You are on page 1of 22

North American Journal of Economics and Finance 53 (2020) 101215

Contents lists available at ScienceDirect

North American Journal of Economics


and Finance
journal homepage: www.elsevier.com/locate/najef

The financial investment decision of non-financial firms in China


T
Chengsi Zhang , Ning Zheng

School of Finance & China Financial Policy Research Center, Renmin University of China, Beijing, PR China

ARTICLE INFO ABSTRACT

JEL classification: This paper constructs a portfolio model to analyze the determinants of the financial investment
E22 decision of non-financial firms in China. Unlike the literature assuming that financial investments
O16 are riskless, our model allows risks in both fixed and financial investments. We show that this
G11 extension provides an analytically similar but economically different model from the literature.
In particular, it is relative risk and risk-adjusted return gap, not pure risk and simple return gap
Keywords:
that enter into firms’ financial investment decision model. Using firm-level panel data of 1902
Financial investment
firms listed in Chinese stock market over the period from 2006 to 2016 with semi-annual fre-
Portfolio choice
Investment risk quency, we find that the ratio of fixed investment risk over total risk dominates financial in-
Return gap vestment decisions of non-financial firms. However, rates of risk-adjusted return gap between
Financialization financial and fixed investments play no role in Chinese firms’ financial investment decisions,
which is in stark contrast to the results using a model assuming riskless financial investments.
The baseline findings are robust to alternative measures of financialization and investment risk
and different firm sizes, ownership structures and time periods.

1. Introduction

Non-financial firms face a portfolio choice between real and financial assets. Given a certain amount of aggregate capital available
for firms’ investments, there is a substitution effect between the real and financial assets (Tobin, 1965). Accordingly, the rise of the
share of financial assets to aggregate capital, in contrast to the falling share of real assets, is referred to as financialization of non-
financial firms.1 Studies for developed economies in Stockhammer (2004), Crotty (2003), Epstein and Jayadev (2005), and Krippner
(2005), and studies for Argentina, Mexico and Turkey in Demir (2009) provide empirical evidence on the structural change in the
portfolio allocation decision of non-financial firms. The impact of such structural change has been identified as adverse to real sector
investments (Orhangazi, 2008; Demir, 2009; Barradas, 2017; Barradas & Lagoa, 2017; Tori & Onaran, 2017) and income inequality
(Alvarez, 2015), and hence is destructive to real economic growth (Moosa, 2018).
Since firms’ financialization generally exerts negative impact on real economic development, it is important to figure out reasons
why non-financial firms prefer financial investment when they make their decisions of investment allocations. In this respect, Demir
(2009) provides an important contribution to answer the question. In particular, Demir proposes a portfolio choice model to explain
why non-financial firms in Argentina, Mexico and Turkey favor financial assets vis-à-vis fixed assets, among other things. His results
appear to indicate that two main factors significantly drive the firms’ financialization behavior in the underlying countries: one is the


Corresponding author.
E-mail addresses: zhangcs@ruc.edu.cn (C. Zhang), victor_zheng@ruc.edu.cn (N. Zheng).
1
In a broader sense, financialization in micro level reflects important changes in the structure of non-financial firms’ balance sheets, including the
growth of income from financial subsidiaries and investment as well as growth in the transfer of earnings to financial markets in the forms of interest
payments, dividend payments and stock buybacks; see Orhangazi (2008) and Davis (2017).

https://doi.org/10.1016/j.najef.2020.101215
Received 19 January 2019; Received in revised form 6 March 2020; Accepted 22 April 2020
Available online 28 April 2020
1062-9408/ © 2020 Elsevier Inc. All rights reserved.
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

rate of return gap between financial and fixed investments and another is the uncertainty related to fixed investments. The results
seem to be reasonable since an intuitive impression is indeed that relatively higher return in financial assets motivates firms to do
more financial investments, and relatively higher risk in fixed assets drives firms to run away from fixed investments to financial
investments.
However, it should be noted that Demir (2009) empirical analysis is based on a stylized portfolio choice model in which only fixed
investments are assumed to have risks, whereas financial investments are assumed to be risk-free. Given that most of the financial
assets in Argentina, Mexico and Turkey are short term T-bills in Demir (2009), the assumption of riskless financial investments might
be reasonable, which also simplifies the derivation of the analytical model for describing firms’ financialization behavior in the
underlying countries. However, financial assets invested by the Chinese firms include stocks, mutual funds and many other risk-
bearing assets, which is why the assumption of riskless financial investments does not apply to the Chinese case. Moreover, the risk
measurement in Demir’s analysis is approximated by macroeconomic uncertainty, which somehow deviates from the baseline de-
finition of the fixed investment risk defined in the portfolio choice model.
Because the financialization behavior of Chinese firms is under study, this paper focuses on the financial investment decision of
non-financial firms in China and intends to complement the existing studies from the following three aspects. First, we relax the
assumption of riskless financial investments in Demir (2009) portfolio choice model and derive an extended model to capture firms’
financial investment behavior. The extended model is not only analytically but also economically different from the stylized model in
the literature. We show that the relative risk of fixed investment to total risk and risk-adjusted return gap between financial and fixed
investments, instead of pure risk and simple return gap, appear in firms’ financial investment decision framework. Second, we
measure investment risks by the baseline definition with the corresponding variations in the returns on the investments. Third, we
compare the determinants of financial investments of non-financial firms in our extended model with the stylized model and we find
the results and the corresponding conclusions are critically different, which reinforces the importance of relaxing the stylized as-
sumption of risk-free financial investments. These findings are robust to alternative models with lagged and dynamic specifications.
Our empirical application utilizes firm level panel data of listed non-financial firms in China over the period from 2006 to 2016
with semi-annual frequency. The choice of China over this period is representative for large developing countries and closely related
to the distinct features of investment behavior of Chinese non-financial firms over the most recent decade. In particular, the share
(ratio) of financial asset to aggregate asset in Chinese non-financial firms exhibits an overall growing trend from 2006 to 2016 (with a
relatively short-period fall from 2011 to 2013), during which the share of fixed asset to aggregate asset has been declining over time,
as depicted in Fig. 1.
The continuous fall in real investment coupled with rapid rise in financial investment for the non-financial firms seems coun-
terintuitive and motivates researchers to explore the determinants of the firms’ financial investment decision. Like the existing
studies, our analysis highlights the importance of risk and return (profit) in determining the firms’ investment decision in the
theoretical model. As mentioned above however, we relax the assumption that financial investment is riskless and derive an extended
framework to capture firm’s portfolio choice. The empirical results confirm that the theoretical extension is critical in understanding
the firms’ financial investment decision in China.
The organization of the paper is as follows: Section 2 develops a portfolio model for firms acknowledging risks in both fixed and
financial investments; Section 3 describes data and variables used in our empirical analysis; Section 4 provides empirical results for
the baseline model; Section 5 provides robustness tests with alternative financialization and risk measures and different firm sizes,
ownership structures and time periods, followed by Section 6 which concludes the paper.

2. Baseline model

In this section, we develop a portfolio choice model to our analysis of financial investment behavior of non-financial firms in
China. The portfolio choice for the firms is investment allocation between financial and fixed assets. By definition, fixed investment
assets refer to the wealth invested in a long-term tangible piece of property that a firm owns and uses in the production of its income
and is not expected to be consumed or converted into cash any sooner than at least one year's time. Financial investments, on the
other hand, include various financial assets available in the financial markets. The model contains a pool of firms living in a country
where they consume their returns from capital invested in one-period investment projects in both fixed and financial assets. Like
Demir (2009), it is assumed that each type of assets can be considered as the sum of multiple investments. It is also assumed that there
is a single homogeneous good2 and the population is normalized to one with a zero growth rate.
Note that Demir (2009) follows Le and Zak (2006) and assumes firms’ financial investments are riskless. For some developing
countries, this might be true since treasury bills are the main financial assets firms invest in. In China, however, financial assets
available for firms include both risk-free financial products such as T-bills and risk-bearing instruments such as stocks and investment
funds.3 Therefore, unlike Demir (2009), we allow risks in firms’ financial investments. As it turns out, this different but more plausible
assumption could lead to very different model representation from the existing literature.
Let It f be the financial assets investment at time t with a return rt f . Firms can also invest Itk in financial assets with a return rtk . In

2
Note that this is not a key assumption in our model settings. It is included as it is necessary for Demir (2009) model.
3
In our sample, riskless assets such as cash and cash equivalents account for 13% of total assets on average, while other financial assets including
investment real estate, financial assets available for sale and so on make up roughly 6.75% of total assets over the sample period. In addition,
Chinese firms are generally insensitive to exchange rate risks.

2
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Fig. 1. The ratios of financial and fixed assets to aggregate assets in non-financial firms Notes: The raw data was obtained from WIND database; the
firms included are those categorized in non-financial industry and listed in China A-share stock market.

our model, both financial assets investment and fixed assets investment are risky. More specifically, the returns of rt f and rtk follows
normal distributions with means and variances specified as follows:

rt f Ñ(E (rt f ), Var (rt f ))


.
rtk Ñ(E (rtk ), Var (rtk )) (1)
4
It is also assumed that the returns in the two different investments (markets) are uncorrelated at time t. Also, both types of
investments are undertaken from the beginning to time t using the initial capital of I0. As in Demir (2009), the standard maximization
by a representative firm of the expected utility from such investments provides us with the following problem:

Max E tU (It ),
t=0 (2)
subject to

It = (1 + rtk ) Itk + (1 + rt f ) It f , (3)


and

I0 = It f + Itk . (4)
The utility function is a constant absolute risk aversion utility function, viz.
U (It ) = e It ,
(5)
where

U (wt )
, t
U (wt ) (6)
Note that in Demir (2009) setup financial assets investment is riskless assuming the return rate of financial investment is time
invariant and by applying the Stein’s Lemma the optimum allocation becomes a very simple expression, viz.

E (rtk r f )
Itk = ,
Var (rtk ) (7)
where is a function of second and first orders of the underlying utility function.
In addition, the aggregate capital invested in the economy is the sum of financial and fixed capital, viz.

