Viral decisions: unmasking the impact of COVID-19 info and behavioral quirks on investment choices

The COVID-19 pandemic precipitated sharp declines and heightened volatility in global equity markets, including Pakistan, where indices fell markedly following the first confirmed cases. Against this backdrop, the study examines how behavioral biases—investor sentiment, overconfidence, over/under-reaction, and herding—affect investment decisions during the pandemic. It further evaluates how COVID-19-related information (e.g., infection rates, lockdowns, vaccines, stimulus) moderates these relationships. The research also explores the role of sociodemographic variables (age, occupation, gender, education, investor type, objectives, reasons for investing, horizon, pre-investment considerations, and advice sources) in shaping investment behavior. The purpose is to clarify the interplay between behavioral biases and pandemic information in determining investment choices, and to inform policy and modeling via cluster analysis and an agent-based model of herding.

The review details how pandemic-driven investor sentiment, shaped by external information, influences returns and volatility, with stronger reactions to earnings and news during high-sentiment periods. Overconfidence is framed as overestimation and over-placement, leading to suboptimal risk-taking and misattribution of outcomes; COVID-19 information can distort risk perception among overconfident investors. Over/under-reaction challenges market efficiency, with pandemic news inducing exaggerated responses and non-linear market behavior across regions. Herding—mimicking others’ trades—intensifies under uncertainty; COVID-19 information overload may foster or alter herding tendencies, particularly among individual investors. The review posits COVID-19 information sharing as a moderator between behavioral biases and investment decisions and highlights the influence of demographics (age, gender, education, income, experience) on biases and trading outcomes. Hypotheses H1–H4 test direct effects of each bias on investment decisions; H5–H8 test COVID-19 information sharing as a moderator of these effects.

Design: Quantitative survey of individual investors in Pakistan during COVID-19. Convenience sampling via online Google Form; 750 invitations; 257 responses; 223 usable (valid response rate 29.73%). Target population: individual investors (Punjab); surveys distributed in major cities (Karachi, Lahore, Islamabad, Faisalabad). Bias checks: Non-response bias tested via early vs late respondents (independent samples t-tests): no significant differences (p>0.05). Common method variance assessed by Harman’s single-factor test: no single factor >50% variance, indicating CMV not a concern. Measures: Five-point Likert scales (1=strongly disagree to 5=strongly agree). Investor sentiments (5 items; adapted from Metawa et al. 2018; Baker & Wurgler 2006). Overconfidence (3 items; Dittrich et al. 2005). Over/under-reaction (4 items; De Bondt & Thaler 1985; Metawa et al. 2018). Herding behavior (3 items; adapted from Bikhchandani & Sharma 2000; Metawa et al. 2018). COVID-19 information sharing (4 self-developed items validated by expert panel; infection rates, lockdowns, vaccination developments, stimulus packages). Investment decisions (10 items; adapted from Metawa et al. 2018). Reliability and validity: Cronbach’s alpha—Investor sentiments 0.888; Overconfidence 0.827; Over/under-reaction 0.858; Herding 0.741; Investment decision 0.933; COVID-19 0.782. Convergent validity: factor loadings >0.60, CR ≥0.70, AVE ≥0.50 for all constructs. Discriminant validity supported (alphas exceed average inter-construct correlations). Analytic strategy: Descriptive stats and bivariate correlations; visualization via parallel coordinates and correlation network graphs. Hypotheses testing: Moderation analysis using Hayes PROCESS Macro (Model 1) in SPSS with bootstrapping (95% CIs). Model: ID = β0 + β1(IS) + β2(OV) + β3(OR) + β4(HB) + β5(IS×COVID) + β6(OV×COVID) + β7(OR×COVID) + β8(HB×COVID) + μ. Clustering: K-means (elbow method to select k), feature importance via Extra Trees Classifier; visualization with seaborn. Agent-based model: 223 agents trading a single stock; agent prototypes reflect empirical clusters and socio-demographic profiles (age, income). Younger/lower-income agents parameterized with higher herding propensity; simulation over 50 steps using Python (Mesa, NumPy, SciPy, pandas, matplotlib). Performance metrics include balance trajectories, strategy efficacy, risk/volatility, market alignment, and adaptability.

