Does Managerial Overconfidence Change with Market Conditions? Risk Management for Financial Institutions

Abstract

Overconfidence (hubris or overestimation of one’s ability to perform) has been viewed in the finance literature as a character trait that is stable over time, e.g., assuming that if a manager is overconfident, he/she is overconfident all the time. In this paper, we aim to show that managerial overconfidence can be state-contingent, i.e., the level of managerial overconfidence could be influenced by an external economic shock such as the global financial crisis in 2008. A novelty of this paper is to provide evidence for and application of the concept of state-based managerial overconfidence, which is new in the finance literature. Two empirical studies were reported. In the first study (Study 1), we analyzed real market data by linear regression. We found that managerial overconfidence could vary according to changes in the state of the macroeconomy or tightening of corporate governance policies. In the second study (Study 2), we conducted a lab experiment simulating how external manipulations could alter participants’ confidence level. Both our empirical studies provide strong evidence of state-contingent overconfidence by Student’s t-test and contribute to the current finance literature, which assumes overconfidence as a personality trait. Our findings have important practical implications for the credit market. According to the state-contingent overconfidence hypothesis, creditors might reduce the loan amount or the loan duration (or other loan contract terms) too excessively by more than the efficient level during an economic downturn if the offsetting effect of state-contingent overconfidence is ignored.

1. Introduction

The finance literature often assumes managerial overconfidence (overestimating one’s own ability to perform) to be trait-based. That is, if a manager is overconfident, he or she will remain overconfident all the time (see Galariotis et al. 2022; Lartey and Danso 2022; Hirshleifer et al. 2012; Voon et al. 2022 for reviews). Unlike the traditional trait-based assumption in the finance literature, the goal of this paper is to demonstrate that managerial overconfidence can be state-contingent. That is, the level of self-overconfidence could change in accordance with external environmental shocks, such as the 2008 global financial crisis (GFC) or the implementation of the 2002 Sarbanes Oxley Act (SOX). We used both real market data and a lab experiment to provide evidence of state-based managerial confidence.

The concept of state-based overconfidence is new in finance. The finance/management literature focused hitherto on the concept of self-attribution bias, which means that individual self-confidence might be attributed, for instance, to his or her own intelligence or character traits (Galariotis et al. 2022; Lartey and Danso 2022; Voon et al. 2022), or to self-employment (Cassar 2010), or individual overconfidence could be made or nurtured (Billet and Qian 2008). There is also research pointing out how hubris could be prevented (Petit and Ballaert 2012). These lines of research do imply that overconfidence intensity may change over time. However, the above studies did not explicitly test the state-contingent managerial overconfidence hypothesis that we propose in this paper. There is a paucity of research in finance positing how one’s self-confidence level could be influenced directly by the external environment. Our current work contributes to the finance literature, as this is the first time the state-dependent managerial overconfidence hypothesis has been empirically tested.

The finance literature often compares risk behavior between overconfident and underconfident managers. For example, Lee et al. (2017) found that the founder chief executive officers (CEOs) of large companies are more overconfident than their non-founder counterparts. Since some CEOs are more overconfident than the others, the finance literature often separates the CEOs into two dichotomous groups, namely, the overconfident group and underconfident group, using option-based measures (Malmendier and Tate 2005). The purpose of such dichotomous grouping is to compare the investment or managerial behavior of overconfident and underconfident groups of managers (Ahmed and Duellman 2013; Christensen et al. 2016; Malmendier et al. 2011) and to examine how firm factors may affect the risk behavior of overconfident and underconfident managers (see Humphery-Jenner et al. 2016). The potential change in self-confidence level for each individual or within each dichotomous group, however, has been ignored. Yet, in this paper, we aim to show that the level of overconfidence in an overconfident manager and hence the extent of his/her investment or risk-taking behavior could go up or down depending on the state of the macroeconomy. We used two empirical studies to test our state-contingent overconfidence hypothesis. In the first study, we analyzed real market data and developed continuous managerial overconfidence measures that allow the overconfidence level to change within an overconfident CEO individual or group. We aimed to show that the same overconfident manager could become less overconfident during a financial crisis. In the second study, we conducted a lab experiment to show that participants’ self-confidence level can be significantly affected by external manipulation.

The research questions in the paper are two-fold. First, in traditional finance research, we find that overconfidence is often assumed to be a personality trait and this characteristic tends to be stable over time, while in psychology, the concept of state-contingent confidence has appeared for years. However, we believe that it would be meaningful for the contemporary finance literature to take state-based overconfidence into account, as the overconfidence level could be affected by external shocks such as policy changes and macroeconomic downturns. In the first hypothesis, we wish to find out whether overconfident managers encountered a significant drop in their overconfidence level when the SOX was implemented or the GFC happened. In addition to market data, in our lab experiment, we predict that participants’ confidence level based on the estimation of the over-precision would decrease when there is an external shock. Second, if the first hypothesis is confirmed that the overconfidence level of CEOs could be negatively affected by the external shocks (SOX and GFC), we hope to discover its implications for the credit market. Many firms, especially big corporations like the Fortune 500 companies in our dataset, adopt debt financing to finance the companies. One of the major ways to borrow is from bank loans. Therefore, loan terms or loan covenants are vital in affecting firms’ behavior. In the second hypothesis, we wish to confirm whether there is a statistically significant drop in the loan amount to the companies managed by overconfident CEOs. As lenders, banks may not be aware that external shocks mitigate and significantly reduce the credit risks initiated by overconfident CEOs, and they could have imposed much stricter loan terms or a great reduction in loan size. This could also carry detrimental effects to the pace of economic recovery, as it hinders the companies’ after-crisis growth.