K t = Itk + It f (8)

Rearranging Eq. (8) and substituting Itk from Eq. (7) gives the equilibrium level of financial investment as shown in Demir (2009):

E (rtk rf)
It f = K t ,
Var (rtk ) (9)
or equivalently

It f 1 E (rtk r f )
=1 × .
Kt Kt Var (rtk ) (10)

4
Table 2 in section 3 provides numerical support for this assumption.

3
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

It is must be emphasized that the neat representation of Eq. (10) was obtained by assuming financial assets investment is riskless.
It shows that the financial investment ratio is related to the return gap between fixed and financial investments over the risk in fixed
assets investment as well as the overall capital.
Demir (2009) claims that taking the natural log of both sides and approximating log(1 x ) with log( x ) can get the following
linear expression:

It f
ln = ln(E (rtk r f )) + ln( ) + ln(Var (rtk )) + ln(K t ).
Kt (11)

Note that the approximation method mentioned in Demir (2009) is likely to be an editing error because in general log(1 x ) is
not approximately equal to log( x ) , although under certain exceptional condition, log(1 x ) can be approximated to be x . The
correct approximation, however, should be log(1 x ) log(x ) which actually transforms Eq. (10) to Eq. (11).
Again, Eq. (10) was derived based on the assumption that financial investment is riskless. Now we come back to our assumption
that both fixed and financial investments are risky and we can obtain:

E [rt f rtk] = [Var (rt f ) It f Var (rtk )(Kta It f )], (12)


5
and

It f Var (rtk ) E [rt f rtk ]


= + .
Kta Var (rt f ) + Var (rtk ) Kta [Var (rt f ) + Var (rtk )] (13)
In contrast to Demir’s Eq. (10), our Eq. (13) shows that the share of financial assets in aggregate capital is affected by the ratio of
fixed investment risk to total risk (i.e. relative risk instead of pure risk) and risk-adjusted rate of return gap between financial and
fixed investments.
Based on theoretical model (13), we can write the corresponding econometric model for empirical analysis, viz.
k f k
FIit = cit + 1 risk it + 2 rgap, it + controlit + it , (14)
where FIit denotes the share of firms’ financial investment in aggregate capital, cit is a constant, riskitk refers to the ratio of fixed
investment risk to total risk, and rgap, it denotes the risk-adjusted rate and is calculated as
f k
E [rt f rtk]/ Kta [Var (rt f ) + Var (rtk )]; 1, 2 ,
and denote parameters. In addition, controlit refers to other variables (i.e. control variables) that may also affect firms’ financial
investments. For example, firms’ financial constraints may affect their decisions in financial investments. Similarly, firms’ leverage
ratio, size and growth potential may also provide information about their ability in financial investments. Therefore, in practice, we
consider firms’ financial constraints (fc), leverage ratio (lev), firm size (asset) and revenue growth rate (rev) as control variables in the
regression model (14). A detailed description for variable definitions, measurement, and calculations are provided in the following
section.

3. Data and variables

3.1. Data source

The datasets are obtained from WIND, a database providing both macro and micro levels of data in China. We use balance sheet
data for listed companies in China A-share stock market in this database. The sample period of our empirical analysis is semi-annual
and covers the period from 2006:1 to 2016:2. The use of this sample period is dictated by the change of Chinese accounting standard
in 2006, which makes the finance and accounting data for the listed companies incomparable before and after 2006. Semi-annual
data also provides an advantage in capturing the effects of sudden changes in profitability and risk on the investment positions of the
firms especially with regards to financial investments (Demir, 2009).
The firms included are non-financial firms in 17 industries categorized by China Securities Regulatory Commissions (CSRC). To
maintain relatively sufficient consecutive time series from the dataset, we dropped those companies listed after year 2011. The final
sample for firms includes 1902 firms with 1140 in manufacturing, 124 in wholesale and retail, 117 in real estate, 92 in information
services, 86 in electricity power generation, 76 in transportation, inventory and post, 55 in mining, 52 in construction, 33 in agri-
culture, forestry, animal husbandry and fishery, 31 in culture, sports and entertainment, 31 in leasing and business services, 22 in
comprehensive industry, 18 in water resources, environment and public facilities management, 10 in scientific research and tech-
nology services, 9 in accommodation and catering industry, 5 in health and social work, and 1 in education.

3.2. Measurement and calculation of variables

The finance and accounting items6 involved in calculating the key variables in our empirical analysis include total assets, total

5
We provide a detailed derivation of Eq. (13) in the Appendix A.
6
Without further details of the firms’ financial and fixed investment such as the specific terms of investments, all the accounting title values are
those at the end of each time period to preserve the number of observations.

4
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

liabilities, liquid liabilities, total owners’ equity, net cash inflow from operating, revenue, operating cost, tax and associate charge,
asset impairment loss, period cost, investment real estate, held-to-maturity investments, financial assets available for sale, long-term
equity investment, dividend receivable, interest receivable, financial assets for trading, cash and cash equivalents, net return on
investment, net income from fair value variation, income from investment to joint venture and cooperative enterprise, exchange net
gain, interest income, interest expense, trading financial liability, derivative financial liability and operating profit. Note that period
cost is calculated based on Trailing Twelve Months (TTM) method, which is measured by the cost statements from a company’s
reports to calculate the cost for the twelve-month period immediately prior to the date of the report. The TTM was used because
quarterly and interim reports often show only income from the preceding 3, 6 or 9 months, not a full year.

3.2.1. The share of financial investment


The share of financial investment to aggregate capital reflects the nature of firms’ financial investment behavior. The higher of
this share, the more willing to invest in financial assets are the underlying firms. According to the current Chinese accounting
standards for enterprises, financial assets consist of 7 accounting items: cash and cash equivalents, held-to-maturity investments,
financial assets available for sale, investment real estate, tradable financial assets, dividend receivable and interest receivable.
Therefore, the share of financial investment is defined and calculated as follows:
FI = (cash and cash equivalents + held-to-maturity investments + financial assets available for sale + investment real es-
tate7 + tradable financial assets + dividend receivable + interest receivable)/total assets.

3.2.2. The rate of return gap


As explained in Eq. (13), the rate of return gap in the regression model is calculated based on the rates of return on financial and
fixed investments, the total risk, as well as the total asset. According to the current Chinese accounting standards for enterprises, the
rates of return on financial and fixed investments are defined as below:
r f = (net return on investments + net income from fair value variations + exchange net gain - income from investment to joint
venture and cooperative enterprise + interest income – interest expense)/(cash and cash equivalents + held-to-maturity invest-
ments + financial assets available for trading + investment real estate + financial assets available for sale + dividend re-
ceivable + interest receivable) and
r k = (revenue – operating cost – tax and associate charge – period cost - asset impairment loss)/ (fixed assets + intangible asset
and other long term asset).8
The risk-adjusted rate of return gap is then calculated by (rt f rtk )/ Kta [Var (rt f ) + Var (rtk )] where the measurement of risk vari-
ables is described below. Note that the return gap captures the market signals regarding future profitability in operational activities
and the effects of opportunity costs of financial investment.

3.2.3. Investment risks


The risk variable in the regression model is the ratio of fixed investment risk to the sum of fixed and financial investments risks. To
calculate this variable, we first obtain individual risk variables for financial and fixed investments respectively (i.e. Var (rt f ) and
Var (rtk ) ). They are measured by conditional variances from a GARCH (1, 1) process based on the following equation:
x t = c + ut
2 2 ,
t = 0 + 1 t 1 + ut2 1

where x t is the return on financial (fixed) investment, c refers to a constant in the mean equation of the GARCH model, and t2 is the
conditional variance of ut , and is used to measure financial (fixed) investment risk. The risk variable riskitk in the regression model is
then calculated as Var (rtk )/[Var (rt f ) + Var (rtk )].

3.2.4. Control variables


The control variables are chosen based on literatures (Hovakimian, 2009; Shi & Zhang, 2018) and their relevancy to the un-
derlying model. Overall, we consider four control variables, including firms’ financial constraints, financial leverage, asset size, and
revenue condition. To be specific, the control variables are defined as follows:
Financial constraints ( fc ) = (net cash inflow from operating)/(total assets). A significantly positive coefficient suggests firms’
investment is sensitive to cash flows, and hence is constrained by how much cash firms hold;
Financial leverage ratio (lev ) = (total liabilities)/(total owners’ equity). The higher the level of leverage is, the higher the financial
risk tends to be, which may hinder firms’ capability to invest;
Firms’ asset size (asset ) = natural log of firms’ asset. Larger firms are usually relatively more mature in main business with stable
and possibly strong profitability, therefore the needs for financial assets are lower than smaller firms; and
Firms’ revenue growth (rev ) = the growth rate of firms’ revenue. Higher revenue growth indicates higher growth potential in

7
According to the accounting standards in China, real estate investment is included in financial assets while plant and property that are used for
firms’ operations are categorized as fixed assets.
8
Different accounting rules might lead to different measurement of main variables of interest. For example, land is excluded when calculating
fixed assets in Demir (2009). However, the rapid growth of real estate has made it a significant element when calculating fixed and financial assets.
By the accounting standards in China, plant and property is categorized as fixed asset and real estate investment is financial asset.

5
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 1
descriptive statistics of main variables.
Variables Mean Std. Min Max

FI 0.211 0.158 0.000 1.000


rf 0.027 0.068 −0.022 0.337
rk 0.075 0.131 −0.238 0.445
risk k 0.600 0.385 0.000 1.000
f k
rgap −0.000 0.003 −0.210 0.332
fc 0.030 0.069 −0.124 0.206
lev 1.282 1.220 0.048 5.556
asset 21.564 1.295 19.020 24.699
rev 0.157 0.413 −0.511 1.774

Notes: all variables are winsorized by 2.5% to rule out outliers.

