1

MINISTRY OF EDUCATION AND TRAINING
UNIVERSITY OF ECONOMICS HO CHI MINH CITY
--------------------

TRUONG DINH BAO LONG

RESEARCH ON CEO’s OVERCONFIDENT BEHAVIOR
AND FINANCIAL DECISIONS IN VIETNAM COMPANIES

Major:

Finance – Banking

Major code: 93 40201

DOCTORAL THESIS SUMMARY

HOCHIMINH CITY – 2018


2

The thesis is completed at:
University of Economics Ho Chi Minh City

Supervisor:

Associate Professor. Ph.D Nguyen Ngoc Dinh


Truong Dinh Bao Long

Supervisor:

Associate Professor. PhD Nguyen Ngoc Dinh,

Course: NCS2010

Keywords:
Overconfidence, Financial Condition Index, Financial
Decisions, Regression Model, CEO (Chief Executive Officer).


4

CHAPTER 1. INTRODUCTION TO THE APPROACH OF THE
THESIS
1.1. The necessity of the Thesis
Nowadays, in Vietnam, CEOs, who directly manage business activities, are often
struggled to make good decisions. Insufficient and inconsistent data, untransparent
market, asymmetric information are the main causes leading to biased CEOs’
decisions. This problem may become very serious and create agency cost for
stockholders, it indirectly reduces the value of the company. In other cases,
companies can be distressed and bankrupted if such problems are not solved
completely.
Although doing research on CEOs’ behaviors is very important in order to contribute
to the theoretical and empirical framework in making right decisions, it is only
implemented in the U.S market. Therefore, it is necessary to implement and conduct
a similar research in Vietnam market because the research can reveal a specific
relationship between behavioral psychology and financial theories.

domestic scholars. For example, Nguyen Ngoc Dinh (2015) and Le Dat Chi (2015)
find out some weak evidence about overconfidence can affect financial decisions.
However, the methodology of overconfidence measurement is still a challenge in
Vietnam. In this thesis, overconfidence measurement is improved significantly to
those two previous papers to avoid the endogeneity.
1.3. The aim of the Thesis
Determining the existence of the relationship between overconfidence and investment
decision in Vietnam. Determining the existence of the relationship between
overconfidence and investment decision in Vietnam under the effect of the financial
conditions. Determining the existence of the relationship between overconfidence


6

and funding decision in Vietnam. Determining the existence of the relationship
between overconfidence and funding decision in Vietnam under the effect of a
financial deficit. Determining the existence of the relationship between
overconfidence and dividend decision in Vietnam. Determining the existence of the
relationship between overconfidence and dividend decision in Vietnam under the
effect of growth opportunity.
1.4. The methodology of the Thesis
The thesis uses regression models on the basis of the previous papers in the U.S
market. The variables and the way to collect data are similar to obtain the aim of the
thesis.
Space dimension of the sample: Sample size includes 136 Vietnamese businesses full
of the audited financial statement and publicly listed in Ho Chi Minh Stock Exchange
and Ha Noi Stock Exchange.
Time dimension of the sample: The data is collected from 2007 to 2016, in which four
first years from 2007 to 2010 will be used to calculate net buying position to confirm
the behavior of CEOs.

THEORETICAL

FRAMEWORK AND

LITERATURE

REVIEW
2.1 Theories of Behavioral Finance about the Biased Recognition
2.1.1. Incomplete rational: Restrictive rational
Some theoretical behavioral finance models show that subjects do not adjust their
belief in the right way when new information arrives. It means the adjustment does
not follow the Bayes rule. Other models believe that subjects’ belief flows Bayes
rules but their behavior is not appropriate with the expected subjective utility.
Therefore, these are the motivation of researchers to discover psychological


