The effect of worked hours and heterogeneous labor adjustment costs on future stock returns

(1)

1

Master Thesis

The effect of worked hours and heterogeneous labor

adjustment costs on future stock returns

Name: Xiaohan Li

Student number:11377259

Supervisor: Dr. R. (Rafael) Perez Ribas

Date: June 2017

MSc Finance (Asset Management)

Amsterdam Business School

(2)

Statement of Originality

This document is written by Student Xiaohan Li who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Acknowledgement

I would like toexpress my sincere gratitude to all the persons who promote the progress of the research documented in this master thesis. The results reported in this thesis was completed with the help and support of my supervisor and fellow students.

First and foremost, I am truly grateful to my supervisor Dr. R. (Rafael) Perez Ribas for the useful comments, remarks, and engagement through the whole process of this master thesis. His patience, motivation, enthusiasm, and enormous knowledge help me a lot in writing of this thesis. During the process, He helped me come up with some novel contributions and told me steps to write an academic paper. He patiently corrected my writing and answered my questions timely. During the most difficult times when writing this thesis, he gave me the moral support.

Furthermore, besides my advisor, I also want to appreciate the various staffs of the Finance Group Team: Dr. J.E. (Jeroen) Ligterink, Dr. E. (Esther) Eiling, Dr. F.S. (Florian) Peters for their encouragement, insightful comments, and hard questions. They provided me with professional guidance and taught me a great deal about scientific research.

Finally, I am really grateful to my friends and family members who gave me immaterial assistance during this study period. I thank my fellow students Hongyue Gong, Yahui Hu, and Wenyi Wang for the stimulating discussions, the sleepless nights we were working together, and their selfless support. Last but not the least, I would like to thank the members of my family: my mom Yanhua Zhang and my father Guoyou Li for giving birth to me and for their constant love and support.

(4)

The effect of worked hours and heterogeneous labor

adjustment costs on future stock returns

Abstract

When the adjustment of firm’s employees is costly, the expenses on the input of labor force can affect firm’s market value. I document that current fast increasing rate of worked hours per employee indicates low abnormal returns in the following year, even after avoiding the predictability arising from other famous factors such as size, book-to-market ratio, and total factor productivity. I also argue that the heterogeneity of labor adjustment costs across different type of industries reflects different levels of expectations about firms’ future conditions. Thus, the predictive power of the increasing rate of each employee’s worked hours is distinct for industries with different characteristics.

(5)

1. Introduction

Forecasting future stock returns has a long tradition in academic finance and is a core issue in asset pricing literature since it is of great significance for both researchers and investors. Many scholars devote themselves to analyze the predictability for future stock returns arising from several firm’s features such as market capitalization, book-to-market ratio, capital investment, labor hiring and worked hours of employees.

In this thesis, I pay attention to the role of worked hours of employees and I examine the relationship between industry-level increasing rate of worked hours per employee and following year’s abnormal return predictability in 459 sub-industries of manufacturing industry. Firstly, I find that industries with currentfast increasing rate of worked hours per employee indicates low abnormal returns in the following year, even after avoiding the predictability arising from other famous factors such as size, book-to-market ratio, and total factor productivity.

Why labor hours have the ability in predicting future stock returns? In this paper, I argue that given future cash flows, firms can either increase employees’ worked hours or expand their capital investment and labor force if the discount rate in the future is low. The reason for this argument is that if firms’ current expenses on investment are less than the discounted value of firms’ future cash flows, firms will decide to increase inputs in their production processes to maximize profits. In other words, when future stock returns decrease, firms tend to expand their business and increase inputs in the production processes. Consequently, firms’ current production decisions can negatively predict future stock returns. Among these production decisions, employee’ s worked hour is the timeliest signal in forecasting future stock returns. This argument comes from the fact that changes in worked hours of employees are more common and frequent in daily operations without large adjustment costs. In contrast, when firms change their labor hiring plans or capital investment, they will face large fixed adjustment costs such as interviewing and training of new workers and purchasing some fixed assets. Unless the positive shock can cover the large adjustment costs arising from

(7)

expanding labor force and capital investment, firms prefer to increase labor hours. Li Gu and Dayong Huang (2014) find that worked hours of employees provide increasing amount of information in regard to future stock returns. However, in their paper, they just collect firms with distinct characteristics together in Fama-Macbeth regressions. This approach implicitly assumes that labor force is a homogenous input in firms’ production processes. In reality, firms differ in many dimensions and labor force is a heterogeneous input for different kind of firms, which may make the previous assumption implausible. For instance, consider firms that are labor intensive and need a significant amount of labor force to produce their goods and services. When this type of firms decides to change their worked hours per employee or labor hiring plan, it will face large labor adjustment costs to secure employees to complete the work. In this case, changes in employees’ worked hours should contain much information about firms’ expectations about future conditions. On the contrary, if a firm almost uses the capital to produce outputs, expenses on labor adjustment are small. Changes in worked hours should contain less information about future stock returns for this type of firms than labor-intensive firms. Similarly, if a firm is in the high-tech sector, the firm uses the most advanced technology and high-skilled workers for daily operation and future development. However, if the firm is in the low-tech sector, it operates with traditional and simple technology and ordinary workers. High-skilled workers are less substitutable than low-skilled workers. The unit change in growth rate of worked hours per employee will lead to higher labor adjustment costs for high-tech firms than low-tech firms. The more expenses on labor adjustment are, the less sensitively worked hours of employees respond to variation of the future stock returns. In addition, another missing part of this line of research is the consideration of the market fluctuations. Prior studies mainly use excess stock returns in the analysis approach, which lack control of market fluctuations. Different stock reacts differently to market fluctuations (different systematic risk). Thus, in this paper, I use Fama-French three-factor model to obtain industry-level abnormal returns in place of excess returns in order to control market fluctuations, since the abnormal return is the return achieved above the return that results from the correlation between the stock and the market.

(8)

Inspired by the ideas above, I propose an extension of Gu and Huang’s research. In this thesis, I also test the predictive power of the increasing rate of worked hours per employee for abnormal returns in the following year across different kind of industries by dividing all the industries in my sample into four sub-samples: labor intensive, capital intensive, high technology level and low technology level. I use two fundamental industry-level variables to classify industries: capital-to-labor ratio and technology respectively. Capital-to-labor ratio is defined as real capital stock scaled by total employment. Technology is calculated as R&D divided by total assets. First, I consider an industry to be capital intensive if it has the capital-to-labor ratio on the top 20th percentile of the cross-sectional distribution. Similarly, I consider an industry to be labor intensive if it has the capital-to-labor ratio below the bottom 20th percentile of the cross-sectional distribution. Second, I consider an industry to be innovative and with high technology level if it has “technology” on the top 20th percentile of the cross-sectional distribution. Similarly, I consider an industry to be traditional and less innovative if it has “technology” below the bottom 20th percentile of the cross-sectional distribution. I find that the coefficients of the increasing rate of worked hours per employee are more negative and significant for labor-intensive industries than capital- intensive industries. Furthermore, the negative connection between the rate of increase in worked hours per employee and abnormal stock returns is steeper in high-tech and more innovative industries than low-tech and less innovative industries.