Sample: N=223; majority aged 20–30 (61%), male (61%), salaried (56.5%), master’s degree (67.3%), seasonal investors (63.7%). Objectives: growth and income (37.2%); purpose: wealth creation (41.3%); horizons: medium term (43.5%) and long term (28.3%); primary factor: high returns (38.6%). Advice sources: family/friends (44.8%), social media (29.6%). Reliability/validity: All constructs reliable (alphas ≥0.741) with adequate convergent and discriminant validity. Correlations: All constructs positively correlated (p<0.01). Direct effects (Table 6): Investor sentiments → Investment decision (R²=0.866, β=0.961, SE=0.083, 95% CI [0.797, 1.125], significant). Overconfidence → ID (R²=0.696, β=0.867, SE=0.118, CI [0.634, 1.099], significant). Over/under-reaction → ID (R²=0.668, β=0.884, SE=0.125, CI [0.638, 1.131], significant). Herding behavior → ID (R²=0.499, β=0.698, SE=0.171, CI [0.361, 1.036], significant). Moderation by COVID-19 information sharing (negative interactions): IS×COVID → ID (β=−0.034, SE=0.026, LLCI −0.086, ULCI −0.018, significant); OV×COVID → ID (β=−0.064, SE=0.037, LLCI −0.136, ULCI −0.009, significant); OR×COVID → ID (β=−0.083, SE=0.038, LLCI −0.159, ULCI −0.007, significant); HB×COVID → ID (β=−0.124, SE=0.051, LLCI −0.225, ULCI −0.022, significant). Interpretation: Behavioral biases positively relate to investment decisions, but COVID-19 information sharing significantly and negatively moderates these relationships, attenuating their impact. Clustering: Elbow method supported k≈3; K-means identified three clusters. Overconfidence and over/under-reaction, together with COVID-19 information sharing (CIS1–CIS4), were key features differentiating clusters; age, occupation, and investor type also influential. Agent-based model: Younger and lower-income agents exhibited stronger herding and underperformed. Example: in one run, Agent 74 (higher income, ≥40 years) reached a peak balance of 911 units (min 732, mean 799, SD 41) when stock price was ~20.03 at step 45; overall average agent balance ~84 units with worst at −670, underscoring superior performance of less-herding, higher-income, older agents. Data quality checks: Non-response bias not significant (p>0.05 across constructs); Harman’s single-factor test indicated CMV not problematic.

Findings confirm that core behavioral biases—investor sentiment, overconfidence, over/under-reaction, and herding—significantly shape investment decisions. However, pandemic-related information sharing reduces the strength of these relationships, suggesting investors adjust their behavior amid heightened uncertainty and risk salience. COVID-19 information appears to temper overconfident and overreactive tendencies, diminish mimicry, and shift decisions toward more cautious or balanced strategies. The strong associations between socio-demographics and behavior indicate that younger, less experienced, and lower-income investors are more susceptible to herding, consistent with the ABM outcomes. These insights are relevant for policymakers and advisors: transparent, accurate crisis communication and financial literacy initiatives can mitigate bias-driven inefficiencies and support more rational investment behavior during systemic shocks.

The study demonstrates that behavioral biases have strong positive links with investment decisions, while COVID-19 information sharing significantly and negatively moderates these effects. Cluster analysis and agent-based simulations reveal that younger and lower-income investors are more prone to herding and tend to underperform, whereas older and higher-income investors perform better by relying less on mimicry. Contributions include: (1) documenting the moderating role of COVID-19 information across multiple biases; (2) integrating sociodemographic profiling with clustering to inform agent design; and (3) illustrating policy-relevant dynamics via an ABM. Future research should expand bias coverage, include institutional investors, adopt longitudinal and cross-country designs, and complement quantitative analyses with qualitative case studies. Enhancing information transparency and investor education can foster resilience and rationality in crisis-period decision-making.

Key limitations include: (1) modest sample size and convenience sampling limiting generalizability; (2) cross-sectional design precluding dynamic causal inference—longitudinal studies are recommended; (3) focus on individual investors within a specific national context—results may differ across countries and regulatory environments; (4) exclusion of other important behavioral factors (e.g., loss aversion, anchoring, recency, personality traits); (5) reliance on self-reported measures subject to response biases despite CMV and non-response checks; and (6) limited external validation of the ABM beyond the study’s empirical clusters. Future work should broaden samples, include institutional investors, add qualitative methods, and test models across markets and time.