Table 2
Correlation matrix of main variables.
FI rf rk risk k f k
rgap fc lev asset rev

FI 1.000
rf −0.018 1.000
rk −0.002 0.001 1.000
risk k −0.027 0.011 0.007 1.000
f k
rgap −0.006 0.001 0.000 −0.002 1.000
fc 0.006 −0.007 −0.005 −0.015 −0.001 1.000
lev −0.012 −0.001 0.000 0.008 0.000 −0.013 1.000
asset −0.145 −0.023 −0.017 −0.044 0.005 0.078 0.019 1.000
rev −0.006 0.000 0.000 0.008 0.000 0.000 0.001 −0.002 1.000

main business. Firms with higher growth potential may tend to expand production leading to a lower financial investment ratio.
Descriptive statistics and a correlation matrix of main variables are reported in Tables 1 and 2 respectively. The mean value of FI
indicates that financial assets account for 21.1% of total assets on average. The rates of return, however, exerts different features:
higher mean values and standard deviation of r k imply that although financial investment has lower return rates but it is less risky
than fixed investments. It may be unsurprising to see that overall the financial investment is less risky than the fixed investment
because the financial assets that Chinese nonfinancial firms invest in are mostly short term products with relatively low rate of return,
while the fixed assets include plant and property for firms’ operational purposes of which the returns are more volatile (especially
under economic policy uncertainties in China). Statistics of riskitk and rgap, it further confirm the differences between the two types of
f k

investments. Regarding control variables, relatively bigger values of standard deviations indicate a big gap between large competent
firms and smaller ones. Table 2 highlights a small value of correlation coefficient between r f and r k which in turn supports the
assumption of zero correlation between r f and r k in Section 2 of our paper.

4. Empirical results

The empirical specification of the econometric model pertains to the theoretical model (14). Therefore, our empirical model is
specified as
k f k
FIit = cit + 1 risk it + 2 rgap, it + controlit + it , (15)
where FI , risk k ,
and control are defined in Section 3.
f k
rgap
Note that the empirical model (15) has not considered delivery lags and adjustment costs. Therefore, our empirical analysis also
examines the following two alternative specifications to capture possible dynamic mechanism in the underlying model, viz.
k f k
FIit = cit + 1 risk it 1 + 2 rgap, it 1 + controlit 1 + it , (16)
and
k f k
FIit = cit + 1 FIit 1 + 2 FIit 2 + 1 risk it + 2 rgap, it + controlit + it . (17)
By incorporating lag terms of FI into the model, Eq. (17) effectively captures the aggregate effects of both the contemporaneous
and lag terms of explanatory variables. Details on the derivations of the dynamic structures of Eq. (17) are provided in the Appendix
B.
To facilitate comparisons with the relevant literature, we also provide estimation results for the stylized model in the literature
with pure risk and simple return gap, as in Demir (2009). In those regressions, the risk variable refers to fixed investment risk

6
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

(denoted as risk unadj ) and rgap


unadj
denotes simple difference in returns between financial and fixed investments, namely r f r k . To
facilitate notations, we denote the estimation results for our models as baseline model results and the estimation results for the
stylized model in the literature as unadjusted model results.9
To estimate the underlying panel data models and obtain valid results, we need to pin down several econometric issues. For Eqs.
(15) and (16), we first examine whether individual effect exists in the regression model with the Breusch-Pagan (Breusch & Pagan,
1980) LM test. The null hypothesis for the test is that there is no individual effect, under which a pooled OLS estimation method is
valid. If the null hypothesis is rejected, we proceed to conduct the Hausman test in panel regressions (Hausman, 1978) to examine
whether a random effect or fixed effect estimation method should be implemented. Be design, the null hypothesis of the Hausman test
favors the random effect while a rejection of the null favors the fixed effect. Another presumably more important issue is potential
endogeneity in the model. To address this issue, we carry out heteroskedasticity-robust C-test of Hayashi (2000) for endogeneity,
which effectively makes use of the difference between two Sargan-Hansen statistics (Sargan, 1958; Hansen, 1982). The C-test results
entail the use of instrumental variable (IV) estimation method or more general name Generalized Method of Moments (GMM). The
instrumental variables in the endogeneity tests include two to four lag terms of the variables on the right hand side of the regression
equations. The validity of the IV set is also checked by the cluster-heteroskedasticity-robust Hansen-J test (Hansen, 1982) for over
identification. In practice, we also examine the validity of the IV choice by Kleibergen-Paap rk LM test (Kleibergen & Paap, 2006),
which verifies whether the product of the endogenous variables matrix and the IV matrix satisfies full rank requirement. Based on the
test results, assuming firm-level and time (semi-annual) fixed effect specification and utilizing GMM estimation are preferred
treatment where all explanatory variables are endogenous. We only report the Kleibergen-Paap LM test statistics (denoted as KP-LM)
and Hansen J statistics (denoted as Hansen J) to save space.
In addition, for Eq. (17) which is a dynamic panel data model, we follow the standard estimation method for endogenous dynamic
panel data model and use system GMM method to obtain estimation results. In this setup, the diagnostic tests focus on autoregressive
(AR) tests for lag one and two periods and over-identification tests.
The estimation results for Eqs. (15), (16), and (17) for both baseline and unadjusted models are reported in Table 3 and 4. Again,
the baseline model refers to our extended model and the unadjusted model denotes the stylized model in Demir (2009). Looking at
Table 3 for the results of Eqs. (15) and (16), our primary interests are the impact of risk and return gap on firms’ financial investment
decision. The results for our baseline model allowing for risks in both fixed and financial investments show that the relative risk of
fixed investment exerts significantly positive impact (at 1% level) on firms’ financial investment decision in both Eqs. (15) and (16),
indicating that increasing risk in real investment motivates firms to invest more in financial markets. This result is intuitive because
rational firms tend to avoid fixed investment and favor financial investment if the corresponding risk in the fixed investment is
relatively higher.
Perhaps a more interesting finding is that the return gap between financial and fixed investments has no statistically significant
effect on firms’ financial investment in either one of the two model specifications, which means that although both the relatively risk
of fixed investment and the profitability of financial investment are two driving factor of firms’ financialization as indicated by Eq.
(13), statistically it is the relatively higher risk k (not rgap
f k
) that motivates non-financial firms to make financial investment decision.
Nonetheless, the impact of risk and return gap in the Demir-type unadjusted model provides a stark contrast. The risk in pure term
(i.e. the risk of real investment), has no significant driving force on financial investment whereas the return gap between financial
investment and fixed investment without total risk adjustment has a significantly positive impact on financial investment. These
results suggest that the empirical findings regarding the impact of risk and return gap on the firms’ financial investment decision
could be very different between the models considering and neglecting risks in the firms’ financial investment. Put it bluntly, em-
pirical results using the stylized model neglecting financial investment risk may lead to misleading conclusions for the Chinese case.
Turning to the coefficient estimates for the control variables in Table 1, we see that the results for the baseline model and the
unadjusted model are in general consistent. Indeed, the financial constraint variable is significantly positive while the asset size and
revenue variables are significantly negative in both regression models. Also, the leverage ratio is insignificant at conventional levels
of significance in the baseline model and only marginally significant at 10% level in the unadjusted model. The significantly positive
impact of financial constraint variable on financial investment is intuitive because fc measures net operating cash inflow (over total
assets). The more net cash a firm has, the higher degree of freedom the firm has for financial investment, given the underlying risk
and return gap the firm is facing. In addition, the negative impact of asset size and revenue growth on financial investment appears
less intuitive but interesting. The negative impact means that an increase in asset size and revenue growth decreases firm’s financial
investment. This may indicate that firms with larger asset size and revenue growth pay more attention to real investment and hence
development in firms’ real business whereas firms with smaller asset size and revenue growth focus more on short-term investment
opportunity in financial markets.
The diagnostic test statistics reported in the lower panel of Table 3 indicate that the model is properly specified and the IV choice
is valid and strong in both regressions. Overall, the results in Table 3 suggest that Chinese non-financial firms’ financial investment
decision is dominated by the relative risk in real investment but not by the return gap between financial and fixed investment.
Next, we further examine the robustness of our baseline findings in the dynamic model (17). The estimation results summarized in
Table 4 generally provide consistent findings: the coefficient of risk k is significantly positive while rgap
f k
shows no significant effect on
FI which again indicate the dominance of fixed investment risk over the return gap between financial and fixed investment regarding

9
To complement the existing literatures on fixed investment behavior and provide further support for the empirical evidence in our paper, we also
re-estimate Eqs. (15)–(17) using fixed investment ratio as dependent variable and report the results in Appendix C.

7
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 3
Estimation results for Eq. (15) and Eq. (16).
Baseline model UNADJUSTED model

Eq. (15) Eq. (16) Eq. (15) Eq. (16)

riskk 0.055*** 0.061***


(0.010) (0.012)
rf-k gap −16.472 −4.331
(15.218) (14.103)
riskunadj 0.442 0.319
(0.579) (0.411)
runadj gap 2.702*** 2.345**
(1.005) (1.040)
fc 0.544*** 1.387*** 3.969*** 3.500**
(0.100) (0.155) (1.238) (1.496)
lev −0.000 0.021 −0.066* −0.029
(0.005) (0.015) (0.037) (0.034)
asset −0.060*** −0.074*** −0.181*** −0.192***
(0.008) (0.009) (0.051) (0.062)
rev −0.139*** −0.085*** −0.647*** −0.452**
(0.040) (0.031) (0.215) (0.198)
Number of firms 1,902 1,902 1,902 1,902
Number of observations 34,554 29,763 25,731 23,628
Year Dummy yes yes yes yes
KP-LM 0.063 0.069 0.005 0.018
Hansen J 0.346 0.287 0.325 0.281

Notes: The table reports GMM estimation results with cluster-heteroskedasticity-robust standard errors reported in parentheses. We choose the IV set
from 2 to 5 lag terms of dependent variables in the corresponding regressions with accordance to diagnosis statistics. ***, **, and * denote statistical
significance at 1, 5, and 10 percent levels. KP-LM and Hansen-J refer to p-values of the diagnostic tests.