In our Study 1, the GFC and SOX are used as examples of changes in the external environment that may alter the levels of individual overconfidence. The GFC and SOX are chosen for a couple of reasons. First, the data on the GFC and SOX are readily available. Second, the GFC and SOX provide the basis for our natural experiments given that both the GFC and SOX are exogenous shocks (see Banerjee et al. 2015; Voon et al. 2022 for review). The choice of the GFC and SOX within the natural experiment setting helps to allay potential endogeneity or reverse causality concerns that may arise. In our Study 2, we resort to using a laboratory experiment to examine if one’s confidence level can be affected by external influence.

The evidence of state-contingent managerial overconfidence has practical implications for credit market in general. Take the case of the GFC as an example. In the presence of the GFC, creditors would, for instance, reduce lending to firms due to the systemic risk from the GFC. However, a downward adjustment of a loan during the GFC could exceed the efficient level of adjustment if the state-contingent overconfidence effect is ignored by the creditor. This is because a decrease in state-based overconfidence (and hence the risk behavior of the CEO) from the GFC would partially offset the increase in systemic risk emanating from the GFC. The overall risk faced by a creditor comprises the systemic risk and the state-contingent overconfidence risk. The GFC raises the systemic risk but at the same time lowers the state-contingent overconfidence risk. Hence, a creditor may reduce its lending by more than the efficient level (or too excessively) if the opposite (positive) effect of state-contingent overconfidence is unaccounted for.

2. Study 1: Empirical Evidence for State-Contingent Managerial Overconfidence from Real Market Data

Managerial overconfidence is defined as the deliberate delay of top executives vested with option compensations in exercising their in-the-money options (Malmendier and Tate 2005). Using this definition, once a manager fails to exercise his/her option when the option price rises over 67% of the exercise price, he/she is identified as overconfident. Conversely, if a manager sold his/her option before the option price reaches the 67% threshold of the exercise price, he/she is defined as being underconfident. In the context of this paper, a rise in the level of managerial overconfidence is represented by a continuous variable or continuum that increases from the 33% threshold to the 50% threshold to the 67% threshold to the 85% threshold. A fixed CEO effect is adopted in our specification as we attempt to rule out the possibility of a change in managerial overconfidence attributed to a jump from an individual CEO who is less overconfident to another individual who is more overconfident.

In Study 1, using real market data over the period 1996–2011 as the data collection work ended in 2015, we examined whether managerial overconfidence could be state-contingent or not. We explored if a change in the state of the macroeconomy induced by a financial crisis (such as the 2008 global financial crisis or GFC) or a change in financial regulation (such as the implementation of the Sarbanes Oxley Act or SOX) would directly alter the overconfidence level of an overconfident manager. In addition to imposing the CEO fixed effect, we controlled for other possible firm factors that may affect the managerial overconfidence level. The GFC and SOX are exogenous economic shocks and therefore their unidirectional effects on the managerial overconfidence level would mitigate endogeneity and reverse causality effects that might bias our empirical findings (see Voon et al. 2022 for a recent review).

Banerjee et al. (2015) posited that the SOX, implemented in 2002, would not change the overconfidence level of a manager but merely restrain the managerial functions of an overconfident CEO as independent directors and auditors are empowered to circumvent the investment decisions made by the risk-loving overconfident CEOs. In Study 1, we tested whether the SOX would lower the overconfidence level of an individual CEO rather than simply restraining the activities of an overconfident CEO. To achieve this, we interacted SOX with the managerial overconfidence level. A significant interaction coefficient implies that the SOX impacted on the managerial overconfidence level (but not vice versa, as SOX is an exogenous shock). The use of an interaction term between the SOX and overconfidence level, therefore, constitutes an appropriate model for capturing the state-contingent evidence of managerial overconfidence. Furthermore, we predicted that managerial overconfidence level for an individual CEO could be influenced by the GFC.

Prior research has stated that overconfident managers invest aggressively and engage in riskier activities than less overconfident managers (see Voon et al. 2022 for a review). During the GFC, for instance, creditors would reduce the size and the duration of the loan in order to alleviate the systemic risk emanating from the GFC. However, managers’ overconfidence level, and hence their risk-taking tendency (see Voon et al. 2022 for review), would also be reduced during the GFC. The state-contingent overconfidence hypothesis suggests that banks or financial institutions should respond to the increase in systemic risk (generated by the GFC) by tightening the loan credit terms (such as loan size and loan duration). At the same time, however, banks or financial institutions should relax their credit terms on loans in tandem with the lowering of the risk-taking propensity of the managers (as their overconfidence levels would have been reduced accordingly during the GFC). The credit market will be distorted (i.e., loan size and/or loan duration could be over-reduced) if the state-based overconfidence phenomenon is overlooked. It is important for the credit market to be informed of the above inefficiency effects.