Nevertheless, in-sample predictive power of predictable variables does not necessarily mean they also have out-of-sample forecasting ability. Thus, I also test the out-of-sample performance of the growth rate of worked hours’ prediction model. I find that the model used in this paper has positive out-of-sample R-square (1.08%) as long as the in-sample period which used for parameters estimation is long enough.

The structure of this paper is organized as follows. Section 2 introduces main theories related to the work in this paper and derives hypothesis. Section 3 reports the data. Section 4 presents the model to test the relationship between the increasing rate of each employee’s worked hours and future abnormal returns. Section 5 discusses empirical results. Section 6 draws conclusions and discusses potential limitations.

(9)

2. Literature review

This part reviews main theories in the prior literature which concentrate on the linkage between companies’ production decisions and stock returns predictability.

My empirical research is in the spirit of a large number of previous empirical studies which examine the predictability of several firm characteristics for stock returns, such as market capitalization, book-to-market ratio, hiring rate and investment rate.

My empirical research is also based on the labor and investment-based asset pricing studies. The study that firstly consider the role of investment initiates from Cochrane (1991). He constructs investment returns (marginal rate of transformation) from investment data and production function and he finds that stock returns equal investment returns. Cochrane (1996) extends his work in 1991 and proposes a simple factor pricing model in which investment returns act as a factor. He finds that investment returns can explain variations in expected stock returns both in cross-section and time series. In addition, Cochrane also discovers this simple investment return model behaves like the CAPM and performs basically better than traditional consumption-based model. The literature as I mentioned above establish a clear link between the role of investment and the expected stock returns, however, they all assume that labor force can be adjusted freely and cannot explain companies’ stock returns.

The pioneering research that considers the role of labor owes to the efforts of Merz and Yashiv (2007). They assume that labor hiring is similar to capital investment and that the two both have frictions in the adjustment. They find both labor hiring rate and investment rate are volatile and vital to explain the volatility of market value. Moreover, when they incorporate labor adjustment costs in the model, the model performs better than traditional specification (no hiring cost) in fitting in the features of the entire stock market price. Moreover, the interaction between investment and hiring is a key determinant of the market behavior. This finding is essential for my research since, without labor adjustment costs, the value of labor input cannot affect firms’ market value and workers can be adjusted freely. In this case, the growth rate of worked hours

(10)

per employee contains no information about firms’ future expectations and future conditions, then it has no ability in predicting future stock returns.

More recently, Bazdresch, Belo, and Lin (2014) extend the work by Merz and Yashiv from the entire stock market to individual company level to learn the cross-section of equity premiums. Furthermore, they pay attention to forecasting stock returns and they find that hiring rate has the ability to predict future stock returns after controlling for the predictability arising from investment rate and other predictors such as size, book-to-market ratio, and momentum. This research also confirms the existence of labor adjustment costs. The expenses on labor force include finding and training the new employees. Through these adjustment costs, they connect employees’ hiring with future stock returns. Inspired of their study, I consider when the worked hours of employees change, firms also face labor adjustment costs such as overtime wages and other benefits. In this case, the increasing rate of each employee’s worked hours is also possible to have ability in forecasting future equity premiums like labor hiring.

Bazdresch, Belo, and Lin (2014) also find firms’ workers with different characteristics have various degrees of influence on cross-sectional stock returns. They propose that labor force is heterogeneous investment in companies’ production processes and workers with different level of skills contribute differently to firms’ cash flows. They find it is more expensive to employ and dismiss a labor with technical skills than a labor with low skill level. Thus, the expected stock return–hiring relation is more negative in companies which have more expenses on the adjustment of proficient workers than companies with lower skilled workers. This study gives me the inspiration of considering the heterogeneity of employees across firms. Different type of firms requires different skill-level staffs. Some firms use the most advanced technology and they need high-skilled and high-educational workers to complete the daily work. However, some firms just operate with traditional and simple technology and they just need normal workers for daily operations. Different type of labor force incurs distinct levels of labor adjustment costs and contributes differently to future cash flows. Thus, I can also expect the predictive power of the increasing rate of worked hours each employee to vary across firms. For firms with many high skilled and educational

(11)

employees, a given magnitude of change in the worked hours should lead to a higher change in future stock returns than firms that just operate with low-skilled employees.

In contrast to Bazdresch, Belo, and Lin’s conclusion, Marcelo Ochoa (2013) argues that firm’s dependence on technical workers is a unique source of risk. He finds that when the market aggregate volatility is high, companies in which technical employees account for a large proportion of total employees gain higher yearly equity premiums over firms with a large percentage of ordinary staffs. This finding is opposite to the argument in my paper: skilled workers contribute more to firms’ cash flows and the large adjustment costs incurred by skilled workers reflect firms’ optimistic attitude towards future conditions.

Chen, Kacperczyk, and Ortiz-Molina (2007) report that companies in more unionized industries possess a higher required stock return for stockholders in general. They clearly demonstrate labor unions produce a significant friction that can reduce firm’s operational flexibility. As a consequence, company’s systematic risk increases and the required stock return for stockholders also rises. This finding also proves the existence of the friction of labor adjustment.

In the light of Miguel Palacios (2013), the wage of employee is a vital factor which influences firm’s cash flows. As a consequence, the company’s salary level should have the ability in explaining its returns and riskiness. He reports that companies which need a large number of employees operate with higher risk than capital-intensive firms and this type of firms tend to have higher factor exposure. This analysis helps me to consider a unique source of firms’ risk and focus on the difference between labor-intensive industries and capital-intensive industries.

Ernst R. Berndt, Catherine J. Morrison, and Larry S. Rosenblum (1995) report the relationship between the investment in corporate innovation and distribution of firms’ workers by testing the type of occupation and the level of education. They find that the increase in the investment in research and development is accompanied by the growth in white-collar workers who deal with administrative, technical, and skilled work. In addition, they also document the growth in the investment in research and development is associated with the increase in the employees’ educational level. This research

(12)

stimulates my interest in focusing on the distinction between firms’ innovative level and distribution of workers. Firms with high R&D investment need more educated and technical workers. This kind of staffs are harder and costlier to find and adjust.

In 2014, Ayşe İmrohoroğlu establishes a connection between the company-level total factor productivity and equity premiums in the following year. He argues that total factor productivity is an important factor which contributes to the values added of firms and it offers a wider measurement for firms’ performance than some traditional measures such as profitability and labor productivity which lacks adequate gauge of entire efficiency, especially in capital-intensive industries. In addition, he finds that firms with lower total factor productivity level earn significant stock premiums over firms with high total factor productivity level. Thus, in my paper, to avoid endogeneity problem, I use the industry-level total factor productivity as an additional control variable to avoid the predictability arising from total factor productivity.