Table 4
Estimation results for Eq. (17).
Baseline model unadjusted model

IV set collapsed IV set untransformed IV set collapsed IV set untransformed

FI(-1) 0.526*** 0.518*** 0.949*** 0.627***


(0.112) (0.047) (0.132) (0.050)
FI(-2) 0.204** 0.195*** −0.063 0.143***
(0.100) (0.044) (0.112) (0.043)
riskk 0.023*** 0.023***
(0.009) (0.007)
rf-k gap −1.563 −1.353
(4.648) (1.594)
riskunadj −0.030 0.001
(0.055) (0.011)
runadj gap 0.217*** 0.008
(0.081) (0.024)
fc 1.123*** 0.090 1.917*** 0.215***
(0.274) (0.060) (0.355) (0.050)
lev −0.007 −0.010*** 0.011 0.003
(0.005) (0.002) (0.007) (0.002)
asset −0.009*** −0.002* 0.019 −0.008**
(0.003) (0.001) (0.012) (0.003)
rev −0.010*** −0.000 −0.141* −0.011
(0.003) (0.002) (0.076) (0.011)
Number of firms 1,902 1,902 1,902 1,902
Number of observations 35,030 35,030 35,030 35,030
Year Dummy yes yes yes yes
AR(1) 0.001 0.000 0.000 0.000
AR(2) 0.821 0.809 0.386 0.292
Hansen-J 0.213 0.000 0.192 0.000

Notes: The “collapse” transformation is applied during system GMM estimation to avoid the “too many instruments” problem (Mehrhoff, 2009);
however, we also report results without applying the transformation for comparison purpose. Note that without the “collapse” transformation, the
Hansen J statistics become relatively smaller due to over identification. The table reports system GMM estimation results with cluster-hetero-
skedasticity-robust standard errors reported in parentheses. We choose the IV set from 2 to 5 lag terms of dependent variables in the corresponding
regressions based on diagnosis statistics. ***, **, and * denote statistical significance at 1, 5, and 10 percent levels. AR(1) and AR(2) refer to p-values
of autocorrelation tests at orders 1 and 2. Hansen-J refers to p-values for over identification tests.

8
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 5
Estimation results for Eq. (15) and Eq. (16) (FI = financial liability ratio).
Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16)

riskk −0.081 −0.446


(0.190) (0.382)
rf-k gap 97.083 182.699
(102.910) (133.760)
riskunadj −0.780 −0.788
(0.730) (0.782)
runadj gap 1.913 −3.836
(2.157) (3.206)
fc −3.644 −1.883 −0.939 −5.674
(2.396) (1.999) (3.022) (4.557)
lev 0.003 −0.070 −0.075 −0.151
(0.079) (0.082) (0.097) (0.121)
asset 0.358** 0.717*** 0.569** 1.055***
(0.140) (0.219) (0.239) (0.341)
rev −0.188 0.737** 0.053 1.351
(0.331) (0.343) (0.699) (0.832)
Number of firms 1,902 1,902 1,902 1,902
Number of observations 25,391 23,568 22,695 20,875
Year Dummy yes yes yes yes
KP-LM 0.005 0.014 0.000 0.005
Hansen J 0.276 0.385 0.480 0.574

Notes: see Table 3.

the driving factors of non-financial firms’ investment. Furthermore, results of the unadjusted model of Demir (2009) exert similar
outcome as those of static models in Table 3. Furthermore, coefficients of the lag terms of FI are significantly positive in both
specifications implying that non-financial firms’ financialization features an inertia effects, driving and deepening itself.
In addition, the statistical nature of coefficient estimates on the control variables are generally consistent with those shown in
Table 3. The diagnostic tests provide supporting evidence for the econometric specification and estimation.
To summarize, we find that the extended portfolio model for non-financial firms relaxing the assumption of riskless financial
investment provide utterly different findings. With the more plausible assumption of risks in both fixed and financial investments, we
confirm that the relative risk of fixed investment significantly drives non-financial firms to invest financial assets while the potential
higher profitability in financial investment is statistically not a significant factor in Chinese firms’ information set for their portfolio
choice. These results refine our understanding on non-financial firms’ financial investment decision, at least for the case in China.

5. Further discussions

In this section, we incorporate alternative variable measures in terms of financializaiton and investment risks and provide ro-
bustness tests with different firm sizes, ownership structures and time periods. We follow the same procedures described in the former
section when determining the appropriate estimation methods for model specifications. Also, we assume firm-level individual effect
and time fixed effect exist and apply GMM estimation to address potential endogeneity.

5.1. Using financial liability ratio as indicator of financialization

As described in Davis (2017), the post-1980 period of USA economy features a dramatically expanding debt in non-financial firms’
balance sheets which could be utilized to acquire financial assets. Thus financialization of non-financial firms is also linked with the
liability side of balance sheets. To ascertain whether the logic and mechanism of liability side analysis resembles that of the asset side
which we adopt in the former sections, we incorporate financial liability ratio which is calculated as: (trading financial liabi-
lity + derivative financial liability)/total liability into our empirical models as an alternative indicator of financialization. Likewise,
we also report the results from Demir’s model specifications. The estimates are reported in Tabless 5 and 6.
Surprisingly, neither the investment risk nor the return gap have significant effect on financial liability ratio, whether we apply
our model or Demir-type specifications, which indicates the driving mechanisms of financial liability ratio does not share similar
channels as financial investment ratio (asset side of balance sheet). Some intriguing implications, however, come from the change of
signs of asset and rev in Table 5, both of which are positive indicating larger firms and those with higher growth potential tend to have
higher ratio of financial liability over total liability. This is in fact intuitive as highly competent firms have easier access to financial
markets and thus are more capable of funding with financial tools.

9
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 6
Estimation results for Eq. (17) (FI = financial liability ratio).
Baseline model Unadjusted model

IV set collapsed IV set untransformed IV set collapsed IV set untransformed

FI(-1) 0.785*** 0.625*** 1.539** 0.564***


(0.253) (0.095) (0.606) (0.090)
FI(-2) 0.006 0.166** −0.534 0.281***
(0.288) (0.082) (0.494) (0.087)
riskk 0.348 −0.178
(0.419) (0.111)
rf-k gap −106.699 14.322
(214.204) (22.019)
riskunadj −0.568 −0.080
(0.465) (0.145)
runadj gap −0.885 −0.345
(1.081) (0.236)
fc −40.626 −0.405 −12.869 0.005
(24.700) (0.562) (18.609) (0.638)
lev 0.148 0.020 0.075 0.082**
(0.123) (0.019) (0.132) (0.032)
asset −0.819 0.026 −0.133 −0.035***
(0.588) (0.029) (0.128) (0.004)
rev 0.298** 0.014 −0.179 0.055
(0.125) (0.015) (0.404) (0.051)
Number of firms 1902 1902 1902 1902
Number of observations 28,442 28,442 28,442 28,442
Year Dummy yes yes yes yes
AR(1) 0.032 0.000 0.054 0.000
AR(2) 0.371 0.967 0.165 0.271
Hansen-J 0.152 0.467 0.242 0.029

Notes: see Table 4.

5.2. Using alternative risk measures

Tables 3 and 4 have shown that the risk factor is a critical driving force of non-financial firms’ financial investment. Therefore it is
crucial to test the sensitivity of the mechanism to different measures of risk related variables.

5.2.1. Market beta and economic policy uncertainty


In Section 3, we utilize a GARCH (1, 1) model to obtain the conditional variances of both return rates as the measures of the
respective investment risks. While the GARCH method is close to the direct definition of uncertainty and conveniently produces time-
variant measures, using only volatility risks of return rates might neglect other aspects of risks non-financial firms are exposed to.
Macro level risk such as economic policy risk could distinctively influence firms’ investments. To analyze the effects macro level risks
have on non-financial firms’ financial investment, we introduce market beta and economic policy uncertainty (EPU) (Baker, Bloom, &
Davis, 2016) into our empirical models. Specifically, we obtain the data of market beta of each individual firm from WIND as an
aggregate measure of investment risks, and use the Chinese EPU10 as fixed investment risk as is similar in Demir (2009). Therefore,
the alternative risk k and rgap
f k
are calculated as: EPU/market beta and (rt f rtk )/(Kta × market beta) . The estimation results are reported
in Tables 7 and 8.
Results of baseline model (15) and (16) are generally consistent with those in Table 3 where the ratio of fixed investment risk is a
significant driving factor of firms’ financial investment. Results of unadjusted model now resemble the hypotheses in Demir (2009)
indicating that both the return gap between financial and fixed investment and economic policy uncertainty drive firms’ financial
investment. Turning to Table 8, results of baseline model, however, show that the risk variable is no longer significant. A more
surprising result is that the coefficient of simple return gap in Demir’s model specification turns negative, which is somehow counter-
intuitive. Nevertheless, given that financial investments are riskless in Demir’s model assumption, the negative sign of return gap
implies that Demir’s theoretical model settings might not be plausible for the Chinese case. In addition, estimation results of control
variables remain robust.

5.2.2. Standard deviations of logarithmic differences of return rates


Apart from market beta and EPU, we also alter the calculation of risk measures, by taking the logarithmic differences of both
return rates and obtain the standard deviations of both logarithmic differences as investment risks. Note that the calculation itself

10
The index calculated by Baker et al. (2016) is normalized to a mean value of 100. To obtain proper values during calculating the alternative
measure of risk k , we de-normalize the EPU via dividing the data by 100.

10
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 7
Estimation results for Eq. (15) and (16) (with market beta and EPU).
Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16)

riskk 0.006** 0.007*


(0.003) (0.004)
rf-k gap 0.277 −0.110
(0.206) (0.162)
riskunadj 0.084* 0.228**
(0.046) (0.095)
runadj gap 2.050** 2.225**
(0.924) (1.087)
fc 1.119*** 0.302*** −0.317 2.636*
(0.112) (0.117) (0.878) (1.432)
lev −0.015*** −0.008** −0.018 −0.017
(0.004) (0.004) (0.062) (0.059)
asset −0.007 −0.016*** −0.422*** −0.421***
(0.005) (0.005) (0.135) (0.150)
rev 0.003 −0.056*** −0.781** −0.643**
(0.006) (0.016) (0.315) (0.274)
Number of firms 1,902 1,902 1,902 1,902
Number of observations 26,630 22,999 25,852 23,829
Year Dummy yes yes yes yes
KP-LM 0.000 0.000 0.018 0.027
Hansen J 0.224 0.242 0.496 0.356

Notes: see Table 3.