2.1. Baseline Model

To provide evidence of the existence of state-contingent managerial overconfidence, we collected actual market data to examine if the 2008 global financial crisis (GFC) and Sarbanes Oxley Act (SOX) implemented in 2002, both exerting exogenous and prolonged changes to the macroeconomy, would affect managerial overconfidence. The managerial overconfidence level is initially modelled as the dependent variable while the GFC and SOX are the independent variables. The empirical model is expressed as:

Overconfidence Level = f (GFC/SOX; CEO controls; firm controls; error)

(1)

CEO characteristics, such as a CEO’s age, gender, tenure, and wealth, and firm factors such as firm size, market-to-book ratio, and debt level, which may influence the managerial overconfidence level, were collected for the regression analysis. Three different option price measures for indicating the differential levels of managerial confidence were used for tests of robustness. Firm, industry, and CEO fixed effects are imposed.

We first examined if the GFC and SOX changed the level of managerial overconfidence for the whole CEO sample that we collected. Specifically, we constructed the managerial overconfidence continuous variable to allow it to span from several cutoff points below the overconfidence 67 level to several cutoff points above the overconfidence 67 level. We then restricted the level of managerial overconfidence to change but only for the cutoff points above the overconfidence 67 continuum (for instance, from overconfidence 67 to the overconfidence 85 mark). This measure adhered strictly to the concept of state-contingent managerial overconfidence, showing that an overconfident individual can become even more overconfident. The regressions above provide direct evidence of the state-contingent effect of the GFC and SOX on the managerial overconfidence level for a homogenous group of CEOs.

2.2. Natural Experiment from a Real Market

To capture evidence of state-contingent overconfidence while mitigating potential endogeneity effects, we adopted the following model for the regression:

Loan size or Loan duration = f (Overconfidence level × GFC (or SOX); firm controls; CEO controls; loan characteristics; error term)

(2)

The interaction term above between the overconfidence level and the GFC, for instance, can be interpreted as follows. The interaction term implies that the GFC may synergize with managerial overconfidence to yield a joint effect additional to the individual effects put together. Given that the GFC is an exogenous shock, it would impact on the overconfidence level, but not vice versa. A significant interaction term therefore implies that the GFC acted on the manager’s overconfidence level, hence capturing the state-contingent managerial overconfidence effect.

Previous studies examine, for instance, managerial overconfidence effects on loan covenant usage (Voon et al. 2022). There is a paucity of research investigating the effects of CEO overconfidence on loan contract terms such as loan amount and loan duration. This paper bridges this gap in the literature. We controlled for firm characteristics in our baseline and interaction term models, including firm size, profitability, tangibility, market-to-book ratio, loan maturity, leverage, S&P rating, cash flow, and z-score, following Graham et al. (2008) and Kim et al. (2011). Appendix A provides the definitions and constructions of all the measured variables used in our empirical analysis in Study 1 over the period 1996–2011.

2.3. Measures

We measured managerial overconfidence by using CEOs’ option exercise behavior (as in Banerjee et al. 2015; Billet and Qian 2008; Campbell et al. 2009; Malmendier and Tate 2005). We followed Campbell et al. (2009) in calculating the average moneyness of the CEO’s option portfolio for each year over the period 1996–2011. First, for each CEO-year, we calculated the average realizable value per option by dividing the total realizable value of the options by the number of options held by the CEO. The strike price was calculated by subtracting the average realizable value from the fiscal year end stock price. The average moneyness of the options was then calculated by dividing the stock price and subtracting one from the result (following Hirshleifer et al. 2012). As we are only interested in options that the CEO could exercise, we included only the vested options held by the CEO. Overconfidence 67 is a dummy equal to 1 if the CEO held at least 67% in-the-money options.

Different levels of managerial overconfidence were measured by the extent to which the manager postponed the exercise of vested options. To achieve this, a continuous variable known as a confidence index was constructed. The analysis was conducted using the overall CEO data sample as well as using only the overconfident CEO sample. For the overall CEO sample, the managerial confidence continuum changed from the normal confidence level continuously up to the overconfidence level. For the overconfident CEO sample, three levels of managerial overconfidence were constructed, with several cutoff points ranging from overconfidence 67 (a weak form of overconfidence) to overconfidence 120 (a very strong form of overconfidence).

3. Data Collection

Our primary data source in relation to bank loan size was DealScan, which included details of loan size, loan purpose, and collateral provisions. Firm data were extracted from CompuStat, and included asset size, profitability, liabilities, S&P rating, etc. The ExecuComp database provided information on CEOs, their education level, gender, and their option compensations, which were used to construct measures on the level of managerial confidence using the different option-based methods. The firm size is important in loan evaluation as larger firms often have better reputations, which gives them more clout to negotiate better terms for their loans. Profitability is the ratio of earnings before interest, taxes, depreciation, and amortization (EBITDA) to total assets. It controls for different abilities of firms to make a profit. Firms with higher profitability are expected to have a lower default risk. Tangibility is the ratio of net property, plant, and equipment to total assets, which measures the quality of loan collateral. As creditors usually have the right under a security document to enforce the security and take over a firm’s secured assets in an event of default, more tangible assets should lower the borrowing cost and covenant usage. The market-to-book ratio is derived by dividing the sum of market value of equity and book value of loans by total assets, a proxy for controlling firms’ growth opportunities. Leverage is the ratio of long-term debt to total assets, which measures the financial status of firms. Firms with higher leverage have higher default risk, which increases their borrowing cost. In addition to the above, we controlled for the z-score (Murfin 2012). The summary statistics of our data are presented in

Table 1. Summary statistics.