Finally, my focus on the predictability of the increasing rate of worked hours per employee is relevant to the work by Li Gu and Dayong Huang (2014). According to Gu and Huang, when firm’s expected stock returns change, the firm can also change its worked hours of employees apart from capital investment and labor hiring. They argue that the increasing rate of worked hours is a timelier forecaster for future stock returns relative to capital investment and labor hiring. The reason for their argument is when a firm varies its capital investment or labor hiring, the large fixed adjustment costs will occur at the same time. Due to the existence of these large fixed adjustment costs, the firm’s investment and hiring may fall into zones of inertia, such that they have little predictive powers for future stock returns. In contrast, there is no fixed cost along with the adjustment of worked hours and change in worked hours per employee is more frequent and common such that the growth rate of worked hours per employee should performs better at forecasting future stock returns. Nevertheless, the work by Gu and Huang lacks analysis of heterogeneity of the predictive power of the increasing rate of worked hours per employee. They just assume that the increasing rate of worked hours per employee has the same predictive power across different type of firms. Obviously, it is not the case. Consider a company which nearly use the capital to create its products,

(13)

in this circumstance, the ability of worked hours for forecasting future equity premiums is weak. However, if a firm has a large number of employees and rarely use the capital to produce output, the growth rate of worked hours per employee should have strong predictive power for this type of firms. In addition, Gu and Huang use excess stock returns in the analysis approach, which lacks control of market fluctuations. Each stock reacts differently to market fluctuations (different systematic risk). In this paper, I use abnormal returns instead of excess returns in order to control market fluctuations.

According to Goyal and Welch (2006), they systematically test the in-sample and out-of-sample performance of linear regressions of some popular variables in the prior literature which are used to predict excess stock returns. They find that not only most predictable variables cannot beat the benchmark (historical average excess stock returns), but also the models outright underperform it. Therefore, the models would not help investors to obtain useful information regarding future stock returns. This paper gives me the inspiration of testing the out-of-sample performance of my predictive model. Significant in-sample forecasting ability of predictable variables does not necessarily mean they can also survive in out-of-sample. For investors, out-of-sample performance is the key.

My extension in this paper is clear and logical. When the growth rate of worked hours per employee changes, the labor adjustment costs are different across different type of industries. Since the labor adjustment costs contain useful information about the companies’ optimism degree of future circumstance and equity premiums, we can expect different predictive power of increasing rate of worked hours per employee for future abnormal stock returns across different industries. Based on ideas above, I come up with two main hypotheses used in this paper:

Hypothesis 1: The coefficients of the increasing rate of each employee’s worked

hours are more negative and significant for labor-intensive industries than capital-intensive industries.

Hypothesis 2: The negative relationship between the increasing rate of each employee’s worked hours and future abnormal stock returns is steeper in high-tech industries than low-tech industries.

(14)

3.Data

In this paper, the key explanatory variable is the increasing rate of worked hours per employee. Data on employees’ worked hours are accessible from NBER-CES Manufacturing Industry Database. This database includes annual data on employees’ worked hours, employment, total real capital stocks and increasing rate of total factor productivity (DTFP) for 459 sub-industries of manufacturing industry. The standard industrial classification number is between 2000 and 3999 and the time period is from 1963 to 2011. For each industry, the worked hours per employee are calculated as total worker hours per year scaled by the number of workers in the same year.The increasing rate of each employee’s worked hours at year t-1 is computed as the log difference of worked hours per employee between year t and year t-1. One of the industry-level classification variable capital-to-labor ratio which used to classify the different type of industries is defined as real capital stock divided by total employment.

The monthly stock data are available from the Center for Research in Security Prices (CRSP share code shrcd =10 or 11) database and monthly excess returns are returns that exceed the riskless rate. Companies’ accounting data is from the CRSP/COMPUSTAT Merged Annual Industrial Files. For stock data and accounting data, I include firms whose four-digit SIC are between 2000 and 3999, which represent firms of manufacturing Industry. The sample period is from January 1963 to December 2011.

Another classification variable which used to classify industries is “technology”. In order to obtain the industry-level classification variable, first I calculate each firm’s technology, which is defined as R&D scaled by total assets and then calculate the average of different firms’ technology within the same industry. The control variable size (SIZE) at year t-1 is defined as the natural log of price times shares outstanding at the end of year t-1. In addition, the control variable book-to-market ratio (BM) at year t-1 is calculated as the natural log of the ratio of book equity to market equity at the end of year t-1. I merge labor hour data with CRSP stock data and accounting data with four-digit SIC code.

(15)

The risk- free rates and the three Fama/French benchmark factors 𝑅𝑚− 𝑅𝑓, SMB, and HML are available from French’s website.

4. Methodology

4.1: Abnormal return

The main dependent variable of this research is abnormal returns, which is based on industry-annual frequency. The main explanatory variable increasing rate of worked hours per employee is also based on industry-annual level. This approach assumes that firms in the same industry have the same growth rate of worked hours per employee and ignore the firm variations within the same industry since firms in the same industry usually make similar real decisions in daily operations.

In order to calculate the abnormal returns for each industry and each year, first I need to calculate factor exposure beta. The model which I used to obtain abnormal returns is Fama-French three factor model.

𝑟_{𝑖,𝑦,𝑡}− 𝑟_{𝑓,𝑦,𝑡} = 𝛽_𝑖,𝑦1 _(𝑟 𝑚,𝑦,𝑡 − 𝑟𝑓,𝑦,𝑡) + 𝑒𝑖,𝑦,𝑡1 𝛽̂𝑖,𝑦1 = 𝑐𝑜𝑣(𝑟𝑖,𝑦,𝑡 ,𝑟𝑚,𝑦,𝑡 ) 𝑣𝑎𝑟(𝑟𝑚,𝑦,𝑡) 𝑒_{𝑖,𝑦,𝑡}1 = 𝛽_𝑖,𝑦2 ∗ 𝑆𝑀𝐵 +𝑒_{𝑖,𝑦,𝑡}2 𝛽̂_𝑖,𝑦2 =𝑐𝑜𝑣(𝑒𝑖,𝑦,𝑡 1 ,𝑆𝑀𝐵 ) 𝑣𝑎𝑟(𝑆𝑀𝐵) 𝑒_{𝑖,𝑦,𝑡}2 = 𝛼𝑖,𝑦+ 𝛽𝑖,𝑦3 ∗ 𝐻𝑀𝐿 + 𝑒𝑖,𝑦,𝑡3 𝛽̂𝑖,𝑦3 = 𝑐𝑜𝑣(𝑒_{𝑖,𝑦,𝑡}2 ,𝐻𝑀𝐿 ) 𝑣𝑎𝑟(𝐻𝑀𝐿) 𝛼_𝑖,𝑦= 𝑒_{𝑖,𝑦,𝑡}2 _{− 𝛽} 𝑖,𝑦3 ∗ 𝐻𝑀𝐿

Where: subscript i represents each industry, y represents each year and t represents each month.

𝑟_{𝑖,𝑦,𝑡}− 𝑟_{𝑓,𝑦,𝑡} is excess stock return for industry i at year y and month t

𝑟_{𝑚,𝑦,𝑡}− 𝑟_{𝑓,𝑦,𝑡} is market excess stock return at year y and month t

𝑆𝑀𝐵 (small minus big) is one of Fama/French three factors, which represents the average return on three small portfolios minus the average return on three big portfolios 𝐻𝑀𝐿 (high minus low) is one of Fama/French three factors, which represents the average return on two value portfolios minus the average return on two growth

(16)

portfolios

𝛼_𝑖,𝑦is abnormal stock returns for industry i at year y.