Table 8
Estimation results for Eq. (17) (with market beta and EPU).
Baseline model Unadjusted model

IV set collapsed IV set untransformed IV set collapsed IV set untransformed

FI(-1) 1.150*** 0.586*** 0.630*** 0.525***


(0.438) (0.064) (0.109) (0.055)
FI(-2) −0.176 0.316*** 0.187** 0.239***
(0.364) (0.055) (0.079) (0.047)
riskk −0.010 −0.000
(0.027) (0.002)
rf-k gap 1.748** 0.013
(0.700) (0.062)
riskunadj 0.012 0.095***
(0.043) (0.016)
runadj gap −0.350*** −0.074***
(0.051) (0.013)
fc 2.876*** 0.235*** 1.241*** 0.150***
(0.996) (0.049) (0.261) (0.057)
lev 0.013 −0.000 0.038*** −0.008***
(0.010) (0.001) (0.004) (0.002)
asset −0.016** −0.008*** −0.003 −0.013***
(0.008) (0.002) (0.007) (0.003)
rev −0.019** −0.003 −0.159*** −0.039***
(0.009) (0.002) (0.019) (0.009)
Number of firms 1,902 1,902 1,902 1,902
Number of observations 28,265 28,265 35,030 35,030
Year Dummy yes yes yes yes
AR(1) 0.015 0.000 0.000 0.000
AR(2) 0.087 0.142 0.203 0.215
Hansen-J 0.413 0.000 0.205 0.000

Notes: see Table 4.

produces time-invariant variables, the coefficients of which a fixed effect model cannot be effectively estimated. Therefore, we adopt
a random effect model specification to obtain estimates of the risk variables. The results are reported in Tables 9 and 10.
Before looking at the estimation results, some critical details are worth mentioning. Taking the logarithmic differences of return
rates actually generates the growth rates of return rates for both assets. Although Table 1 shows that the return rates of financial
investment are averagely lower than those of fixed investment, the growth rates of return rates are, nonetheless, at more complicated

11
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 9
Estimation results for Eq. (15) and (16) (with Standard deviations of logarithmic differences).
Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16)

riskk −0.032*** −0.040***


(0.011) (0.010)
rf-k gap −4.030 −5.352
(3.554) (4.641)
riskunadj 0.001** −0.000
(0.000) (0.000)
runadj gap −0.077 −0.077
(0.124) (0.570)
fc 0.476*** −0.201** 0.290*** −0.109
(0.097) (0.086) (0.071) (0.361)
lev −0.036*** −0.031** −0.069*** −0.003
(0.010) (0.013) (0.022) (0.062)
asset −0.036*** −0.027*** −0.010 −0.022
(0.011) (0.010) (0.012) (0.059)
rev −0.140** −0.092** −0.047 −0.007
(0.063) (0.044) (0.067) (0.350)
Number of firms 1902 1902 1902 1902
Number of observations 24,104 22,254 22,708 22,752
Year Dummy yes yes yes yes
KP-LM 0.000 0.000 0.000 0.000
Hansen J 0.372 0.307 0.553 0.575

Notes: see Table 3.

Table 10
Estimation results for Eq. (17) (with Standard deviations of logarithmic differences).
Baseline model Unadjusted model

IV set collapsed IV set untransformed IV set collapsed IV set untransformed

FI(-1) 0.550** 0.497*** 1.354*** 0.501***


(0.215) (0.055) (0.398) (0.059)
FI(-2) 0.144 0.237*** −0.412 0.265***
(0.179) (0.046) (0.375) (0.053)
riskk −0.727 −0.088***
(0.515) (0.028)
rf-k gap −8.881 −1.370
(8.777) (0.979)
riskunadj −0.003 0.000
(0.006) (0.000)
runadj gap 0.332* −0.038
(0.180) (0.029)
fc 1.540*** 0.115* 2.848*** 0.161***
(0.488) (0.062) (1.063) (0.062)
lev 0.000 −0.009*** 0.036*** −0.012***
(0.009) (0.002) (0.010) (0.002)
asset −0.043** −0.015*** −0.064 −0.013**
(0.020) (0.003) (0.070) (0.005)
rev 0.030 −0.021** −0.187*** −0.014
(0.072) (0.009) (0.058) (0.009)
Number of firms 1902 1902 1902 1902
Number of observations 34,247 34,247 35,030 35,030
Year Dummy yes yes yes yes
AR(1) 0.009 0.000 0.014 0.000
AR(2) 0.710 0.301 0.330 0.141
Hansen-J 0.178 0.000 0.180 0.000

Notes: see Table 4.

situations. Appendix D provides detailed statistics of average values of r f and r k over sample period and corresponding histograms of
both variables.
Along with the rapid development of financial markets in China, the growth rates of return rates of financial investment have been
volatile leading to a much higher standard deviation than that of growth rates of fixed investment return rates. Furthermore, fixed
investment of the Chinese real sector has been declining over the sample period with an increasingly stable growth rates of profits,

12
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 11
Estimation results for Eq. (15) and (16) (with EPU as an individual variable).
Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16)

riskk 0.055*** 0.061***


(0.010) (0.012)
rf-k gap −16.472 −4.331
(15.218) (14.103)
riskunadj −0.015 −0.016
(0.019) (0.022)
runadj gap −0.145*** −0.242***
(0.022) (0.035)
EPU 0.004 0.074*** 0.018*** 0.038***
(0.007) (0.006) (0.005) (0.005)
fc 0.544*** 1.387*** 0.269*** 0.713***
(0.100) (0.155) (0.083) (0.116)
lev −0.000 0.021 −0.005* 0.002
(0.005) (0.015) (0.003) (0.003)
asset −0.060*** −0.074*** −0.032*** −0.034***
(0.008) (0.009) (0.004) (0.004)
rev −0.139*** −0.085*** −0.007** −0.012***
(0.040) (0.031) (0.003) (0.003)
Number of firms 1902 1902 1902 1902
Number of observations 34,554 29,763 34,169 32,267
Year Dummy yes yes yes yes
KP-LM 0.063 0.069 0.000 0.000
Hansen J 0.346 0.287 0.256 0.391

Notes: see Table 3.

thus the standard deviation of the growth rates of r k tends to be smaller, while a relative growing value of standard deviation of the
growth rates of r f indicates that r f is increasing at a faster pace. Ceteris paribus, an increasing standard deviation of the growth rates
of r f will lead to a declining trend of the alternative measure of risk k in our empirical models. Given this situation, the results of
Table 9 may not be surprising since a rapid growing return rates of financial investments provide stronger incentives for firms to
invest more in financial assets, which explains the negative coefficients of risk k in the baseline model results. Regarding Demir-type
models and controls variables, results are generally robust.

5.2.3. Adding EPU as an individual variable


As explained in Section 5.2.1, using a GARCH model might neglect other forms of risks non-financial firms are exposed to. Because
macro level risk influences firms’ investment, it is possible that the return rates of investment partially contain information of macro
risks. To separate the possible effects of macro level risks mixed in return rates, we introduce the Chinese EPU into empirical
framework as an independent variable. Note that in Demir (2009), similar methods are used to compensate the riskless financial asset
assumption, in this section however, we still report the results from unadjusted model for comparison. Provided the nature of EPU, it
is treated as an exogenous variable in our model. Results are reported in Tables 11 and 12.
Estimation result are generally consistent with those in the former sections. After separating potential influences EPU has on
return rates from the original risk related variables, the ratio of fixed investment risk still significantly drives financial investment of
non-financial firms, in both the baseline and the unadjusted models. EPU itself is also a significant factor except in Eq. (15) in baseline
model implying the effects of economic policy shock are lagged in time. Estimation results of the other variables remain robustly
consistent.

5.3. Robustness test

Our sample data contains 1902 non-financial firms over the period of 2006 to 2016 which are inevitably various in asset size and
ownership structures. Thus the financial investment mechanism might vary across firms with different financial characteristics. It is
also necessary to distinguish sample time period between pre-2009 and post-2009 as financial crisis may have considerable effects on
the Chinese real sector. These robustness tests can complement the findings in the former sections and shed light on further research.

5.3.1. Firm size


Former results show that asset significantly affecting firms’ financial investment, hence it is likely that firms with different asset
size might also exert different outcomes with respect to financial investment mechanism. To separate larger firms from smaller ones,
we firstly calculate the average value total asset of each firm; by using the median of average values as a separatrix, we then assume
firms with average total asset higher than the median value as large firms while the others are taken as small firms. Results are
reported in Tables 13 and 14.

13
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 12
Estimation results for Eq. (17) (with EPU as an individual variable).
Baseline model Unadjusted model

IV set collapsed IV set untransformed IV set collapsed IV set untransformed

FI(-1) 0.526*** 0.518*** 0.708*** 0.508***


(0.113) (0.047) (0.198) (0.042)
FI(-2) 0.205** 0.195*** 0.058 0.200***
(0.101) (0.044) (0.160) (0.036)
riskk 0.024*** 0.023***
(0.009) (0.007)
rf-k gap −0.988 −1.353
(4.122) (1.594)
riskunadj −0.024 −0.011
(0.018) (0.010)
runadj gap 0.074 −0.079***
(0.097) (0.020)
EPU 0.059*** 0.031*** 0.077*** 0.048***
(0.014) (0.007) (0.013) (0.007)
fc 1.122*** 0.090 1.595*** 0.130**
(0.280) (0.060) (0.482) (0.055)
lev −0.007 −0.010*** −0.002 −0.007***
(0.005) (0.002) (0.006) (0.002)
asset −0.009*** −0.002* −0.013*** −0.004***
(0.003) (0.001) (0.003) (0.001)
rev −0.010*** −0.000 −0.009*** −0.005**
(0.003) (0.002) (0.003) (0.002)
Number of firms 1902 1902 1902 1902
Number of observations 35,030 35,030 35,030 35,030
Year Dummy yes yes yes yes
AR(1) 0.001 0.000 0.008 0.000
AR(2) 0.822 0.809 0.348 0.485
Hansen-J 0.205 0.000 0.331 0.000

Notes: see Table 4.