8

recognition of the subjects. From that, behavioral financial models (behavioral
economics) are based on the irrational behavior of the subjects resulting from the
biasness in the personal belief and interest. The models are designed to explain how
subjects can create their expectations.
2.1.2. Cognitive psychology: Biased patterns in Behavior
Heuristics or the rule of experiences is a thinking activity based on experienced
events, it helps the subject make decisions easier. Cognitive accounting is a type of
behavior which helps the subject separate different decisions of the same resource.
This term is first expressed by Thaler (1985), when he tries to depict a process of
human in which the economic results are standardized, classified and evaluated.
Later, Ritter (2003) contribute a descriptive process to make the model complete.
Another term given by Shefrin and Thaler, 1988 is framing. Framing is a social

Malmendier and Tate (2005a, 2005b, 2008) suggest two approaches to measure the
overconfidence of CEOs. The first approach is based on revealed beliefs or the option
and stock-based measure of CEO confidence. The second approach is based on the
way other people think of the CEO and is called the press-based measure of CEO
confidence.
2.3.1 Option and stock-based measure of CEO confidence
In this measure, there are some different methods: Holder 67, Holder 150,
Longholder and Net Buyer. CEO holds their options at a level that excess their
rational threshold of 67% or 150%, so those methods are Holder 67 and Holder 150,
respectively. CEO tries his or her best to hold the option 5 years longer from the
maturity point even though the value of options excesses their rational level. So, the
long holder measure is built on that basis. On the other hand, Net Buyer measure is
created to capture the tendency of buying more stocks of some CEOs although they
know the sensitivity of business risk.
2.3.2 Press-based measure of CEO confidence


10

This measure is based on the correlation between the number of articles about
overconfidence/optimistic and unconfidence/ pessimistic. According to Malmendier
and Tate (2005b, 2008), we can distinguish who is the overconfident CEO based on
the press. Following each CEO on each year, they collect keywords such as a)
Overconfident/

overconfidence,

b)

optimistic/

investigate biasness in funding decisions via optimism/overconfidence. Those models
give them clues that optimistic and overconfident CEOs will choose the higher
financial leverage rate. Hackbarth (2008) argues that overconfident CEOs believe the
cash flow in their companies are less variated, then their companies have lower
chances to get distressed despite that this is not true. In terms of dividend policy,


11

Allen and Michaely (2003), Bouwman (2009), Sanjay Deshmukh et al (2013) give
out empirical evidence that overconfident CEO pays less dividend than rational CEO.
However, under the effect of growth opportunity, overconfident CEO pay more
dividend than usual.

CHAPTER 3. THE METHODOLOGY OF THE THESIS
Hypothesis 1: In companies which have overconfident CEOs, the sensitivity of
investment-cashflow is higher than in companies which have rational CEOs.
The thesis tests the hypothesis 1 via checking whether CEO confidence increases the
sensitivity of investment-cashflow. The empirical model is based on the one creating
by Malmendier and Zheng (2012) as follows:
𝐼𝑖𝑡 = 𝛽1 + 𝛽2 𝐶𝐹𝑖𝑡 + 𝛽3 𝑂𝐶𝑖 + 𝛽4 𝑂𝐶𝑖 ∙ 𝐶𝐹𝑖𝑡 + 𝛽5 𝑋𝑖𝑡′ + 𝛽6 𝑋𝑖𝑡′ ∙ 𝐶𝐹𝑖𝑡 + 𝜀𝑖𝑡 (1)
Where Iit denotes the investment (or capital expense – capex) of company i in year t
using the initial year as the benchmark. CFit denotes the cash flow of company i in
year t, standardized and winsorized at 1% according to the asset value of the initial
year. OCi represents the CEO overconfidence of company i and is a dummy variable.
Finally, X’it is a set of controlling variables for CEO and company properties.
Hypothesis 2: The sensitivity of investment-cashflow in companies which have
overconfident CEOs is affected by the effect of the funding deficit.
The static model of investment (1) has an issue that it has not considered the dynamics
and persistence of investment. For example, investment at this moment can depend

Research on the influence of CEO overconfident on debt financing in the context of
the financial deficit under the approach of Malmendier et al. (2011) and Malmendier
and Zheng (2012). The equation is based on the theoretical framework of Shyam-