4.2: Predictive regression

To test the relationship between the abnormal return and previous year’s growth rate of worked hours per employee, I run pooled OLS regressions and use both entity(SIC) and time(year) fixed effects, controlling for three other well-known predictors: market capitalization (SIZE), book-to-market ratio (BM), and the growth rate of total factor productivity (DTFP). First, I run the pooled OLS regression for the full sample (459 industries) to examine whether the increasing rate of each employee’s worked hours has the ability in predicting future abnormal returns. Then, redo the predictive regression as below for each of the subsamples (labor-intensive industries, capital-intensive industries, high-tech industries, low-tech industries) to test the predictive power of the increasing rate of worked hours per employee across different kind of industries.

𝛼_𝑖,𝑡 = 𝛽₁𝐺𝐿𝐻_{𝑖,𝑡−1}+ 𝛽₂𝑆𝐼𝑍𝐸_{𝑖,𝑡−1}+ 𝛽₃𝐵𝑀_{𝑖,𝑡−1}+ 𝛽₄𝐷𝑇𝐹𝑃_{𝑖,𝑡−1}+ 𝛾_𝑡−1+ 𝑤_𝑖+𝜀_{𝑖,𝑡−1}

Where: subscript i represents each industry and t represents each year. 𝛼_𝑖,𝑡 is abnormal return for industry i at year t.

𝐺𝐿𝐻𝑖,𝑡−1 is the growth rate of worked hours per employee for industry i at year t-1.

𝑆𝐼𝑍𝐸𝑖,𝑡−1 is natural log of the market value of equity for industry i at year t-1.

𝐵𝑀𝑖,𝑡−1 is natural log of the ratio of book equity to market equity for industry i at year

t-1.

𝐷𝑇𝐹𝑃_{𝑖,𝑡−1} is the growth rate of total factor productivity for industry i at year t-1.

𝛾𝑡−1 is time-specific effects and 𝑤𝑖 is industry fixed effects.

𝜀_{𝑖,𝑡−1} is a disturbance term.

To compare the difference between abnormal returns and excess returns, I also test the relationship between excess returns and lagged one year growth rate of worked hours per employee.

(17)

𝑟𝑖,𝑡− 𝑟𝑓= 𝛽1𝐺𝐿𝐻𝑖,𝑡−1+ 𝛽2𝑆𝐼𝑍𝐸𝑖,𝑡−1+ 𝛽3𝐵𝑀𝑖,𝑡−1+𝛽4𝐷𝑇𝐹𝑃𝑖,𝑡−1+ 𝛾𝑡−1+ 𝑤𝑖+𝜀𝑖,𝑡−1

4.3: Out-of-sample performance

Next step I test out-of-sample performance of the prediction model proposed in the previous section. Following Campbell and Thompson (2008), I first split the dataset into an in-sample period and an out-of-sample period. The in-sample period is applied for parameters estimation and the out-of-sample period is used for evaluating the performance of the model. This out-of-sample test can also be treated as a robustness check for in-sample prediction.

Using all abnormal returns up to year t, I estimate the above predictive regression. Then, I use the estimated parameters to construct the forecast for the abnormal returns. (I use at least 25 years data to acquire preliminary parameters estimation and compare the out-of-sample performance of the model based on different length of the in-sample period).

𝜶̂𝒊,𝒕+𝒌 = 𝜷̂𝟏∗ 𝐺𝐿𝐻𝑖,𝑡+𝑘−1+ 𝜷̂2∗ 𝑆𝐼𝑍𝐸𝑖,𝑡+𝑘−1+𝜷̂3∗ 𝐵𝑀𝑖,𝑡+𝑘−1+𝜷̂4∗ 𝐷𝑇𝐹𝑃𝑖,𝑡+𝑘−1

To evaluate out-of-sample performance, I calculate out-of-sample R-square (𝑹_𝒐𝒐𝒔𝟐 _), where the benchmark is the historical mean abnormal returns 𝜶̅_{𝒊,𝟏:𝒕} calculated by using data up to year t. The out-of-sample R-square (𝑹_𝒐𝒐𝒔𝟐 _{) is calculated as:}

𝑹𝒐𝒐𝒔𝟐 = 𝟏 − ∑ ∑𝑵 (𝜶_{𝒊,𝒕+𝒌}− 𝜶̂_{𝒊,𝒕+𝒌})𝟐 𝒊=𝟏 𝒏 𝒌=𝟏 ∑ ∑𝑵 (𝜶_{𝒊,𝒕+𝒌}− 𝜶̅_{𝒊,𝟏:𝒕})𝟐 𝒊=𝟏 𝒏 𝒌=𝟏

Where: 𝜶̅_{𝒊,𝟏:𝒕} is the historical mean abnormal returns for industry i by using data up

to year t. n is the length of the out-of-sample period. N is the number of industries. If 𝑹𝒐𝒐𝒔𝟐 >0, the squared forecast error incurred by the growth rate of worked hours per employee is smaller than the forecast error comes from the historical mean abnormal returns. That is to say, the growth rate of worked hours per employee has better predictive ability than historical mean abnormal returns. The in-sample evidence of its predictability can also survive in out-of-sample test. In contrast, if 𝑹_𝒐𝒐𝒔𝟐 _{0, the} historical mean abnormal return is a better predictor than the increasing rate of worked hours per employee and the increasing rate of worked hours per employee has poor

(18)

out-of-sample forecasting ability.

5. Empirical Results

5.1. Summary statistics

Table 1 shows detailed industry-level summary statistics of both dependent and independent variables used in my empirical research during period 1963 to 2011. The table presents common statistics for abnormal returns (AlPHA), growth rate of worked hours per employee (GLH), natural log of the market capitalization (SIZE), natural log of book-to-market ratio (BM), growth rate of total factor productivity (DTFP) and Fama/French three factors (SMB, HML, MKT) for full sample and four different sub-samples: labor-intensive industries, capital-intensive industries, high-tech industries, and low-tech industries.

Panel A reports the summary statistics for the full sample. It can be seen from panel A that the largest annual increasing rate of worked hours per employee (GLH) is approximately 26.55%. In addition, the mean of the increasing rate of each employee’s worked hours is about -0.08% and its standard deviation is about 3.61%, which indicates changes in worked hours per employee in manufacturing industry is quite smooth. The average value of abnormal return (AlPHA) for the full sample is approximate -0.46 and from statistics percentile25, percentile50 and percentile75 of abnormal returns, we can speculate most abnormal returns of the full sample are negative during the sample period 1963 to 2011. As for control variables, the mean of the natural log of the market capitalization (SIZE), natural log of book-to-market ratio (BM), and growth rate of total factor productivity (DTFP) for the full sample is about 5.77, -0.5833 and 0.0051 respectively. From the statistics of total factor productivity, we can easily infer that most industries in my data have positive values of total factor productivity. Total factor productivity is a wider measurement which can gauge companies’ efficiency of using inputs. Positive values mean industries in my data see increases in technological innovations and improvements during this period.

