from the yielded returns of the new stock. So, in case of trading stocks, the participant receives a random return and an additional − 5% fee. The returns of all trading rounds are accumulated and displayed at the end of the experiment, and participants are instructed to try to maximize this number, although no monetary or other external incentives were provided. In any case, each round starts with a new stock allocation, and trading rounds were independent in the sense that a participant’s decisions in a given round could never influence another round.

To measure gambling behavior in the sample, two established benchmarks, the 'Consumption Screen for Problem Gambling' (Rockloff 2012) and 'Problem Gambling Severity Index' (Miller et al. 2013; Loo et al. 2011), are included in the questionnaire. The Consumption Screen for Problem Gambling (CSPG) is based on the conceptual analogue 'Alcohol Use Disorders Identification Test—Consumption' (AUDIT-C) (Daniel R et al. 1998), which measures gambling consumption rather than the negative consequences caused by gambling. Participants answer three brief multiple-choice questions regarding their gambling behavior, and according to their answers, they are assigned a score ranging from 0 (lowest) to 4 or 5 (highest). The Problem Gambling Severity Index (PGSI) is another benchmark used in this study to measure gambling consumption and study problem gambling further (Miller et al. 2013; Loo et al. 2011). Similar to the CSPG, this benchmark also uses 9 multiple-choice questions with items on an ordinal scale. The items are associated with scores from 0 to 3, which are accumulated at the end.

The trading experiment takes place after the personal questions, where the decision to trade (1) or hold (0) the stock is coded as the dichotomous variable Trading. To measure gambling behavior, the two benchmarks mentioned above are used with CSPG scores of 4+ considered indicative of problematic gambling. The PGSI, on the other hand, categorizes the population into four different groups: non-problem gambler (score 0), low-risk gambler (score 1–2), moderate-risk gambler (score 3–7), and problem gambler (score above 8). While the categories themselves are irrelevant for further analysis, as the score is used in its numeric value, they are useful for quickly evaluating an individual’s gambling behavior. Both scores are divided by the maximum possible score to normalize them, resulting in the final variables named CSPG and PGSI.

The sample consists of a total of 106 participants. The sample consists of 55 male and 37 female participants; additionally 14 participants did not declare their gender. Diverse gender has not been recorded. Of the 106 participants, 55 belonged to the German language group, 36 to the Italian language group, and 1 to another language group. 14 persons did not provide any information on their language affiliation or dropped out of the questionnaire before this information was collected. Table 1 shows some key descriptive statistics of the sample.

To test for differences between Austrians or Germans, on the one hand, and Italians, on the other hand, in their gambling and trading behavior, we use two-sample t-tests. We test for differences in the PGSI variable (for gambling behavior) and TRADE variable (for trading behavior) with respect to groups formed according to the NATION variable. As shown in Table 2, both tests reveal significant national differences at \(\alpha <0.01\).

Comparing the arithmetic means of the PGSI variable between German-speaking and Italian-speaking participants, we find that the Italian average (PGSI = 0.0824) is almost 8 times higher than the average of German-speaking participants (PGSI = 0.0105). The TRADE variable takes on an average value of 29.31% for German-speaking participants and an average value of 41.25% for Italian-speaking participants. This difference is highly significant with a p-value of 0.002, falling within the 99% range. The results of these two t-tests are shown graphically in Fig. 1. We thus confirm national differences within the sample in both gambling and trading behavior.

The differences in trading behavior can be further analyzed with a regression analysis. Table 3 presents the coefficient estimates from various logit regression models. Different columns correspond to different model specifications. In each case, trade is the dependent variable, while we interchange different combinations of nation, PGSI, and CSPG as independent variables. Our unit of observation is a participant’s trading decision in a given round of the trading game. Each participant provided ten such decisions.

The first three columns of Table 3 present the respective univariate regressions of TRADE on NATION, PGSI, and CSPG. The first column displays a significant coefficient of 0.1193 (\(\alpha <0.01\)), indicating an 11.93% increase in trading activity among Italian-speaking participants compared to their German counterparts. This result is consistent with the previously conducted t-test. The influence of problem gambling (PGSI) is highly significant with a p-value of 0.004%. Multiplying its regression coefficient with the respective standard deviation reveals that 'problem gamblers' trade on average 7.54% more than those who do not gamble. The influence of gambling activity in general (CSPG) is also significant with a p-value of 1.6%. Analogous to the previous calculation, the standardized effect of CSPG gambling consumption on trading activity is 6.02%.