13

Sunder and Myers (1999) about the financial deficit of the company and is
empirically tested by Frank and Goyal (2003):
𝐷𝑒𝑏𝑡𝑖𝑡 = 𝛽1 + 𝛽2 𝐹𝐷𝑖𝑡 + 𝛽3 𝑂𝐶𝑖 + 𝛽4 𝑂𝐶𝑖 ∙ 𝐹𝐷𝑖𝑡 + 𝛽5 𝑋𝑖𝑡′ + 𝛽6 𝑋𝑖𝑡′ ∙ 𝐹𝐷𝑖𝑡 + 𝜀𝑖𝑡 (4)
Where Debtit denotes the net debt, which is calculated by using incremental debt of
the company i in the year t. FDit denotes the financial deficit of the company I in the
year t (it is calculated by adding up the cash dividend, net capital expense, net
working capital and then minusing for internal cashflow). Other variables have the
same as notations as the equation (1), (2) and (3) except for controlling variables.
Although the thesis still uses the same controlling variables as in the paper of Frank
and Goyal (2003) and Malmendier et al (2011), most variables are re-calculated on
the basis of the incremental changes. For instance, they are the change in profitability
(Δ𝑃𝑟𝑜𝑓𝑖𝑡𝑖𝑡 ), in tangible asset (Δ𝑇𝑎𝑛𝑔𝑖𝑡 ), in size of the company (Δln(𝑆𝑎𝑙𝑒𝑠)𝑖𝑡 ) and
in the growth opportunity (Δ𝑄𝑖𝑡 ).
Hypothesis 5: In companies which have overconfident CEOs, the dividend is less
than in companies which have rational CEOs.
Hypothesis 6: Rational CEOs pay more dividend than overconfident CEO under the
effect of growth opportunities.
According to Malmendier and Tate (2005, 2008) and Malmendier et al (2011), the
thesis classifies overconfident and rational CEO based on the investment level of
him/her in the company. The model presents the relationship between the CEO
overconfidence and dividend payment which results from the original paper of Sanjay
Deshmukh et al (2013):
𝐷𝐼𝑉𝑖𝑡 = 𝛽0 + 𝛽1 𝑂𝐶𝑖𝑡 + 𝛽2 𝑋𝑖′ (5)

0

1

Owp

816

0.0621

0.0971

0.0000

0.5501

Tenure

816

9.7471

4.1874

9

I

816


0.5603

Equity

816

0.0043

0.0281

-0.1609

0.2259

FD

816

0.0061

0.0696

-0.2713

0.5603

Profit

816


17.0552

Q

816

0.9560

0.3488

0.3785

3.3435

Lev_bk

816

0.3621

0.2548

0.0000

0.8725

Lev_mk

816


Properties of company

Financial conditional index
FCI

816


15

CHAPTER 4. THE RESULTS OF THE THESIS
After regressing a sample data including 136 non-financial listed company which is
collected from both HNX and HSX exchanges on the period from 2011-2016, the
thesis has found some profound evidence.
Firstly, in Vietnam, CEO confidence does not impact on the sensitivity of investment
and cashflow. This evidence seems to be different from that of Manmeldier et al
(2005, 2008, 2012). Secondly, the financial condition plays an important role in
reducing the sensitivity of investment-cashflow. Thirdly, based on effective and
efficient estimation models, the thesis finds out that CEO overconfidence increases
the sensitivity of using debt under the effect of financial deficit. Fourthly, although
there is no evidence to prove that CEO overconfidence affects the capital structure of
the company, under the effect of financial conditions there is a weak evidence that
overconfident CEOs reduce financial leverage of the company. Fifthly, overconfident
CEOs pay more dividend than rational ones do. Finally, under the effect of growth
opportunities, overconfident CEOs still pay more dividend than rational ones do.