(19)

Panel B reports statistics for labor-intensive industries. Two points in Panel B are worth emphasizing. First, the mean of the increasing rate of worked hours each employee (GLH) is about -0.09% and its standard deviation is about 3.91%, indicating changes in the increasing rate of worked hours of employee in labor-intensive industries are more frequent than the changes in the full sample. Second, the average value of the natural log of the market capitalization (4.4711) for labor-intensive industries is smaller than that of full sample and the natural log of the book-to-market ratio (-0.2641) is bigger than that of the full sample. These statistics indicate firms in labor-intensive industries generally have smaller sizes and they have value stocks which tend to trade at lower prices. These firms are considered with higher risk due to their small sizes. In addition, the mean of the increasing rate of total factor productivity for labor-intensive industries equals -0.0029 and is smaller than the mean for the full sample, indicating labor-intensive industries usually have lower efficiencies in using inputs of their processes.

Panel C reports statistics for capital-intensive industries. The average of the increasing rate of worked hours per employee (GLH) is about -0.12% for capital-intensive industries and its standard deviation is about 3.85 %, which is smoother than that of labor-intensive industries but more volatile than the value of the full sample. The mean of the natural log of the market capitalization (6.5255) for capital intensive industries is bigger than the means of both full sample and labor-intensive industries. Moreover, the average value of natural log of the book-to-market ratio for capital-intensive industries (-0.6998) is lower than those of full sample and labor-capital-intensive industries. These statistics imply firms in capital-intensive industries have large sizes and this type of firms usually well established and considered to be safe for investment. It is worth mentioning that the increase rate of total factor productivity is quite high in capital-intensive industries, with the average value of 0.0135 during period 1963 to 2011. The level of total factor productivity usually determines how efficiently and intensely firms use the inputs in their production process. Therefore, capital-intensive firms utilize inputs more efficiently.

(20)

D, we can find that the maximum value of annual increasing rate of worked hours each employee is approximately 21.96% for industries with high technology level. In addition, the mean of the increasing rate of each employee’s worked hours (GLH) is about 0.05% and its standard deviation is about 3.67%, which is more volatile than the standard deviation of the full sample. In addition, the average values of the natural log of the market capitalization, natural log of book-to-market ratio, and growth rate of total factor productivity for high-tech industries is about 6.0418, -1.0982 and -0.0052 respectively. These statistics signify firms in the high-technology sector generally have large sizes and low book-to-market ratio. High-tech firms usually demand high investment of information, knowledge, and research and development. This type of firms is usually considered to be safe for investment.

Panel E represents data of industries with low technology level. From this panel, we can see that the maximum value of annual increasing rate of worked hours each employee (GLH) is approximately 20.75% for industries with low technology level. Furthermore, the change in growth rate of worked hours per employee is much smoother than the change in high-tech industries since the standard deviation (3.23%) is a few percentages less than that of high-tech industries. Moreover, the mean of the natural log of market capitalization (5.7273) for low-tech industries is smaller than that of high-tech industries and the natural log of the book-to-market ratio (-0.3347) is bigger than that of high-tech industries. These statistics imply that firms in the low-technology sector generally have small sizes and high book-to-market ratios. This kind of firms operate with traditional and simple technology. It is natural for investors to believe firms with low technology level are risky, since they operate with lower investment in capital and knowledge, thus they cannot easily adapt to the rapid change of socio-economic conditions.

(21)

5.2. Predictive regressions

In this section, I run pooled OLS regressions of industry-level yearly abnormal

returns on the increasing rate of each employee’s worked hours. I control for three other famous predictors: size, book-to-market ratio, and growth rate of total factor productivity to avoid endogeneity problem in the regressions and report the regression coefficients and t-statistics in the tables. To better evaluate the forecasting ability of the increasing rate of each employee’s worked hours for abnormal returns in the following year across different type of industries, I also separately report regression results across different categories of industries: (1) labor-intensive and capital-intensive industries, measured by capital-to-labor ratio. (2) high-tech and low-tech industries, judged by industry-level “technology”. In all the regressions, I use full sample period from 1963 to 2011. I summarize this part by making comparisons of the magnitude and statistical significance of estimated results across four different type of industries.

5.2.1. Predictability in the full sample

Table 2 reports results of regressing yearly abnormal return and excess return on

lagged one-year increasing rate of each employee’s worked hours with full sample. The estimates in column 1 represent regression results of regressing abnormal return on worked hours per employee and the estimates in column 2 represent results of regressing excess return on worked hours per employee. Obviously, the results confirm the fact that current increase in worked hours of employees can negatively and significantly forecast abnormal returns in the following year. While for future excess returns, the increasing rate of each employee’s worked hours seems to have no predictive power for the full sample in my data. For instance, one unit increase in the growth rate of worked hours per employee is associated with a decrease of 6.14 % in the firm’s one-year forward abnormal returns. The t-statistic of the coefficient for abnormal returns is about -1.95, suggesting that the coefficient is statistically significant different from zero at the 90% confidence level. Turn to excess returns, we can see that the coefficient of the increasing rate of worked hours is about -0.0034, implying that

(22)

one unit increase in the growth rate of worked hours per employee is associated with 0.34 % decrease in the firm’s one-year forward excess returns. However, the t-statistic for excess returns equals -0.18, which is not statistically significant, indicating the increasing rate of worked hours per employee is not a good predictor for future excess returns.

As for control variables, both the variable market capitalization (SIZE) and the variable book-to-market ratio (BM) succeed to produce correct signs reported in the prior studies. For example, in prior studies, researchers argue that factor size has a negative connection with future equity premiums, since small stocks are much riskier than big stocks. From Table 2, we can see that in each column, the estimate of factor size is negative and the t-statistic is significant, implying that one point growth of industry’s market cap is combined with 1.04% decrease of one-year forward abnormal stock returns and 0.83% decrease of one-year forward excess returns. Similarly, researchers also argue that stocks with high book-to-market ratio tend to earn more equity premiums over stocks with low book-to-market ratio, since value stocks usually trade at lower price relative to their fundamentals. Table 2 clearly shows positive and significant relationships between industries’ book-to-market ratio and future abnormal returns (2.09%) and excess returns (2.47%). In addition, the sign of control variable “increasing rate of total factor productivity” is also consistent with the results in prior literature. Companies which have low level of total factor productivity are riskier than companies with high level of productivity, thus earn higher equity premiums. In column1 and column 2 of Table 2, we can find that the coefficients of the increasing rate of total factor productivity is approximate -0.0613 and -0.0514 respectively, suggesting that one point increase in the industries’ productivity level is accompanied by 6.13% decrease of abnormal returns and 5.14% decrease of excess returns in the following year.

(23)

5.2.2. Predictability across industries with different capital-to-labor ratio

In the previous section, I pooled all the industries in my dataset together in

regressions. This procedure implicitly assumes the growth rate of worked hours per employee has the same predictive power across industries. However, in reality, industries vary with different characteristics. Therefore, in this part, I examine the predictive power of growth rate of worked hours per employee across two different type of industries. Table 3 shows results of regressing yearly abnormal return and excess return on the increasing rate of each employee’s worked hours across industries with different capital-to-labor ratios. The estimates in column 1 and 2 stand for regression results for labor-intensive industries and the estimates in column 3 and 4show results for capital-intensive industries. From this table, we can find that in each column, the estimated coefficient of the increasing rate of worked hours is negative, whereas the size and t-statistic of the estimates change with the variation of industries’ capital-to-labor ratios.