Variable	N	Mean	Standard Deviation	Min	Max
Overconfidence Measures
Overconfidence 67	6045	0.2967742	0.4568739	0	1
Overconfidence Index	6045	0.9680728	1.185538	0	3
Average Option Price	6044	0.2663608	0.2641887	0	4.154439
Loan Characteristics
Ln(Loan Amount)	6043	19.56283	1.224556	13.99783	24.12446
Ln(Maturity)	5745	3.593328	0.6986203	0.6931472	5.484797
Secured	6045	0.262531	0.440046	0	1
Firm Characteristics
Ln(Asset)	6045	7.882641	1.497603	4.611431	12.39703
Market-to-book	6045	1.850499	1.036431	0.7120423	7.170837
Profitability	6044	0.1468398	0.0756185	−0.0818947	0.4362031
Tangibility	6045	0.3388973	0.2284383	0.0171859	0.8953143
Z-score	5776	2.618075	2.553105	−0.8035191	17.5571
Ln(Firm Age)	6045	3.251725	0.6949042	0.6931472	4.110874
Cash Flow Vol.	5614	63.2186	136.4043	0.6073399	1502.4

.

4. Results and Discussion

4.1. Results of the Baseline Model

Evidence of state-contingent managerial confidence can be obtained directly by observing whether the self-confidence level of the managers changed in accordance with the state of the macroeconomy as mimicked by a financial crisis or a financial regulation. Our empirical data from about 4570 managers across different industries show that the managerial self-confidence levels were negatively and significantly influenced by the GFC and by the SOX (see

Table 2. Baseline results: effects of the GFC and SOX on managerial overconfidence.

	(1)	(2)	(3)	(4)	(5)	(6)
	Overconfidence Index	Avg Option Price	Overconfidence 67	Overconfidence Index	Avg. Option Price	Overconfidence 67
GFC	−0.0987 **	−0.0276 ***	−0.0370 **
	(−2.23)	(−2.81)	(−2.10)
Ln(CEO Age)	−0.253 *	−0.0431	−0.108 *	−0.281 **	−0.0482	−0.120 **
	(−1.80)	(−1.38)	(−1.93)	(−2.01)	(−1.54)	(−2.14)
Ln(Tenure)	0.137 ***	0.0277 ***	0.0513 ***	0.132 ***	0.0268 ***	0.0491 ***
	(6.78)	(6.16)	(6.37)	(6.51)	(5.94)	(6.09)
Male	0.382 ***	0.0859 ***	0.105 *	0.369 ***	0.0833 ***	0.1000 *
	(2.71)	(2.74)	(1.88)	(2.62)	(2.66)	(1.78)
Ln(Wealth)	0.0491 **	0.0111 **	0.0128	0.0709 ***	0.0151 ***	0.0217 **
	(2.34)	(2.39)	(1.53)	(3.26)	(3.13)	(2.51)
Firm Size	−0.0454 ***	−0.00472	−0.0190 ***	−0.0507 ***	−0.00572 *	−0.0211 ***
	(−3.01)	(−1.41)	(−3.16)	(-3.35)	(−1.70)	(-3.51)
Market-to-book	0.986 ***	0.221 ***	0.334 ***	0.980 ***	0.221 ***	0.331 ***
	(25.93)	(26.20)	(22.05)	(25.85)	(26.20)	(21.95)
Lagged Cash Flow	0.0476 ***	0.0107 ***	0.0169 **	0.0457 ***	0.0101 ***	0.0163 **
	(2.74)	(2.78)	(2.45)	(2.64)	(2.63)	(2.36)
SOX				−0.157 ***	−0.0325 ***	−0.0628 ***
				(−4.29)	(−3.99)	(−4.31)
Constant	0.718	0.104	0.389 *	0.765	0.113	0.409 *
	(1.26)	(0.82)	(1.71)	(1.34)	(0.89)	(1.80)
N	4572	4571	4572	4572	4571	4572
R2	0.1807	0.1799	0.1396	0.1832	0.1813	0.1423

This table contains the OLS results using overconfidence levels spanning from underconfidence (option price threshold below 67%) to overconfidence (option price threshold above 67%) as the dependent variables and the external shocks (GFC and SOX) as the independent variables. Contingent-based self-confidence is evidenced by the direct effect of the GFC and SOX on managerial self-confidence. t-statistics are shown in parentheses; * p < 0.10, ** p < 0.05, *** p < 0.01.

), demonstrating state-contingent overconfidence, even after controlling for various CEO/demographic, firm, and loan characteristics that may also correlate with managerial overconfidence. The coefficients were significant after controlling for CEO, the firm, and industry fixed effects. Our results are robust to the three continuous option-based measures of managerial overconfidence.

4.2. Results of the Natural Experiment

We first conducted a natural experiment using actual financial market data to investigate if the managerial overconfidence level interacted with an exogenous shock such as the GFC or SOX. Since both the GFC and SOX are exogenous shocks, the interaction effect between overconfidence and the GFC should be exclusively driven by the impact of the GFC on the overconfidence level, hence capturing the state-contingent overconfidence effect. The interaction effects between the overconfidence level and the SOX using the overall CEO sample are reported in

Table 3. Effects of the SOX–overconfidence interaction on loan amounts.