Table 4.1 CEO overconfidence and the sensitivity of investment - cashflow
Dependent variable: I(t)

Independent variable:


OLS

OLS

OLS

OLS

CUE

0.2702

0.3436

0.3918

0.3750

1.1163

0.3051

(4.08)***

(2.53)**

(2.24)**

(1.99)**

-0.2927

0.0675

(-0.93)

(-0.72)

(-0.91)

(0.33)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes


for

test

for

AR(1) (p-value)
Arellano-Bond
AR(2) (p-value)

Yes

Yes

Yes

Yes

Yes

0.0277

-0.0229

-0.0277

-0.0790

0.0034


0.4999

0.5052

0.5155

0.2620

0.0806

0.0050

Sargan J test (p-value

0.0511

Hansen J test (p-value)

0.6434

All standard deviations (in the parenthesis) are adjusted errors to adapt to
heteroskedasticity. GMM-CUE is used because it is beneficial and effective to solve
the cases in which heteroskedasticity and autocorrelation exist. GMM-CUE use the
first lag of Q and all interactive variables with CF and dummy variable as its
instrument varibles. ***, **, * are significance levels at 1%, 5% and 10%,
respectively. Hausman test is used to check whether FEM or REM is more suitable.
(H0: REM is better).


17


SGMM

SGMM

SGMM

SGMM

0.2644

0.2871

0.2890

0.2800

0.2874

0.2919

(3.47)***

(4.18)***

(4.32)***

(4.01)***

(3.94)***


-0.0001

-0.0001

(0.93)

(1.01)

(0.87)

(-0.05)

(-0.01)

(-0.01)

-0.1072

-0.1318

-0.1306

0.0880

0.0829

0.0588

(-0.54)


0.0436

0.0382

0.0332

(1.21)

(1.29)

(0.93)

-0.9127

-0.8893

-0.8577

(-2.8)***

(-2.69)***

(-2.74)***

CEO-level Control

Yes

Yes


Yes

Yes

Yes

Firm-level Control*CF

Yes

Yes

Yes

Yes

Yes

Yes


18

Year Fixed Efffects

Yes

Year Fixed Efffects*CF


544

544

0.1500

0.1390

0.0361

Sargan J test (p-value)
Hansen J test (p-value)

Intercept

Obs
Arellano-Bond test for
AR(1) (p-value)
Arellano-Bond test for
AR(2) (p-value)

Yes

Yes

Yes

Yes

544


0.0059

0.2620

0.8590

0.6740

0.6010

0.8780

0.6320

0.3840

All standard deviations (in the parenthesis) are adjusted errors to adapt to
heteroskedasticity. GMM-IV is adjusted for the autocorrlation and error cases.
SGMM (System GMM) uses instrument variables including all controlling
variables and dummy variables. ***, **, * are significance levels at 1%, 5% and
10%, respectively. Hausman test is used to check whether FEM or REM is more
suitable. (H0: REM is better).

Table 4.3 CEO overconfidence and the capital structure
Dependent variable: Lev_mk
Independent
variable:

FD


GMM -IV

0.5623

0.2749

0.2654

0.2354

0.2118

0.2240

0.4425

0.4586

(5.55)***

(3.24)***

(3.22)***

(3.03)***

(2.19)**

(2.28)**

-0.6213

(-3.98)***

(1.10)

(1.44)

(0.74)

(0.67)

(0.71)

(-2.62)***

(-2.43)**

0.1190

-0.0260

-0.0247

-0.0122

-0.0236

-0.0315


0.0370

0.0556

0.0556

(4.47)***

(0.89)

(0.81)

(0.79)

(0.72)

(0.74)

(3.07)***

(2.79)***

-0.2018

-0.0334

-0.0369

0.0544


0.3654

-0.0262

-0.0263

(5.65)***

(5.93)***

(1.18)

(1.59)

(-0.21)

(-0.19)

-0.0524

-0.0483

-0.0459

-0.0198

-0.0015

(-2.71)***


-0.0129

(-5.81)***

(-5.76)***

(-3.19)***

(-2.9)***

-0.0149

-0.0282

0.0097

0.0109

(-0.6)

(-1.07)

(1.00)

(1.07)