When the response variable is industry-level abnormal returns, the coefficient and t-statistic of the increasing rate of each employee’s worked hours for labor-intensive industries are -0.299 and -9.44 respectively. While for capital-intensive industries the coefficient and t-statistic are -0.228 and -2.99 respectively. This result confirms the hypothesis 1 proposed in the literature review section: The coefficients of the increasing rate of each employee’s worked hours are more negative and significant for labor-intensive industries than capital-labor-intensive industries. Industries with high labor intensity make products and services depending mainly on employees’ effort in daily operations. When a positive shock appears, firms in labor-intensive industries will firstly consider raising worked hours of their employees to maximize production profits. In this case, firms face a significant number of labor adjustment costs such as overtime wages and other benefits. In contrast, capital intensive refers to the process mainly depends on equipment, vehicles, and facilities rather than labor force. Thus, when this type of firms changes the worked hours of employees, they will not face large labor adjustment costs as much as firms in labor-intensive industries. In other words, one unit

(24)

increase in growth rate of worked hours per employee will lead to higher labor adjustment costs for labor-intensive industries than capital-intensive industries. Since labor adjustment costs reflect firms’ expectations about their future operational status, we can expect a given magnitude increase in growth rate of worked hours per employee will lead to higher decrease in future abnormal returns for labor-intensive industries than capital-intensive industries.

Similarly, when the dependent variable is excess returns, the estimated results also verify the analysis above. The negative connection between the increasing rate of employee’s worked hours and excess stock returns in the following year is steeper in labor-intensive industries (-0.246) than capital-intensive industries (-0.11).

There are two points regarding the results worth mentioning: First, in each column, the estimated coefficient of the increasing rate of each employee’s worked hours is negative and significant, whereas the magnitude and t-statistic of the coefficients for future excess returns are less than the magnitude and t-statistic of the coefficients for future abnormal returns in both labor-intensive industries and capital-intensive industries. This result suggests that the increasing rate of each employee’s worked hours is a better predictor for future abnormal returns rather than excess returns, indicating when we take into account market fluctuations, the increasing rate of each employee’s worked hours performs better at capturing information for stock returns. Second, from Table 3, we can find no matter in labor-intensive industries or capital-intensive industries and no matter abnormal returns or excess returns, the estimated coefficients of the increasing rate of each employee’s worked hours are all more negative and significant than the coefficients for the full sample in Table 2. This outcome is a little puzzling since as I argue in the previous section, capital-intensive industries mainly depend on equipment and facilities rather than employees. Therefore, changes in worked hours per employee should contain less information about future stock returns for this type of firms, thus the coefficient of capital-intensive industries is supposed to be less significant than the coefficient of the full sample. Nevertheless, the truth is just the reverse. There are two dimensions of possible explanations regarding this contradiction. First, labor-intensive usually refers to a basic mode of the production

(25)

process which makes traditional and simple products and needs workers’ muscle and hands to complete work. Employees in this kind of industries are plentiful, unskilled, with low wages and educational level. On the contrary, capital-intensive production processes are mostly automated and mainly depend on equipment. Capital-intensive firms need to use the most advanced technology to better adapted to their environment and tend to survive in the stage of rapid economic development. Therefore, they need skilled workers with high educational level to develop the core competitiveness of the firms and maintain the most advanced machines. Obviously, employees with high skill and educational level are more difficult to find and less alternative than traditional common workers. High-skilled workers are more expensive to adjust. When the worked hours of this kind of employees are raised, they demand more overtime wages or extra benefits. As a consequence, when capital-intensive firms alter the worked hours per employee, they also face a certain number of labor adjustment costs. The estimated coefficients of the increasing rate of each employee’s worked hours for capital-intensive industries are significant as well. The size and t-statistic of the coefficients are steeper than those of the full sample. Second, there may exist economies of scale effect in capital-intensive industries. In other words, an equal unit increase in the inputs of the production process for capital-intensive industries will generate more products and yields than labor-intensive industries. Due to the existence of capital intensity in this type of industries, they tend to have a higher level of productivity and have the ability to generate more revenue and more profits. Therefore, when capital-intensive industries raise the same number of worked hours of employees as labor-intensive industries, both types of industries have equal inputs (labor adjustment costs). However, capital-intensive industries can generate more revenues and profits, hence lower stock returns than labor-intensive industries. Thus, we can also find the negative and statistically significant results between the increasing rate of each employee’s worked hours and future abnormal returns in capital-intensive industries.

Turn to control variables, for labor-intensive industries, both factor size and factor book-to-market ratio fail to provide correct signs: positive coefficients for factor size and negative estimates for factor book-to-market ratio. On the contrary, the results of

(26)

growth rate of total factor productivity are consistent with prior literature. For capital-intensive industries, all the control variables succeed to offer standard signs: negative coefficients for factor size, positive estimates for factor book-to-market ratio, and negative results for predictor total factor productivity.

[Insert Table 3 here]

5.2.3. Predictability across industries with different technology level

In this part, I test the forecasting ability of the increasing rate of each employee’s worked hours across industrieswhich are measured by different technology level. Table 4 shows results of regressing yearly abnormal return and excess return on the increasing rate of each employee’s worked hours for industries with high technology level and industries with low technology level respectively. The estimates in column 1 and column 2 represent regression results for industries with high technology level and the estimates in column 3 and column 4 represent results for industries with low technology level. From this table, we can confirm the negative and significant connection between the increasing rate of each employee’s worked hours and future abnormal returns and the connection between worked hours and excess returns for industries with high technology level, whereas, for industries with low technology level, the results are negative but not significant. For instance, when the dependent variable is industry-level abnormal returns, the estimated coefficient and t-statistic of the growth rate of worked hours per employee for high-tech industries are -0.137 and -9.3 respectively, indicating that one unit increase in the growth rate of worked hours per employee is associated with 13.7 % decrease of the firm’s one-year forward abnormal returns. While for industries with low technology level, the coefficient and t-statistic of the increasing rate of each employee’s worked hours are -0.0435 and -0.38 respectively. This outcome verifies the hypothesis 2 proposed in the previous section: The negative relationship between the increasing rate of each employee’s worked hours and future abnormal stock

(27)

returns is steeper in high-tech industries than low-tech industries. There are two possible reasons for this result: First, firms in innovative industries use the most advanced technology and require workers with high skills and educational level for daily tasks. This kind of firms spends a large amount of money on R&D investment and talented workers’ finding and training to maintain firms’ competitiveness and promote long-term development. On the contrary, if firms are in the low-tech sector, they operate with traditional and simple technology and they just need ordinary workers to complete the simple work which does not need high skills. Workers with higher skill and educational level are less substitutable than common workers. In other words, the adjustment of high-skill workers is much more expensive than the low-skill workers. When high-tech firms raise the worked hours of employees, they have to pay higher profits to their workers as compensation for overtime work, implying they are much more optimistic about the future conditions, thus higher revenues and lower stock returns than firms with low technology level. Second, the more expenses on employees’ adjustment, the less sensitively companies react to the changes of future stock returns. Therefore, when expenses on employees’ adjustment are higher, an equal change of the increasing rate of each employee’s worked hours lead to a higher change of future stock returns. Thus, we can expect the negative relation between the increasing rate of each employee’s worked hours is steeper in high-tech industries than low-tech industries.