Table 13
Estimation results for Eq. (15) and (16) (with different firm size).
Large firms Small firms

Baseline model Unadjusted model Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16) Eq. (15) Eq. (16) Eq. (15) Eq. (16)

k *** ***
risk −0.007 −0.009 0.103 0.090
(0.008) (0.009) (0.018) (0.018)
rf-k gap 26.360 38.241 −10.592 −11.132
(29.425) (28.995) (9.106) (11.707)
riskunadj 0.139 0.083 −0.009 0.017
(0.166) (0.124) (0.179) (0.162)
runadj gap 0.814* 0.492 1.848** 1.296**
(0.427) (0.311) (0.751) (0.652)
fc 0.786*** 1.162*** 2.235*** 1.194*** 1.108*** 0.079 2.073* 0.996
(0.121) (0.151) (0.574) (0.453) (0.264) (0.204) (1.177) (1.169)
lev −0.007 −0.005 −0.038*** −0.015 −0.013* 0.000 −0.096** −0.054
(0.005) (0.005) (0.014) (0.010) (0.008) (0.007) (0.039) (0.034)
asset −0.032*** −0.038*** 0.027 0.003 −0.062*** −0.065*** −0.191*** −0.172***
(0.009) (0.009) (0.018) (0.013) (0.016) (0.013) (0.064) (0.061)
rev −0.143*** −0.142*** −0.164** −0.110** −0.024 −0.051 −0.635*** −0.352**
(0.043) (0.039) (0.069) (0.045) (0.061) (0.040) (0.242) (0.157)
Number of firms 951 951 951 951 951 951 951 951
Number of observations 17,057 16,105 13,112 12,160 15,656 14,705 11,277 10,327
Year Dummy yes yes yes yes yes yes yes yes
KP-LM 0.025 0.054 0.004 0.007 0.025 0.096 0.044 0.068
Hansen J 0.515 0.357 0.481 0.510 0.372 0.341 0.560 0.470

Notes: see Table 3.

14
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 14
Estimation results for Eq. (17) (with different firm size).
Large firms Small firms

Baseline model Unadjusted model Baseline model Unadjusted model

IV set IV set IV set IV set IV set IV set IV set IV set


collapsed untransformed collapsed untransformed collapsed untransformed collapsed untransformed

FI(-1) 0.399*** 0.606*** 0.438*** 0.663*** 1.098*** 0.561*** 1.141*** 0.588***


(0.145) (0.061) (0.153) (0.063) (0.380) (0.054) (0.345) (0.043)
FI(-2) 0.385*** 0.232*** 0.384*** 0.122** −0.313 0.191*** −0.238 0.166***
(0.111) (0.056) (0.112) (0.060) (0.316) (0.049) (0.257) (0.037)
riskk 0.017 0.009* 0.008 0.085***
(0.014) (0.006) (0.028) (0.013)
rf-k gap 14.692 −0.138 16.961 1.982
(40.643) (5.609) (22.659) (2.460)
riskunadj −0.010 0.020 −0.014 −0.004
(0.157) (0.014) (0.038) (0.007)
runadj gap 0.025 −0.016 0.305* 0.005
(0.118) (0.014) (0.177) (0.028)
fc 1.335*** 0.066 1.594*** 0.104* 2.873*** 0.337*** 2.397*** 0.182**
(0.405) (0.055) (0.404) (0.056) (1.065) (0.026) (0.821) (0.088)
lev 0.023 −0.002 0.008 −0.005** 0.012 −0.009* −0.026*** −0.008***
(0.019) (0.003) (0.012) (0.002) (0.014) (0.005) (0.008) (0.003)
asset −0.025*** −0.005*** 0.046 −0.012* −0.187*** −0.001 0.067 −0.015**
(0.010) (0.002) (0.038) (0.006) (0.033) (0.008) (0.057) (0.006)
rev −0.088 −0.037*** −0.153** 0.010 −0.028*** −0.032* −0.014* 0.007
(0.077) (0.011) (0.066) (0.009) (0.010) (0.019) (0.008) (0.011)
Number of firms 951 951 951 951 951 951 951 951
Number of 18,134 18,134 18,134 18,134 16,896 16,896 16,896 16,896
observations
Year Dummy yes yes yes yes yes yes yes yes
AR(1) 0.005 0.000 0.001 0.000 0.029 0.000 0.015 0.000
AR(2) 0.239 0.652 0.136 0.128 0.117 0.390 0.116 0.911
Hansen-J 0.275 0.000 0.448 0.000 0.263 0.000 0.243 0.000

Notes: see Table 4.

Tables 13 and 14 combined reveal very different result between large and small firms. For large firms, neither the ratio of
investment risk nor the return gap is a significant factor driving financial investment, while for small firms, results are consistent with
those of the whole sample indicating that risk k significantly stimulates non-financial firms’ financial investment. It is noteworthy that
the magnitude of coefficient of rgap
f k
is larger in the large firm sample than that in the small firm sample, suggesting that large firms’
financial investment is more sensitive than that of small firms to the return gap. These statistical evidences further confirm that the
reason non-financial firms in China favor financial assets is the relatively higher fixed investment risk rather than return gap between
both investments. Results of Demir’s model and control variables remain generally consistent.

5.3.2. Ownership structures


Among 1902 firms in our sample, 850 of them are controlled by the Chinese government, namely state-owned enterprises (SOEs),
which usually have easier access to bank loans and tend to be less constraint financially due potential political ties. Non-state-owned
enterprises (non-SOEs), however, face severe finance environment and thus might rely on additional financial channel compared to
SOEs. Given the situation, the financial investment mechanisms of SOEs and non-SOEs might be different. Therefore, we separate our
sample into a group of SOEs and that of non-SOEs, and re-estimation results are reported in Tables 15 and 16.
Results of Tables 15 And 16 are similar to those with different firms since many SOEs are also indeed large firms. Looking at
Table 15, for SOEs, there is no strong evidence that risk factor drives the financial investment, while rgap
f k
has significantly (at 10%
level) positive effect on firms’ financializaiton in Eq. (16); non-SOEs’ financialization is significantly motivated by higher ratio of
fixed investment risk implying the harsh financial environment non-SOEs are faced with. Turning to Table 16, rgap f k
significantly
drives SOEs’ financialization while the relative risk has no significant influence. Non-SOEs’ financialization, however, is significantly
motivated by risk k , but only in the untransformed IV set model. Other results are robust and consistent with former findings.

5.3.3. Time period


Since our sample period covers the 2009 financial crisis which may have significant impact on Chinese real sectors. In November
2008, the Chinese policy makers implemented an economic stimulus plan that was later called the “four trillion” plan to counteract
the negative impact from the global financial crisis. The stimulus package helped the Chinese economy to stabilize and maintain a
high GDP growth rate in the following year which might lead to possible changes in the investment environment for nonfinancial
firms after 2009. To analyze the possible shift of financial investment characteristics pre and post crisis, we separate the sample into
two groups: pre 2009 and post 2009 (starts from 2010). Respective results are reported in Tables 17 and 18.

15
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 15
Estimation results for Eq. (15) and (16) (with different ownership structures).
SOEs non-SOEs

Baseline model Unadjusted model Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16) Eq. (15) Eq. (16) Eq. (15) Eq. (16)

k *** ***
risk 0.000 −0.001 0.074 0.069
(0.009) (0.010) (0.015) (0.020)
rf-k gap 5.544 10.885* −13.099 −21.732
(5.206) (6.414) (10.971) (17.404)
riskunadj 0.008 0.032 0.247 −0.039
(0.066) (0.085) (0.327) (0.333)
runadj gap 0.357** 0.327 3.935** 3.168*
(0.151) (0.247) (1.880) (1.768)
fc 0.674*** 1.233*** 1.477*** 2.712*** 0.929*** −0.127 6.485** 3.618
(0.119) (0.163) (0.347) (0.798) (0.202) (0.343) (3.247) (2.402)
lev −0.012*** −0.011** −0.022*** −0.026* −0.001 0.021 −0.160* −0.040
(0.004) (0.004) (0.008) (0.013) (0.008) (0.020) (0.094) (0.065)
asset −0.037*** −0.032** −0.002 −0.003 −0.042*** −0.115*** 0.159 −0.218**
(0.014) (0.014) (0.010) (0.015) (0.011) (0.023) (0.098) (0.085)
rev −0.082* −0.047 −0.085*** −0.089** −0.083* −0.176** −0.464** −0.583**
(0.046) (0.039) (0.030) (0.043) (0.043) (0.075) (0.213) (0.292)
Number of firms 850 850 850 850 1052 1052 1052 1052
Number of observations 15,618 14,768 11,763 10,913 17,096 15,992 13,968 12,916
Year Dummy yes yes yes yes yes yes yes yes
KP-LM 0.019 0.035 0.000 0.008 0.057 0.027 0.067 0.097
Hansen J 0.255 0.263 0.298 0.403 0.413 0.262 0.356 0.426

Notes: see Table 3.

Table 16
Estimation results for Eq. (17) (with different ownership structures).
SOEs non-SOEs

BASELINE model Unadjusted model Baseline model Unadjusted model

IV set IV set IV set IV set IV set IV set IV set IV set


collapsed untransformed collapsed untransformed collapsed untransformed collapsed untransformed

FI(-1) 0.571** 0.575*** 0.479** 0.548*** 0.669*** 0.676*** 0.463*** 0.631***


(0.242) (0.050) (0.243) (0.060) (0.217) (0.046) (0.152) (0.041)
FI(-2) 0.282 0.290*** 0.337* 0.299*** 0.058 0.077** 0.231* 0.117***
(0.201) (0.046) (0.178) (0.053) (0.182) (0.039) (0.126) (0.039)
riskk −0.004 0.003 0.005 0.008**
(0.009) (0.005) (0.010) (0.004)
rf-k gap 13.177** 0.309 9.645 1.251
(5.206) (0.393) (23.775) (1.433)
riskunadj −0.105 −0.029 −0.007 0.008
(0.067) (0.018) (0.039) (0.010)
runadj gap −0.110 −0.022 0.022 −0.043
(0.087) (0.015) (0.064) (0.031)
fc 1.830*** 0.323*** 1.633** 0.179*** 1.863*** 0.533*** 1.238*** 0.299***
(0.659) (0.049) (0.643) (0.043) (0.540) (0.054) (0.155) (0.025)
lev 0.017 −0.002 0.025*** 0.001 −0.009 0.001 −0.024*** −0.012***
(0.015) (0.001) (0.007) (0.001) (0.008) (0.003) (0.006) (0.003)
asset −0.027*** −0.007*** −0.034** −0.015*** −0.068*** −0.006*** 0.008 −0.002
(0.009) (0.003) (0.015) (0.003) (0.011) (0.002) (0.009) (0.005)
rev 0.031 −0.003 −0.104** −0.004 −0.013 −0.021** 0.051 −0.014
(0.088) (0.010) (0.051) (0.009) (0.009) (0.010) (0.046) (0.009)
Number of firms 850 850 850 850 1052 1052 1052 1052
Number of 16,547 16,547 16,547 16,547 18,483 18,483 18,483 18,483
observations
Year Dummy yes yes yes yes yes yes yes yes
AR(1) 0.015 0.000 0.027 0.000 0.023 0.000 0.003 0.000
AR(2) 0.512 0.407 0.843 0.412 0.482 0.177 0.225 0.542
Hansen-J 0.248 0.000 0.282 0.000 0.225 0.000 0.209 0.000

Notes: see Table 4.