OC

Return



(-1.84)*

-0.3393

-0.4994

-0.3697

-0.4658

0.7594

1.0353

-0.1133

-0.0954

(-1.4)

(-0.77)

(-0.78)

(-0.74)

(0.91)

(1.17)

Obs
R-squared
Hausman
test

816

816

816

816

816

816

544

544

0.2668

0.8771

0.8771

0.8831

0.8880


0.3270

ArellanoBond
for

test
AR(1)

(p-value)
ArellanoBond
for

test
AR(2)

(p-value)
Sargan J test
(p-value)
Hansen
test

J
(p-

value)

All standard deviations (in the parenthesis) are adjusted errors to adapt to heteroskedasticity.
GMM-IV is adjusted for the autocorrlation and error cases. GMM-IV uses instrument variables
including two differntial values of Retur and dummy variables. ***, **, * are significance


REM

FEM

GMM-CUE

0.8484

0.8454

0.8350

0.8091

0.3275

0.7362

(14.63)***

(11.96)***

(10.09)***

(9.29)***

(1.75)*

(5.04)***


(2.02)**

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

OC

Yes

-0.0034

0.0236

0.0233

0.0185

0.0106

-0.0006

(-2.72)***

(10.42)***

(8.77)***

(2.2)**

(1.14)

(-0.31)

816

816


3.01***

2.06**

3.27***

0.1137

0.0409

0.0778

0.0131

Log likelihood
χ2
Hausman test

0.0103


22

Arellano-Bond

test

for

test


Owp

OC

Q

CF

ln(Sales)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

FEM

FEM

REM


(0.89)

(0.92)

(-0.72)

3.4715

3.0757

3.5166

3.1923

3.5245

3.6316

61.2179

(2.12)**

(2.30)**

(1.99)**

(2.24)**

(2.12)**


(1.60)

0.4829

1.0373

0.7372

0.4854

0.3267

0.4686

1.8409

(0.21)

(0.29)

(0.36)

(0.14)

(0.09)

(0.13)

(0.41)

23

0.5531

Tenure

(1.82)*
0.6568

Lev_bk

(0.55)
-2.50e-14

Tang

(-0.03)
0.2972

OC*Q

(2.33)**
2.7280

OC*CF

(0.18)
-5.58e-13

OC*Tang

-13.9973

-14.7461

-73.7374

(-1.03)

(-1.28)

(-1.18)

(-0.08)

(-1.02)

(-1.07)

(-1.73)*

Yes

Yes

Yes

Yes

Yes


544

5180.49

4512.71

2511.24

4513.62

4476.41

4529.13

527.49***

283.21***

572.45***

422.22***

553.42***

387.62***

0.0316

0.0051


(p-value)

0.3671

0.897

All standard deviations (in the parenthesis) are adjusted errors to adapt to heteroskedasticity.
GMM-IV uses instrument variables including two differntial values of Return and dummy
variable. ***, **, * are significance levels at 1%, 5% and 10%, respectively. Hausman test is
used to check whether FEM or REM is more suitable. (H0: REM is better).

CHAPTER 5. THE CONCLUSIONS AND IMPLICATIONS OF THE THESIS
5.1. Main conclusions
Hypothesis 1: In companies which have overconfident CEOs, the sensitivity of
investment-cashflow is higher than in companies which have rational CEOs.
Although on the whole investment is affected by cash-flow, in both dynamic and
static models, there is no evidence of the relationship between CEO overconfidence
and the sensitivity of investment-cashflow. The interactive coefficient of CEO
overconfidence and cashflow has no statistical significance.
Hypothesis 2: The sensitivity of investment-cashflow in companies which have
overconfident CEOs is affected by the effect of the funding deficit. CEO
overconfidence affects the sensitivity of investment-cashflow under the effect of
macro-financial conditions. The thesis shows that CEO overconfidence has an
indirect role in that relationship. Particularly, the impact has the negative sign and
there are levels in the range of 0.86 to 0.91. That means CEO overconfidence reduces
the effect of (good) financial conditions on the sensitivity. This implies that


25