The fourth column in Table 3 presents the results of a logit regression analysis that examines the influences of the two variables on trading behavior. Interestingly, it shows that the variable measuring general gambling activity, CSPG, no longer has a significant effect on trading behavior when controlling for PGSI (p-value = 0.8267). Moreover, the coefficient is now negative, indicating that individuals with high levels of general gambling activity tend to trade less. This result is not consistent with the univariate regression analysis conducted between CSPG and TRADE. On the other hand, the variable PGSI increases its influence on trading behavior to a coefficient of 1.0647 and remains significant. This suggests that the effect of general gambling activity is 'absorbed' by pathological gambling and that it is not gambling in general, but rather pathological gambling, that has a significant influence on trading behavior. This finding highlights the importance of considering different levels of gambling behavior when examining its relationship with trading behavior.

Lastly, we test for interdependencies between the participants’ nationality, gambling behavior, and trading behavior. Since PSGI dominates CSPG in its effect on trading activity, we will not consider CSPG further as a measure of gambling activity. Column five of Table 3 shows the effect of nation on trade while controlling for gambling activity. Strikingly, we find that the influence of the nation variable on trading behavior falls to 7.37% from 11.93% when considered in isolation. The influence of gambling also diminishes in magnitude.

We take this as evidence that the measured effects overlap, indicating that some of the observed effect of national differences on trading behavior can be attributed to differences in gambling behavior among participants. Crucially, PGSI appears to have a more dominant effect than nation, as the effect of nation does not reach significance at the 0.05% level when controlling for gambling preferences while problem gambling remains significant at the 0.01% level. This provides further evidence that the effect on trading observed for language affiliation is primarily due to the variable PGSI or problem gambling and that the main factor influencing excessive trading is not language affiliation but rather the participants’ involvement in the gambling market.

It should be noted that our categorization of nationalities according to participants’ mother tongues may not perfectly capture their cultural affiliation. Some participants may have grown up speaking one language but due to migration lived large shares of their lives in countries speaking another language, so that cultural backgrounds are mixed. We partly attempt to account for such dynamics by recoding our nation variable according to the language each participant chose for answering the survey (e.g., German, Italian, or English). No participants with German as their mother tongue answered the survey in Italian, while two participants with Italian as their mother tongue answered the survey in German. 8 participants answered in English. We generate a new nation variable by dropping all English-answered entries and coding all answers provided in Italian with 1 and 0 for the answers in German. Using this new nation variable, we rerun our main regression exercise shown in Table 3 and find that our findings qualitatively remain unchanged except that the effect of a participant’s language affiliation on trading activity, while sharply decreasing in magnitude, remains significant at the 5% level when controlling for problem gambling.

The study aims to investigate the national differences between German- and Italian-speaking countries in Europe regarding their trading activity on a stylized stock market, with a particular focus on whether the gambling preferences of individuals could explain these observed differences. The research confirms substantial and significant differences in both gambling and trading behavior across the language groups. Moreover, gambling-motivated trading should be seen as an addiction rather than entertainment. The findings suggest that the observed national differences are mostly due to differences in gambling preferences rather than other cultural attributes.

We believe that these findings have significant implications for households for two reasons. Firstly, they highlight that pathological gambling not only has a direct negative impact on household finances, but it also leads to excessive trading behavior, which further depletes the household budget (Barber and Odean 2000). Secondly, the research underscores the importance of educating individuals not only about financial literacy but also about the consequences of gambling, especially pathological gambling. The regression analyses presented in Table 2 suggest that gambling behaviors transfer to trading behavior, emphasizing the need to address the issue of pathological gambling to prevent financial losses on the stock market.

For future studies, it would be very interesting to investigate what happens when comparing a group of problem gamblers with a group of non-gamblers without making distinctions based on language. The effects shown could be even stronger and the differences even more pronounced, which would give this study further scientific relevance. Additionally, it would be useful to determine whether pathological gambling genuinely causes excessive behavior or whether excessive behavior conversely increases the tendency toward pathological gambling. This study only showed a positive correlation. Studies with larger, multinational, and multicultural samples could be conducted to serve this purpose.