	(1)	(2)	(3)
Variables	Ln(Loan Amount)	Ln(Loan Amount)	Ln(Loan Amount)
Overconfidence 67	0.0280
	(0.67)
Overconfidence 67 x SOX	0.170 ***
	(3.50)
SOX	−0.0644 **	−0.0751 **	−0.0763 **
	(−2.28)	(−2.30)	(−2.53)
Firm Size	0.642 ***	0.640 ***	0.642 ***
	(89.86)	(89.57)	(89.88)
Leverage	0.509 ***	0.507 ***	0.505 ***
	(7.00)	(6.96)	(6.96)
Profitability	1.566 ***	1.548 ***	1.539 ***
	(10.68)	(10.52)	(10.48)
Tangibility	−0.426 ***	−0.420 ***	−0.422 ***
	(−9.08)	(−8.96)	(−9.01)
Z-score	−0.0218 ***	−0.0224 ***	−0.0230 ***
	(−4.24)	(−4.31)	(−4.45)
Ln(Loan maturity)	0.257 ***	0.257 ***	0.255 ***
	(16.98)	(16.98)	(16.88)
Avg Option Price		0.0888
		(1.27)
Avg Option Price x SOX		0.232 ***
		(2.82)
Overconfidence Index			0.0177
			(1.12)
Overconfidence Index x SOX			0.0653 ***
			(3.52)
Constant	13.46 ***	13.46 ***	13.47 ***
	(149.09)	(148.23)	(148.97)
N	5588	5587	5588
R2	0.6236	0.6228	0.6242

This table contains the results of OLS regressions between loan amount as the dependent variable and the interactions of the SOX and self-confidence as the independent variables. The dependent variable in columns (1) to (3) is expressed as the natural log of the loan amount. The interaction terms provide evidence of potential changes in managerial confidence levels brought by the external shock. Numbers in parentheses are standard error. *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.

and the interaction effects between the overconfidence level and the GFC using the same data sample are presented in

Table 4. Effects of the GFC–overconfidence interaction on loan amounts.

	(1)	(2)	(3)
Variables	Ln(Loan Amount)	Ln(Loan Amount)	Ln(Loan Amount)
Overconfidence 67	0.130 ***
	(5.21)
Overconfidence 67 x GFC	0.144 **
	(2.25)
Fin. Crisis	−0.0126	−0.0433	−0.0202
	(−0.39)	(−1.16)	(−0.58)
Firm Size	0.639 ***	0.637 ***	0.639 ***
	(90.56)	(90.39)	(90.61)
Leverage	0.516 ***	0.512 ***	0.511 ***
	(7.16)	(7.10)	(7.10)
Profitability	1.560 ***	1.546 ***	1.533 ***
	(10.63)	(10.52)	(10.44)
Tangibility	−0.419 ***	−0.416 ***	−0.415 ***
	(−8.95)	(−8.89)	(−8.89)
Z-score	−0.0224 ***	−0.0228 ***	−0.0237 ***
	(−4.38)	(−4.42)	(−4.61)
Ln(Loan Maturity)	0.253 ***	0.253 ***	0.252 ***
	(17.05)	(17.06)	(16.98)
Avg Option Price		0.206 ***
		(4.73)
Avg Option Price x GFC		0.291 ***
		(2.78)
Overconfidence Index			0.0561 ***
			(5.77)
Overconfidence Index x GFC			0.0531 **
			(2.21)
Constant	13.45 ***	13.45 ***	13.45 ***
	(148.99)	(148.80)	(149.06)
N	5588	5587	5588
R2	0.6231	0.6228	0.6237

This table contains the results of OLS regressions between loan amount as the dependent variable and the interactions of the GFC and CEOs’ self-confidence as the independent variables. The dependent variable in columns (1) to (3) is expressed as the natural log of the loan amount. The interaction terms provide evidence of potential changes in managerial self-confidence levels brought by the external shock. Numbers in parentheses are standard error. *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.

. The coefficient of the interaction term (confidence level * SOX) in Table 3 and the coefficient of the interaction term (confidence level*GFC) in Table 4 are reported to be positive and highly significant, implying that the exogenous shock impacted on the overconfidence level, which then raised the loan amount offered by the creditors.

Using the same model as above, we also conducted regression analyses using only the overconfident CEO sample. We examined if the overconfidence level of an overconfident CEO changed with the SOX or GFC. Similar results were obtained (see

Table 5. Overconfidence subsample: effects of SOX–overconfidence interactions on loan amounts.

	(1)	(2)	(3)
Variables	Ln(Loan Amount)	Ln(Loan Amount)	Ln(Loan Amount)
Overconfidence 67	−0.0174
	(−0.38)
Overconfidence 67 x SOX	0.146 ***
	(2.71)
SOX	−0.0595 *	−0.0477	−0.0601 *
	(−1.79)	(−1.31)	(−1.68)
Firm Size	0.660 ***	0.658 ***	0.659 ***
	(81.16)	(80.94)	(81.18)
Leverage	0.264 ***	0.256 ***	0.263 ***
	(3.59)	(3.43)	(3.59)
Profitability	1.383 ***	1.394 ***	1.360 ***
	(8.53)	(8.40)	(8.37)
Tangibility	−0.313 ***	−0.310 ***	−0.312 ***
	(−6.14)	(−6.06)	(−6.13)
Z-score	−0.0302 ***	−0.0314 ***	−0.0312 ***
	(−6.02)	(−5.85)	(−6.20)
Ln(Loan Maturity)	0.270 ***	0.270 ***	0.268 ***
	(15.78)	(15.81)	(15.66)
Avg Option Price		0.0423
		(0.66)
Avg Option Price x SOX		0.132 *
		(1.65)
Overconfidence Index			0.00688
			(0.39)
Overconfidence Index x SOX			0.0477 **
			(2.32)
Constant	13.31 ***	13.30 ***	13.31 ***
	(127.92)	(126.60)	(127.50)
N	4134	4132	4134
R2	0.6465	0.6446	0.6468

This table contains the results of OLS regressions between loan amount as the dependent variable and the interactions of the SOX and CEO overconfidence as the independent variables. The dependent variable in columns (1) to (3) is expressed as the natural log of the loan amount. The results provide evidence on the change in managerial overconfidence among the overconfident CEOs only. Numbers in parentheses are standard error. *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.

and

Table 6. Overconfidence subsample: effects of GFC–overconfidence interactions on loan amounts.