16
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 17
Estimation results for Eq. (15) and (16) (with different time period).
Pre 2009 Post 2009

Baseline model Unadjusted model Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16) Eq. (15) Eq. (16) Eq. (15) Eq. (16)

k *** ***
risk 0.026 −0.105 0.068 0.034
(0.054) (0.097) (0.012) (0.011)
rf-k gap 1.396 0.627 −3.861 −13.644
(1.515) (1.476) (7.140) (18.048)
riskunadj 0.054 0.071 0.032 0.115
(0.188) (0.127) (0.202) (0.302)
runadj gap 0.220** 0.300** 2.747** 3.258**
(0.095) (0.138) (1.081) (1.636)
fc 0.047 0.328 0.098 0.035 1.138*** 0.098 4.974** 7.890**
(0.081) (0.231) (0.091) (0.249) (0.124) (0.117) (1.947) (3.689)
lev 0.026 0.059* −0.007 0.025 −0.010** 0.003 −0.064* −0.071
(0.023) (0.035) (0.025) (0.052) (0.005) (0.005) (0.036) (0.054)
asset −0.123*** −0.216*** −0.161*** −0.239* −0.042*** −0.049*** −0.183*** −0.221***
(0.034) (0.072) (0.046) (0.138) (0.010) (0.009) (0.055) (0.085)
rev −0.005 0.001 −0.018 0.021 −0.038 −0.061*** −0.654*** −0.606**
(0.012) (0.021) (0.015) (0.028) (0.028) (0.021) (0.239) (0.293)
Number of firms 1669 1669 1669 1669 1902 1902 1902 1902
Number of observations 6949 5372 6949 5372 25,743 24,693 25,730 23,828
Year Dummy yes yes yes yes yes yes yes yes
KP-LM 0.006 0.003 0.001 0.038 0.033 0.015 0.011 0.058
Hansen J 0.402 0.449 0.265 0.272 0.429 0.238 0.398 0.370

Notes: see Table 3.

Table 18
Estimation results for Eq. (17) (with different time period).
Pre 2009 Post 2009

Baseline model Unadjusted model Baseline model Unadjusted model

IV set IV set IV set IV set IV set IV set IV set IV set


collapsed untransformed collapsed untransformed collapsed untransformed collapsed untransformed

FI(-1) 0.742 0.611*** 0.886*** 0.589*** 0.667*** 0.560*** 0.664*** 0.541***


(0.616) (0.132) (0.314) (0.080) (0.212) (0.046) (0.210) (0.060)
FI(-2) 0.133 0.224** 0.063 0.209*** 0.072 0.204*** 0.092 0.212***
(0.533) (0.105) (0.225) (0.065) (0.170) (0.040) (0.171) (0.053)
riskk 0.009 −0.002 −0.035 0.027***
(0.021) (0.010) (0.040) (0.010)
rf-k gap −52.659 −16.314 6.735 −3.697
(37.199) (10.270) (31.863) (3.165)
riskunadj −0.018 0.004 −0.045 0.013
(0.077) (0.017) (0.042) (0.012)
runadj gap 0.135 0.052* −0.046 −0.033
(0.114) (0.031) (0.130) (0.028)
fc 0.621* 0.529*** 0.944*** 0.056 1.960*** 0.140** 1.619*** 0.127**
(0.340) (0.142) (0.267) (0.055) (0.585) (0.064) (0.551) (0.061)
lev 0.022 −0.002 0.005 0.001 −0.012 0.004 −0.004 −0.008***
(0.016) (0.006) (0.013) (0.004) (0.022) (0.003) (0.006) (0.003)
asset −0.046** −0.003 −0.004 −0.010* −0.032* −0.007*** 0.000 −0.001
(0.020) (0.002) (0.022) (0.006) (0.020) (0.001) (0.019) (0.004)
rev −0.014 −0.000 −0.074 −0.004 0.148 −0.053*** 0.044 −0.029***
(0.015) (0.006) (0.123) (0.014) (0.208) (0.012) (0.043) (0.010)
Number of firms 1669 1669 1669 1669 1902 1902 1902 1902
Number of 8716 8716 8716 8716 26,314 26,314 26,314 26,314
observations
Year Dummy yes yes yes yes yes yes yes yes
AR(1) 0.092 0.000 0.032 0.000 0.003 0.000 0.017 0.000
AR(2) 0.919 0.603 0.595 0.987 0.889 0.384 0.428 0.533
Hansen-J 0.254 0.000 0.200 0.000 0.202 0.000 0.414 0.000

Notes: see Table 4.

17
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Results for pre 2009 period show little significance, while the results for post 2009 period are highly consistent with former
findings.11 An interesting finding is the significant effect of asset for pre 2009 period, which indicates that before the financial crisis,
smaller firms prefer financial investment in comparison to large firms.

6. Conclusions

A small but growing number of studies have provided evidence showing that non-financial firms prefer to invest in reversible
short term financial investments rather than investing in irreversible fixed investment projects. Such evidence is also observed in the
largest developing country China over the recent decade. In essence, the firms in real sector face a portfolio choice between financial
and fixed assets. Empirical analyses of the portfolio choice highlight the importance of risk and return in the financial and fixed
investment. However, the stylized framework of firms’ portfolio choice model assumes fixed investment is risky but financial in-
vestment is riskless. This strong assumption, to a large extent, is imposed to facilitate the derivation of analytical expression for the
regression model in empirical analysis and also based on the financial assets invested in specific countries. The empirical results based
on this framework found that non-financial firms in developing countries take into account alternative investment opportunities in
financial markets when making their decisions on physical investment. The firms’ decision depends on pure risk in fixed investment
and simple return gap between financial and fixed investment. Based on these findings, the existing studies provided policy sug-
gestions that the underlying developing countries need to develop a strategy to link short-term distortions with the medium and long-
term domestic development objectives with special attention to the determinants of productive investment. In particular they sug-
gested that domestic savings should be directed towards productive investments instead of financial ones.
However, the stylized model is negligent to risks in financial investment. For the case of China, the riskless financial investment
assumption might not be practical since financial assets firms invest in also include risk-bearing instruments such as stocks and
mutual funds. Once risks in financial investment is taken into account in the theoretical framework of firms’ portfolio choice model,
the stylized model for firms’ financial investment decision is extended to a new formulation highlighting relative risk instead of pure
risk of fixed investment and risk-adjusted return gap instead of simple return gap between financial and fixed investments. The
theoretical extension also leads to significant change in empirical studies. Our analysis with semi-annual panel data from non-
financial firms listed in China A-share stock market over 2006–2016 period shows that the relative risk term plays a key (statistically
significant) role in driving the firms’ financial investment decision. In contrast, the rates of returns on the respective investments,
after adjusted by total risk, plays no role in affecting the firms’ financial investment decision.
Further examinations with alternative measures of financialization and investment risks and different firm sizes, ownership
structures and time periods provide robustness findings. These findings suggest that the driving mechanism of financial liability ratio
is different from financial investment ratio; alternative risk measures including market beta, EPU and standard deviations of loga-
rithmic differences of return rates are still crucial factors concerning financial investment of non-financial firms; large firms and SOEs’
financial investment are not affected by the relative risk of fixed investment, while empirical results of small and non-SOEs are
consistent with those of the whole sample where fixed investment risk rather than return gap significantly drives financialization; the
financial investment of firms in the post 2009 period are also motivated by relative risk, whilst results of pre 2009 period feature
significant effects from firm size (asset ) on financial investment.
These findings suggest a different policy strategy for Chinese government from the policy strategy recommended by the literature
to other developing countries. In particular, the policy makers in China may need to give more attention to the risk and uncertainty
prevailing in real investment relative to the overall risk instead of spending too much energy in intervening interest rates in the
market. This suggestion, of course, is by no means to debase the importance of returns or financial prices (e.g. interest rates) in
affecting firms’ development, but to propose a macro policy strategy in the specific direction for balancing financial and real in-
vestments of the non-financial firms.

Funding

This research was supported by the National Natural Science Foundation of China (No. 71873131).

CRediT authorship contribution statement

Chengsi Zhang: Writing - original draft, Writing - review & editing. Ning Zheng: Data curation, Conceptualization, Methodology,
Software.

Appendix A. Detailed derivation of the portfolio choice model in Section 2

A standard maximization problem by a representative firm of the expected utility is:

11
Note that the number of observations in the pre-2009 period is much smaller than that in the pro-2009 period. In the pre-2009 period,
accounting titles concerning financial assets are poorly disclosed leading to a lack of effective observations.