Similarly, when the dependent variable is industry-level excess returns, the results are also consistent with the analysis above. The negative connection between the increasing rate of each employee’s worked hours and excess stock returns in the following year is steeper and more significant in high-tech industries (-0.0316) than low-tech industries (-0.0126). It is worth mentioning that the size and t-statistic of the coefficients for future excess returns are less than those for future abnormal returns in both high-tech industries and low-tech industries. This result also implies that the increasing rate of each employee’s worked hours performs better at forecasting future abnormal returns than excess returns, indicating that when we take into account market fluctuations, the increasing rate of each employee’s worked hours becomes a better predictor at capturing information for future stock returns. Turn to control variables,

(28)

for industries with high technology level, both factor size and factor book-to-market ratio succeed to provide correct signs: negative coefficients for factor size and positive coefficient for factor book-to-market ratio. On the contrary, the estimations of growth rate of total factor productivity are not consistent with prior literature. For industries with low technology level, only factor book-to-market ratio succeed to offer standard signs.

5.3. Out-of-sample performance

Nevertheless, in-sample predictive power of predictable variables does not necessarily mean they also have out-of-sample forecasting ability. In this section, I examine whether the in-sample proof of predictability for future abnormal stock returns arising from the increasing rate of each employee’s worked hours can also survive in the out-of-sample performance test. I do this out-of-sample test to be consistent with prior studies and this out-of-sample test can also be viewed as a robustness check for the in-sample proof. Out-of-sample performance is economically meaningful for investors since investors will only be able to invest today based on the estimation results made in the previous time period and earn tomorrow's strategy performance, not yesterday's. However, some academic economists argue that most predictive models have poor out-of-sample performance due to a systemic problem and a large number of predictive variables cannot even beat historical average returns in forecasting future stock returns. Thus, in this part, I use historical mean abnormal returns as the benchmark to evaluate the performance of worked hours of employees.

Following Campbell and Thompson (2008) and Rapach, Strauss, and Zhou (2010), I compare the performance between two models which are based on growth rate of worked hours per employee and historical mean abnormal returns respectively by calculating out-of-sample R-square (𝑹𝒐𝒐𝒔𝟐 ). Table 5 reports the estimated coefficient

(29)

𝛽̂ , t-statistic, in-sample R-square and out-of-sample R-square of the predictive regression of growth rate of worked hours per employee (GLH) based on different length of in-sample period and out-of-sample period. The total sample period is from 1963 to 2011. Panel A, Panel B, and Panel C use 37 years, 32 years, and 27 years respectively for in-sample parameters estimation. From this table, we can obviously find that the estimated coefficient of the growth rate of worked hours per employee keeps significant in all of three specifications. In addition, it is important to note that as the increase of the length of the in-sample period, in-sample R-square (𝑅_𝐼𝑆2_{) grows at} the same time, ranging from 16.75%, 18.98% to 20.53%. Similarly, when the in-sample period is not long enough, not surprisingly, I get negative out-of-sample R-square (-16.54% for 27 years in-sample period and -9,38% for 32 years in-sample period). While for the longest length of in-sample period estimation (Panel A), I obtain a positive out-of-sample R-square (1.08%). A positive value of the out-out-of-sample R-square indicates an increase in accuracy in forecasting future abnormal returns when moving from historical mean abnormal returns to the growth rate of worked hours per employee. Owing to the limitation on the length of the total sample period, I can only obtain a slightly positive value of out-of-sample R-square (1.08%). Nevertheless, this result is economically meaningful for mean-variance investors who desire to make utility and monetary gains based on information provided by changes of firms’ worked hours of employees.

5.4. Trends of the increasing rate of each employee’s worked hours and abnormal returns.

To clearly show the trends of worked hours per employee and abnormal returns, I plot the time series average values of the growth rate of worked hour per employee (GLH) and abnormal returns for the full sample and four different type of industries during period 1963 to 2011. Figure 1 represents time series trend for the full sample.

(30)

From this figure, we can easily find that the change of the growth rate of worked hour per employee (GLH) is within 20 percent and all average abnormal returns for the full sample during this time period are negative. It is clear that from 1963 there is a continuous decline in the abnormal returns and reaches its lowest point in 1980. Then, a sharp increase was witnessed from 1980 to 2011. It is also worth mentioning that the change in the growth rate of worked hour per employee and the change in abnormal returns are generally in opposite trends, indicating the negative relationship between these two variables. Figure 2 and Figure 3 plot time series trend of the increasing rate of each employee’s worked hours and abnormal returns for labor-intensive industries and capital-intensive industries respectively. Obviously, the variation range of the growth rate of worked hour per employee in labor-intensive industries is wider than that in capital-intensive industries, indicating changes of worked hours of employees in labor-intensive industries are more frequent and volatile than that of capital-intensive industries. This result is also consistent with the summary statistics in Table 1, where the standard deviation of the increasing rate of each employee’s worked hours for labor-intensive industries is 0.06% higher than the standard deviation for capital-labor-intensive industries. Figure 4 and Figure 5 show time series trend of the increasing rate of each employee’s worked hours and abnormal returns for high-tech industries and low-tech industries respectively. In both cases, the change in the growth rate of worked hour per employee and the change in abnormal returns show the opposite trends.

6. Conclusion

In this paper, I shed light on the effect of employees’ worked hours and heterogeneous labor adjustment costs on asset prices and I test the relationship between the increasing rate of each employee’s worked hours and future abnormal stock returns in 459 sub-industries of manufacturing industry. I document that the increasing rate of each employee’s worked hours is a strong predictor regarding future abnormal stock returns. In addition, my findings imply that labor adjustment cost reflects firm’s

(31)

expectations about its future conditions and it has a significant influence on asset price, and this influence varies with the characteristics of firms. Thus, the heterogeneity of labor adjustment costs across industries affects the predictive power of the increasing rate of each employee’s worked hours.

Nevertheless, there are also potential limitations regarding the approach used in this paper. For instance, the data of employees’ worked hours are only accessible at industry level of manufacturing industry. Therefore, this approach cannot capture firm variations and cannot figure out the relationship between labor hour and stock returns in other industries, making the results less convincing. In addition, although there is the convincing evidence for the negative relationship between the increasing rate of each employee’s worked hours and future abnormal returns in this paper, there may also exist another explanation. For instance, the negative relationship might just be the result of systematic mispricing when the market is inefficient. Studying another possible interpretation regarding the findings in the paper is vital for future research.

(32)

References

[1] Ayşe İmrohoroğlu, 2014, Firm-Level Productivity, Risk, and Return, Management Science 60, 2073-2090

[2] Bazdresch, Santiago, Frederico Belo, and Xiaoji Lin, 2014, Labor Hiring, Investment and Stock Return Predictability in the Cross Section, Journal of Political Economy 122, No.1, 129-177.

[3] Belo, Frederico, and Xiaoji Lin, 2012, Labor Heterogeneity and Asset Prices: The Importance of Skilled Labor, Working Paper, University of Minnesota.