	(1)	(2)	(3)
	Ln(Loan Amount)	Ln(Loan Amount)	Ln(Loan Amount)
Overconfidence	0.113 ***
	(4.37)
Overconfidence x GFC	0.143 **
	(2.31)
GFC	0.0134	−0.00753	0.0112
	(0.41)	(−0.20)	(0.32)
Firm Size	0.648 ***	0.647 ***	0.648 ***
	(80.52)	(80.47)	(80.61)
Leverage	0.370 ***	0.368 ***	0.368 ***
	(4.62)	(4.58)	(4.59)
Profitability	1.613 ***	1.588 ***	1.586 ***
	(9.55)	(9.38)	(9.38)
Tangibility	−0.341 ***	−0.335 ***	−0.339 ***
	(−7.04)	(−6.92)	(−7.00)
Z-score	−0.0215 ***	−0.0213 ***	−0.0226 ***
	(−3.89)	(−3.81)	(−4.08)
Ln(Loan maturity)	0.434 ***	0.433 ***	0.432 ***
	(15.11)	(15.05)	(15.05)
Avg Option Price		0.190 ***
		(4.07)
Avg Option Price x GFC		0.243 **
		(2.39)
Overconfidence Index			0.0510 ***
			(4.97)
Overconfidence Index x GFC			0.0460 **
			(1.97)
Constant	12.04 ***	12.04 ***	12.04 ***
	(79.54)	(79.39)	(79.56)
N	4346	4345	4346
R2	0.6745	0.6743	0.6749

This table contains the results of OLS regressions between loan amounts as the dependent variable and the interactions between the GFC and CEO overconfidence as the independent variables. The dependent variable in columns (1) to (3) is expressed as the natural log of the loan amount. The results provide evidence of the change in managerial overconfidence level among the overconfident CEO sample only. Numbers in parentheses are standard error. *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.

). The reported results are consistent with findings in the literature (Voon et al. 2022) that loan default risk level decreases with the decline in managerial self-confidence level and creditors would therefore increase the loan size accordingly.

The above model was replicated to explore the effects of state-contingent change in managerial overconfidence on loan duration (another important loan contract term used by creditors). The interaction effects between the overconfidence level and the SOX using the overall CEO sample are reported in

Table 7. Effects of SOX-overconfidence interactions on loan duration.

	(1)	(2)	(3)
Variables	Ln(Loan Duration)	Ln(Loan Duration)	Ln(Loan Duration)
Overconfidence 67	0.0574 ***
	(3.59)
Overconfidence 67 x SOX	−0.0324 *
	(−1.68)
Avg Option Price		0.0706 ***
		(2.99)
Avg Option Price x SOX		−0.0113
		(−0.37)
Overconfidence Index			0.0234 ***
			(3.85)
Overconfidence Index x SOX			−0.0121 *
			(−1.65)
SOX	0.114 ***	0.105 ***	0.116 ***
	(8.65)	(7.27)	(8.29)
Firm Size	−0.0450 ***	−0.0449 ***	−0.0450 ***
	(−8.46)	(−8.45)	(−8.48)
Leverage	0.113 ***	0.113 ***	0.114 ***
	(4.12)	(4.12)	(4.14)
Profitability	0.390 ***	0.392 ***	0.386 ***
	(7.20)	(7.23)	(7.11)
Performance Price	0.0285 ***	0.0280 ***	0.0282 ***
	(3.13)	(3.08)	(3.09)
Collaterals	0.0320 ***	0.0323 ***	0.0320 ***
	(2.88)	(2.91)	(2.88)
Ln(Loan Amount)	0.114 ***	0.114 ***	0.113 ***
	(18.67)	(18.69)	(18.65)
Ln(Loan Spread)	−0.0480 ***	−0.0480 ***	−0.0479 ***
	(−6.49)	(−6.49)	(−6.48)
Ln(Credit Spread)	−0.266 ***	−0.264 ***	−0.265 ***
	(−8.89)	(−8.82)	(−8.84)
Constant	2.493 ***	2.493 ***	2.491 ***
	(23.54)	(23.54)	(23.53)
N	6219	6217	6219
R2	0.7483	0.7484	0.7485

This table contains the results of OLS regressions with fixed effects between loan maturity as the dependent variable and the interactions of the SOX and CEO overconfidence as the independent variables. The dependent variable in columns (1) to (3) is expressed as the natural log of the loan maturity. The results provide evidence on the change in managerial overconfidence among overconfident CEOs only. Numbers in parentheses are standard error. *, and *** denote significance at the 1%, and 10% levels, respectively.

and the interaction effects between the overconfidence level and the GFC using the overall data sample are presented in

Table 8. Effects of GFC–overconfidence interactions on loan duration.