18
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Max E tU (It )
t=0 (1)
subject to:

It = (1 + rtk ) Itk + (1 + rt f ) It f (2)

(3)
f
K t = It + Itk

Assume a representative firm’s utility function is a constant absolute risk aversion utility function:
U (It ) = e It
(4)
Combining Eqs. (2) and (4) gives us
[(1 + rt f ) It f + (1 + rtk ) Itk]
U (It ) = e (5)
Return rates are uncorrelated and follow a normal distribution:

rt f Ñ(E (rt f ), Var (rt f ))


rtk Ñ(E (rtk ), Var (rtk )) (6)
Therefore the exponent of Eq. (5) is a linear combination of two uncorrelated random variables, thus also follows a normal
distribution:

[(1 + rt f ) It f + (1 + rtk ) Itk ]Ñ( [(1 + E (rt f )) It f + (1 + E (rtk )) Itk ], 2Var (r f )(I f )2
t t + 2Var (r k )(I k )2)
t t (7)
Taking the expectation of Eq. (5) gives us
[(1 + E (rt f )) It f + (1 + E (rtk )) Itk ] + 1 [ 2Var (rt f )(It f )2 + 2Var (rtk )(Itk )2]
E [U (It )] = e 2 (8)
Combining Eq. (1) and (8), the maximization problem becomes
1
Max [(1 + E (rt f )) It f + (1 + E (rtk )) (Kt It f )] [Var (rt f )(It f ) 2 + Var (rtk )(Kt It f ) 2]
2 (9)
The first order condition with respect to It is: f

E [rt f rtk] = [Var (rt f ) It f Var (rtk )(Kt It f )] (10)


Rearranging Eq. (10) yields:

It f Var (rtk ) E [rt f rtk ]


= +
Var (rt f ) + Var (rtk ) (11)
f
Kt Kt [Var (rt ) + Var (rtk )]

Appendix B. Detailed derivation on the dynamic structures of Eq. (17)

To show the dynamic mechanism of Eq. (17) explicitly, we utilize the lag operator L and a polynomial expression
(L) = 1 L + 2 L2 to transform Eq. (17) to the following:
k f k
FIit = cit + (L) FIit + 1 riskit + 2 rgap, it + controlit + it , (1)
or equivalently,
k f k
(1 (L)) FIit = cit + 1 risk it + 2 rgap, it + controlit + it (2)
Multiplying both sides with (1 (L)) 1 =1+ 1L +
2
2L + ( i (i = 1, 2, ) is functions of i ) eventually yields

k f k
FIit = 1n ci, t + n 1 + 2n risk i, t + n 1 + 3n rgap, i, t + n 1 + n controli, t + n 1 + 4n i, t + n 1,
n=1 n=1 n=1 n= 1 n=1 (3)
where and are combinations of the original coefficients. Hence, Eq. (17) captures the aggregate effects of both contemporaneous
and lag terms of explanatory variables.

Appendix C. Using fixed investment ratio as the dependent variable in Eqs. (15)–(17)

The fixed investment ratio FK is calculated as (fixed assets + intangible asset and other long term asset)/total assets. Other main
variables are consistent with Eqs. (15)–(17). We follow the same estimation procedure to obtain efficient estimates of main variables
of interests. For comparison purposes, we also report results from the unadjusted models.
Table 19 reveals that a higher relative risk of fixed investment discourages firms to invest in fixed assets in baseline models and

19
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Table 19
Estimation results for Eq. (15) and Eq. (16) (with fixed investment ratio as the dependent variable).
Baseline model Unadjusted model

Eq. (15) Eq. (16) Eq. (15) Eq. (16)

riskk −0.017*** −0.018**


(0.006) (0.009)
rf-k gap 9.217 4.904
(8.515) (5.644)
riskunadj −0.053 −0.094***
(0.033) (0.035)
runadj gap −0.050* −0.117***
(0.030) (0.042)
fc −0.326*** −0.189* −0.238** 0.162
(0.082) (0.102) (0.101) (0.112)
lev −0.088*** −0.062*** −0.090*** −0.062***
(0.003) (0.004) (0.004) (0.004)
asset 0.009** 0.003 0.009** 0.018**
(0.004) (0.008) (0.004) (0.008)
rev 0.028** 0.004 0.031*** 0.075***
(0.013) (0.019) (0.010) (0.026)
Number of firms 1902 1902 1902 1902
Number of observations 30,443 27,091 32,243 30,454
Year Dummy yes yes yes yes
KP-LM 0.092 0.012 0.000 0.000
Hansen J 0.244 0.459 0.205 0.478

Notes: See Table 3.

Table 20
Estimation results for Eq. (17) (with fixed investment ratio as the dependent variable).
Baseline model Unadjusted model

IV set collapsed IV set untransformed IV set collapsed IV set untransformed

***
FK(-1) 0.044 0.377 −0.008 0.471***
(0.223) (0.039) (0.130) (0.073)
FK(-2) 0.271*** −0.080*** 0.328*** 0.137**
(0.069) (0.031) (0.068) (0.055)
riskk 0.032 0.001
(0.019) (0.006)
rf-k gap −90.280*** −8.879**
(27.874) (3.829)
riskunadj 0.374 −0.002
(0.249) (0.017)
runadj gap −0.084*** 0.034
(0.031) (0.027)
fc −1.444*** −0.043 −1.392*** −0.076
(0.510) (0.051) (0.384) (0.055)
lev −0.145*** −0.117*** −0.127*** −0.054***
(0.039) (0.005) (0.026) (0.006)
asset 0.042*** 0.016*** 0.013 0.005***
(0.011) (0.002) (0.028) (0.002)
rev 0.140*** 0.015 0.001 −0.011
(0.042) (0.013) (0.003) (0.010)
Number of firms 1902 1902 1902 1902
Number of Observations 34,402 34,402 34,402 34,402
Year Dummy yes yes yes yes
AR(1) 0.011 0.000 0.000 0.000
AR(2) 0.523 0.240 0.851 0.153
Hansen-J 0.451 0.000 0.341 0.000

Notes: See Table 4.

20
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

Eq. (15) in the unadjusted model while the return gap has no significant influence on FK . Results from the unadjusted models,
however, show that the simple return gap between financial and fixed investment also negatively influences firms’ fixed investment
suggesting higher profitability from financial investment crowds out firms’ fixed investment. Regarding control variables, the ne-
gative signs of fc and lev imply that higher operating cash ratio and financial leverage suppress firms’ fixed investment, while firms
with larger asset size and stronger main business tend to increase fixed investment.
Results in Table 20 are more intriguing since risk k no longer has significant influence on FK , whilst the return gap significantly
crowds out firms’ fixed investment in baseline model unadjusted model with collapsed IV set which indicates that, indicated by the
dynamic nature of Eq. (17), the aggregate influence of return gap on firms’ fix investment outstrips that of the risk factor.

Appendix D. Statistics and histograms of average values of r f and r k over sample period

Note that average values of r f and r k over our sample period are both right skewed. With a near zero skewness and a kurtosis of
about 3.5, the average r k is distributed in a normal pattern, while average r f is more crowded around 0. Although these statistics do
not provide support for our assumption that r f and r k both follow normal distributions, they might shed light on our further research
where a more practical distribution will be considered (Table 21) (Fig. 2 and 3).

Table 21
Descriptive statistics of average values of r f and r k over sample period.

rf rk

Mean 0.027 0.075


Std. 0.034 0.081
Median 0.015 0.070
Skewness 2.301 0.217
Kurtosis 10.407 3.528

Fig. 2. Histogram of average values of r f .

Fig. 3. Histogram of average values of r k .

21
C. Zhang and N. Zheng North American Journal of Economics and Finance 53 (2020) 101215

References

Alvarez, I. (2015). Financialization, non-financial corporations and income inequality: The case of France. Socio-Economic Review, 13(3), 449–475.
Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty. The Quarterly Journal of Economics, 131(4), 1593–1636.
Barradas, R. (2017). Financialisation and real investment in the European Union: Beneficial or prejudicial effects? Review of Political Economy, 29(3), 376–413.
Barradas, R., & Lagoa, S. (2017). Financialization and Portuguese real investment: A supportive or disruptive relationship? Journal of Post Keynesian Economics, 40(3),
413–439.
Breusch, T. S., & Pagan, A. R. (1980). The Lagrange multiplier test and its applications to model specification in econometrics. The Review of Economic Studies, 47(1),
239–253.
Crotty, J. (2003). The neoliberal paradox: The impact of destructive product market competition and impatient finance on nonfinancial corporations in the neoliberal
era. Review of Radical Political Economics, 35(3), 271–279.
Davis, L. E. (2017). Financialization and investment: A survey of the empirical literature. Journal of Economic Surveys, 31(5), 1332–1358.
Demir, F. (2009). Financial liberalization, private investment and portfolio choice: Financialization of real sectors in emerging markets. Journal of Development
Economics, 88(2), 314–324.
Epstein, G. A., & Jayadev, A. (2005). The rise of rentier incomes in OECD countries: Financialization, central bank policy and labor solidarity. In G. A. Epstein (Ed.).
Financialization and the world economy (pp. 46–74). Northampton: Edward Elgar Publishing.
Hansen, L. P. (1982). Large sample properties of generalized method of moments estimators. Econometrica, 50(4), 1029–1054.
Hausman, J. (1978). Specification tests in econometrics. Econometrica, 46(6), 1251–1271.
Hayashi, F. (2000). Econometrics. Princeton: Princeton University Press.
Hovakimian, G. (2009). Determinants of investment cash flow sensitivity. Financial Management, 38(1), 161–183.
Kleibergen, F., & Paap, R. (2006). Generalized reduced rank rests using the singular value decomposition. Journal of Econometrics, 133(1), 97–126.
Krippner, G. (2005). The financialization of the American economy. Socio-Economic Review, 3(2), 173–208.
Le, Q. V., & Zak, P. J. (2006). Political risk and capital flight. Journal of International Money & Finance, 25(2), 308–329.
Mehrhoff, J. (2009). A solution to the problem of too many instruments in dynamic panel data GMM. Bundesbank Series 1 Discussion Paper No. 2009, 31, available at
https://ssrn.com/abstract=2785360.
Moosa, I. A. (2018). Does financialization retard growth? Time series and cross-sectional evidence. Applied Economics, 50(31), 3405–3415.
Orhangazi, Ö. (2008). Financialisation and capital accumulation in the non-financial corporate sector: A theoretical and empirical investigation on the US economy:
1973–2003. Cambridge Journal of Economics, 32(6), 863–886.
Sargan, J. D. (1958). The estimation of economic relationships using instrumental variables. Econometrica, 26(3), 393–415.
Shi, J., & Zhang, X. (2018). How to explain corporate investment heterogeneity in China’s new normal: Structural models with state-owned property rights. China
Economic Review, 50, 1–16.
Stockhammer, E. (2004). Financialisation and the slowdown of accumulation. Cambridge Journal of Economics, 28(5), 719–741.
Tobin, J. (1965). Money and economic growth. Econometrica, 33(4), 671–684.
Tori, D., & Onaran, Ö. (2017). The effects of financialisation and financial development on investment: Evidence from firm-level data for Europe. SSRN working paper,
available at https://doi.org/10.2139/ssrn.3064062.

Chengsi Zhang, Professor of Finance, China Financial Policy Research Center and School of Finance, Renmin University of China. His recent researches are published
in JIMF, JMCB, and IJCB, among many other journals.

22

You might also like