[4] Bloom, Nicholas, 2009, The Impact of Uncertainty Shocks, Econometrica 77, 623-685.

[5] Campbell, John Y. and Samuel B. Thompson, 2008, Predicting Excess Stock Returns Out of Sample: Can Anything Beat the Historical Average? Review of Financial Studies 21(4), 1509-1531.

[6] Campbell, J.Y., 1996, Understanding risk and return, Journal of Political Economy, 104, 298—345

[7] Carlson, Murray, Adlai Fisher, and Ron Giammarino, 2004a, Corporate investment and asset price dynamics: Implications for the cross section of returns, The Journal of Finance 59 (6), 2577-2603.

[8] Chan, Louis K.C., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, The Journal of Finance, LI (5), 1681-1713

(33)

[9] Chen, Jason, Marcin Kacperczyk and Hernan Oritz-Molina, 2008, Labor unions, operating flexibility, and the cost of equity, Working paper, University of British Columbia

[10] Chen, Long, and Lu Zhang, 2011, Do Time-Varying Risk Premiums Explain Labor Market Performance? Journal of Financial Economics 99, 385-399.

[11] Cochrane, John H., 1991, Production-Based Asset Pricing and the Link Between Stock Returns and Economic Fluctuations, Journal of Finance 46, 209-237.

[12] Cochrane, John H., 1996, A Cross-Sectional Test of an Investment-Based Asset Pricing Model, Journal of Political Economy 104, 572-621.

[13] Cochrane, John H., 2011. Discount Rates: American Finance Association Presidential Address, Journal of Finance 66, 1047-1108.

[14] Cooper, Michael J., Huseyin Gulen, and Michael J. Schill, 2008, Asset Growth and the Cross-Section of Stock Returns, Journal of Finance 63, 1609-1651.

[15] Ernst R. Berndt, Catherine J. Morrison and Larry S. Rosenblum, 1995, High-tech capital formation and economic performance in U.S. manufacturing industries An exploratory analysis, Journal of Econometrics 65, 9-43.

[16] Fama, Eugene F. and G. William Schwert, 1977, Human capital and capital market equilibrium, Journal of Financial Economics, vol. 4(1), pages 95-125

[17] Fama, Eugene F., and Kenneth R. French, 1992, The Cross-Section of Expected Stock Returns, Journal of Finance 47, 427-465.

(34)

Returns on Stocks and Bonds, Journal of Financial Economics 33, 3-56.

[19] Fama, Eugene F., and Kenneth R. French, 1995, Size and Book-to-Market Factors in Earnings and Returns, Journal of Finance 50, 131-155.

[20] Gali, Jordi, 2005, Trends in Hours, Balanced Growth, and the Role of Technology in the Business Cycle, Federal Reserve Bank of St. Louis Review 87, 459-486.

[21] Hamermesh, Daniel S, 1989, Labor demand and the structure of adjustment costs, American Economic Review, 79 (4), 674-689.

[22] Hou, Kewei., and David Robinson, 2006, Industry concentration and average stock returns, The Journal of Finance, 61 (4), 1927-1956.

[23] Jermann, Urban J. 2010. The Equity Premium Implied by Production, Journal of Financial Economics 98, 279-96.

[24] Li Gu and Dayong Huang, 2016, The Effect of the Growth in Labor Hours per Worker on Future Stock Returns, Hiring, and Profitability, Review of Finance, rfw049.

[25] Marcelo Ochoa, 2013, Volatility, Labor Heterogeneity and Asset Prices, Working paper, Finance and Economics Discussion Series.

[26] Merz, Monika, and Eran Yashiv, 2007, Labor and the Market Value of the Firm, American Economic Review 97, 1419-1431.

[27] Miguel Palacios,2013, Labor Intensity, Risk and Return, SSRN Electronic Journal

[28] Petersen, Mitchell A., 2009, Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches, Review of Financial Studies 22, 435-480.

(35)

[29] Rapach, David E., Jack K. Strauss, and Guofu Zhou, 2010, Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy, Review of Financial Studies 23, 821-862.

[30] Rosett, J.G., 2001, Equity Risk and the Labor Stock: The Case of Union Contracts, Journal of Accounting Research, Vol. 39, pp. 337—64

[31]Santos, T. and P. Veronesi 2006, Labor Income and Predictability of Stock Returns, Review of Financial Studies, 19

[32] Sloan, Richard G., 1996, Do stock prices fully reflect information in accruals and cash flows about future earnings? The Accounting Review, 71, 289-315.

[33] Stambaugh, Robert F., Jianfeng Yu, and Yu Yuan, 2012, The Short of It: Investor Sentiment and Anomalies, Journal of Financial Economics 104, 288-302.

[34] Titman, Sheridan, K.C. John Wei, and Feixue Xie, 2003, Capital investments and stock returns, Journal of Financial and Quantitative Analysis, 39: 677-700

[35] Vuolteenaho, Tuomo, 2002, What drives firm-level stock returns? Journal of Finance LVII (1), 233-264

[36] Xing, Yuhang, 2008, Interpreting the value effect through the Q-theory: an empirical investigation, Review of Financial Studies, Forthcoming

[37]Yashiv, Eran, 2000, Hiring as investment behavior, Review of Economic Dynamics, 3, 486-522

(36)

Tables and figures

Table 1: Summary Statistics

Panel A: Full sample

statistics ALPHA GLH SMB HML MKT SIZE BM DTFP

min -1.5204 -0.2932 -15.28 -11.25 -23.24 -1.9021 -8.7586 -0.64 p25 -0.6749 -0.0192 -1.4 -1.21 -2.29 4.0239 -1.0815 -0.031 p50 -0.4343 -0.0005 0.07 0.25 0.74 5.8141 -0.5586 0.003 p75 -0.2149 0.0173 2.31 1.87 3.43 7.4491 -0.0462 0.039 max 0.1357 0.2655 18.73 12.91 16.1 13.1308 3.5900 1.2610 mean -0.4612 -0.0008 0.3072 0.3398 0.3708 5.7688 -0.5833 0.0051 sd 0.3162 0.0361 3.0586 3.0347 4.7142 2.4136 0.8476 0.0768

Table 1 reports summary statistics of industry level abnormal returns (AlPHA), increasing rate of each employee’s worked hours (GLH), three Fama/French benchmark factors (SMB, HML, MKT), natural log of market capitalization (SIZE), natural log of book-to-market ratio(BM) and growth rate of total factor productivity (DTFP) for full sample and four different sub-samples: labor intensive, capital intensive, high-tech and low-tech industries. Panel A reports summary statistics for the full sample. The sample period is from 1963 to 2011.

The effect of worked hours and heterogeneous labor adjustment costs on future stock returns

Master Thesis

The effect of worked hours and heterogeneous labor

adjustment costs on future stock returns

Name: Xiaohan Li

Student number:11377259

Supervisor: Dr. R. (Rafael) Perez Ribas

Date: June 2017

MSc Finance (Asset Management)

Amsterdam Business School

Statement of Originality

Acknowledgement

The effect of worked hours and heterogeneous labor

adjustment costs on future stock returns

Abstract

Contents:

1. Introduction

2. Literature review

3.Data

4. Methodology

5. Empirical Results

6. Conclusion

References

Tables and figures