	(1)	(2)	(3)
	Ln(Loan Maturity)	Ln(Loan Maturity)	Ln(Loan Maturity)
Overconfidence	0.0388 ***
	(3.89)
Overconfidence x GFC	−0.0956 **
	(−2.25)
Avg Option Price		0.0707 ***
		(4.33)
Avg Option Price x GFC		−0.216 ***
		(−3.28)
Overconfidence Index			0.0173 ***
			(4.51)
Overconfidence Index x GFC			−0.0493 ***
			(−3.21)
GFC	−0.00385	0.00543	0.00175
	(−0.29)	(0.39)	(0.13)
Firm Size	−0.0390 ***	−0.0391 ***	−0.0389 ***
	(−7.35)	(−7.38)	(−7.34)
Leverage	0.0738 ***	0.0754 ***	0.0748 ***
	(2.69)	(2.76)	(2.73)
Profitability	0.420 ***	0.424 ***	0.416 ***
	(7.71)	(7.80)	(7.64)
Performance Price	0.0290 ***	0.0293 ***	0.0290 ***
	(3.16)	(3.20)	(3.17)
Collaterals	0.0355 ***	0.0351 ***	0.0350 ***
	(3.17)	(3.14)	(3.13)
Ln(Loan Amount)	0.115 ***	0.115 ***	0.115 ***
	(18.86)	(18.84)	(18.79)
Ln(Loan Spread)	−0.0425 ***	−0.0420 ***	−0.0419 ***
	(−5.69)	(−5.63)	(−5.62)
Ln(Credit Spread)	−0.128 ***	−0.118 ***	−0.121 ***
	(−4.48)	(−4.10)	(−4.22)
Constant	2.352 ***	2.344 ***	2.350 ***
	(22.30)	(22.25)	(22.30)
N	6219	6217	6219
R2	0.7450	0.7454	0.7454

This table contains results of OLS regressions between loan maturity as the dependent variable and interactions between the GFC and overconfidence as independent variables. The dependent variable in columns (1) to (3) is expressed as the natural log of loan maturity. The results provide evidence of the change in managerial overconfidence level among the overconfident CEO sample only. Numbers in parentheses are standard error. *, **, and *** denote significance at 1%, 5%, and 10% levels, respectively.

. The coefficient of the interaction term (confidence level * SOX) in Table 7 and the coefficient of the interaction term (confidence level*GFC) in Table 8 are reported to be significant, implying that the exogenous shock interacted with (by impacting on) the overconfidence level to significantly alter the duration of the loan offered by the creditors.

5. Study 2: Empirical Evidence from a Lab Experiment

5.1. Data Collection

In addition to real market data in Study 1, we also conducted a lab experiment in Study 2 to examine if individuals’ self-confidence level can be influenced by external factors. The questionnaires are designed for testing over-precision, one of the distinctions of overconfidence in psychology where over-precision reflects the confidence of people knowing the truth or reality. By snowball sampling, a total of 82 college undergraduate students, 44 from psychology and 38 from economics, were recruited to participate in Study 2. Data were collected in Autumn 2018. Following prior studies (Ulkumen et al. 2008; Anderson et al. 2012), participants were randomly assigned to two conditions, namely, the high-confidence condition (n = 41) and the low-confidence condition (n = 41). In both conditions, participants were asked to complete a 10-item Interval Production Task (Langnickel and Zeisberger 2016). They were presented with 10 descriptions of numerical general knowledge questions, and were asked to estimate the answer for each question, and texts served as an anchor. They needed to provide a lower bound and upper bound estimation for each question. To give an example, participants were asked to estimate the population for the city of Berlin, and give the lower bound (e.g., 2 million) and upper bound (e.g., 5 million) of their estimation. Participants were reminded not to use the internet to search for their answers. Before participants started to complete the task, participants in each condition were manipulated to form differential expectations to start with. In the low-confidence condition, participants formed a low initial expectation. Specifically, they were told that other students in their university had also completed the same general knowledge questions before, and results showed that the vast majority of them were very inaccurate in estimating the answers (i.e., other students could not answer the questions correctly). In the high-confidence condition, participants formed a high initial expectation. Specifically, they were told that the other students who had completed the task were very accurate in estimating the answers (i.e., other students could answer the questions correctly). After forming differential expectations, participants in both conditions started to complete the 10 general knowledge questions and made the corresponding estimations. Following the method from Anderson et al. (2012), we provided differential feedback to remind participants about their performance on the task halfway through the task (i.e., on the fifth question). For instance, in the low-confidence condition, participants received overly negative feedback. They were told: “Unfortunately, comparing to other students in your university, your performance in the last 5 questions is WORSE than 77% of the students”. In the high-confidence condition, participants received overly positive feedback. They were told: “Congratulations! Comparing to other students in your university, your performance in the last 5 questions is BETTER than 77% of the students”. After the differential feedback, participants continued to complete the remaining five questions. It should be noted that for each of the 10 questions, besides providing their answer and bound estimations, participants were further asked to rate their subjective general knowledge level: “How do you compare your general knowledge level with other students? Please rate on a scale from 1 (I am at the very bottom; worse than 99% of the students in this study) to 100 (I am at the very top; better than 99% of the others in this study)”, and to rate their subjective task performance: “How do you compare your task scores with other students? Please rate on a scale from 1 (I am at the very bottom; worse than 99% of the students in this study) to 100 (I am at the very top; better than 99% of the people in this study)”. Our analyses were mainly based on participants’ rating on their confidence level on general knowledge and task performance. At the end, participants were debriefed and thanked.

5.2. Results and Discussion

In order to examine if external factors would affect the self-confidence level of the participants, we employed the differential expectations and the differential feedback methods to simulate external environments/factors.

Table 9. Independent-sample t-test between high- and low-confidence conditions.

		High-Confidence Condition (n = 41)		Low-Confidence Condition (n = 41)
		Mean	SD	Mean	SD	t-Test
Differential Expectation	Subjective general knowledge level	58.0	16.1	46.5	19.8	−1.94 ^
	Subjective task performance	59.5	15.5	46.5	19.2	−2.27 *
Differential Feedback	Subjective general knowledge level	55.8	17.0	36.3	21.2	−3.10 **
	Subjective task performance	58.5	14.3	35.9	20.3	3.93 ***
Overall	Subjective general knowledge level	56.9	16.1	41.4	18.1	−2.75 **
	Subjective task performance	59.0	14.3	41.2	17.4	−3.40 **

This table provides evidence of contingent-based overconfidence from the laboratory experiments. Differential expectations and differential feedback methods (from the psychology literature) were used to test whether or not individuals’ self-confidence level can be changed. * p < 0.05. ** p < 0.01. *** p < 0.001. ^ Due to the sample size, the statistical result was marginally below the level of significance of p = 0.05.

shows the results of the lab experiment. We calculated participants’ mean self-reported confidence level for the high-confidence condition (column 3 of Table 9) and low-confidence condition (column 5 of Table 9), respectively. A simple t-test was conducted (column 7), showing that the confidence level in the high-confidence group was significantly higher than that of the low-confidence group. This is true regardless of whether participants were asked to rate their general knowledge or task performance (see Table 9). Overall, our laboratory data showed that the confidence level of individuals could indeed be affected by external manipulation, indicating that individual confidence can be state-contingent.

6. Conclusions and Implications

State-contingent managerial overconfidence is a new idea in finance. This is the first time (in this paper) that the state-contingent overconfidence hypothesis has been tested. An understanding of this concept and its application constitute our original contributions to the finance literature.

The prior finance literature has usually assumed that overconfidence is a character trait that is stable over time. Consequently, most finance research compared differences in behavioral effects between two different individuals or between overconfident and underconfident groups of managers. In this paper, we reported two empirical studies. Our Study 1 based on real market data and Study 2 based on experimental lab data consistently showed that individuals’ overconfidence could be state-contingent, i.e., the confidence level of individuals (e.g., overconfident managers in Study 1, and college students in Study 2) could be altered by external factors (e.g., the state of the macroeconomy in Study 1, and differential expectations and feedback in Study 2).

The new state-contingent managerial overconfidence hypothesis presented in this paper has important practical implications for the credit market. First, this new proposition implies that since overconfidence could be state-contingent, creditors should review or renegotiate their loan credit terms (such as the loan size and loan duration) with CEOs more frequently according to drastic changes in the external state of the macroeconomy. Second, creditors should not overreact or over-reduce the loan amount or the loan duration, for instance, too excessively in response to a financial crisis that is perceived to raise the systemic risk. This is because a financial crisis has the opposite offsetting effect to lower the level of overconfidence and hence overinvestment risk and the risk to creditors. When the attenuating state-contingent overconfidence risk effect during a financial crisis is overlooked or misunderstood, creditors may alter the loan contract terms by more than the efficient amounts. Third, knowledge of state-contingent managerial overconfidence effects would reduce the overall inefficiency inherent in the credit market, suggesting that creditors such as banks should review the profiles of companies’ senior officers and weighting of firms’ credit risk assessment more frequently, since credit risk aroused by overconfident CEOs could be mitigated by external economic shocks. Bank loans are an important source of capital for many big corporations. Overreaction of banks during a financial crisis, such as heavy reduction in loan size or imposition of very strict loan covenants, could slow down economic recovery, as these actions hinder firms’ after-crisis growth.

There is a limitation in this paper. In our Study 2, we resorted to using a laboratory experiment to examine if one’s confidence level can be affected by external influence. In psychology, overconfidence is a multi-faceted phenomenon, of which major variants include over-precision, over-placement, and over-optimism. Our lab experiment is based specifically on over-precision. However, it should be noted that in the finance literature, overconfidence is defined as non-exercise of in-the-money vested options (Malmendier et al. 2011; Voon et al. 2022). There is no distinction between over-precision, over-placement, and over-optimism: these individual variants are hard to measure using real market data. Thus, caution should be taken to avoid generalization.

A couple extensions may be contemplated for future research. First, our paper currently focuses on the GFC and SOX: these exogenous shocks are reported to have lowered the overconfidence level of senior managers. Other possible external environmental factors may also contribute to variations in managerial overconfidence. For instance, the COVID-19 pandemic, which erupted in January 2020, may also constitute another exogenous shock that may alter the overconfidence level of senior managers. Second, our analysis currently focuses on the CEOs of the Fortune 500 companies. Future research may be extended more generally to managers of smaller companies. Likewise, the lab experiment may be extended to subjects encompassing business managers as the questionnaire respondents. Third, we currently used the non-exercise of options as our measure of managerial overconfidence. Future research may be extended to using multi-faceted measures (such as over-precision) as adopted in most psychological studies.