Does the presence of industry-specific human capital help
explain the equity home bias?
Abstract
This paper assesses the effect of industry-specific labor income risk on the level of international diversification. When domestic equity is suitable as hedging instrument for labor income risk, it may help rationalize the strong preference for domestic equity in portfolios, a phenomenon called equity home bias. An analysis of the implication on optimal portfolio weights in face of labor income risk is conducted for the US, Germany and Finland. The results show that, primarily in the US, labor income risk in some industries should be hedged by adjusting an optimal portfolio towards domestic equity. Hence, the results offer some rational for the presence of equity home bias in the US, while domestic equity in Finland and Germany seems largely unsuitable to hedge labor income risk. Moreover, the results corroborate the notion that any analysis of optimal portfolio weights in the face of labor income risk should be conducted on the industry-level. The hedging demand for investors working in different industries is heterogeneous across all countries.
Master Thesis
Paul Gerd Halm
11943173
Thesis Supervisor: Dr. Esther Eiling
July 2018
Statement of originality
This document is written by, Paul Gerd Halm, 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 of the University of Amsterdam is responsible solely for the supervision of completion of the work, not for the contents.
1. Table of Content
1. INTRODUCTION 4
2. RELATED LITERATURE 7
2.1. THE CASE FOR HUMAN CAPITAL 10
3. DATA AND METHODOLOGY 13
3.1. DATA 13
3.2. EQUITY HOME BIAS MEASURE 15
3.3. METHODOLOGY 17
4. RESULTS 20
4.1. DESCRIPTIVE STATISTICS 20
4.2. CORRELATIONS OF HUMAN CAPITAL AND EQUITY RETURNS 21
4.3. HEDGING DEMAND DUE TO INDUSTRY-SPECIFIC HUMAN CAPITAL 26
5. ANALYSIS OF PORTFOLIO ADJUSTMENTS 27
5.1. DATASET 1 27
5.2. DATASET 2 31
6. DISCUSSION 34
7. CONCLUSION 35
1. Introduction
Wealth levels of different countries do not always move together. Prosperity of economies is dependent on diverse factors and primarily determined on a country level. Subsequently, investors can benefit from holding an internationally diversified set of assets. Table 1 illustrates that equity returns in different countries are not strongly correlated, indicating rather weak linear relationships between the variables. The observation that people show a strong preference for domestic assets in spite of this fact is called home bias. Literature has discussed the presence of home bias from a financial and macroeconomic perspective. Consumption home bias describes the potential diversification benefits of international consumption risk sharing.1 Equity home bias describes the lack of international diversification in equity
portfolios, and will be the focus of this study.
According to the International Capital Asset Pricing Model (ICAPM), an investor should hold domestic equity proportional to the size of her countries relative market capitalization to the world-market capitalization. For many investors, this implies a share of domestic equity below 1% in their portfolio. Empirically, however, a strong preference for domestic equity is observed and some countries show almost no foreign investment at all. French and Poterba (1991) demonstrate that already small levels of international diversification can yield substantial benefits. They reject institutional constraints such as transaction costs as explanation for equity home bias, rather suggesting that observed diversification levels stem from investor choices. The question arises whether this observation constitutes a behavioral bias or if market frictions shift optimal diversification towards domestic equities, rationalizing the home bias.
This paper investigates the effect of industry-specific labor income risk on the proportion of domestic equity in internationally diversified portfolios. Asset pricing theory implies that investors choose to hold assets to maximize the expected utility derived from their lifetime level of consumption. Human capital returns, defined as the growth rate of labor
1 Lewis (1999) explains consumption home bias well in a brief example: “[…]production is
exogenously given in each country. If individuals in each country share risk from their country-specific production processes, then they hold securities that pay out claims against each other’s production processes. In a complete market world economy, these claims represent Arrow-Debreu securities that encompass all states of the possible production outcomes. Thus, in equilibrium, individuals in different countries equalize their marginal utilities in each state of production outcomes.”
Table 1.
Correlations of Equity Index Returns
Panel A (1) (2) (3) (4) (5) (6) S&P500 (US) DAX (Germany) FTSE100 (UK) S&P/TSX (Canada) CAC40 (France) Nikkei225 (Japan) US 1.0000 0.5650* 0.8016* 0.2312* 0.1741 0.0234 Germany 1.0000 0.6686* 0.5694* 0.6739* 0.3821* UK 1.0000 0.3907* 0.1847* 0.1349 Canada 1.0000 0.7813* 0.6692* France 1.0000 0.6246* Japan 1.0000
This table shows simple regressions between quarterly index returns of six countries. The correlations are based on sample data from 1986Q through 2014Q4. * denotes significance at the 5% level.
income in this paper, are non-tradable and represents the largest share of household wealth. Estimates about the share of human capital in total wealth reach from 48% in Heaton and Lucas (2000) (see also: Eiling, 2013), to 90% for Lustig, van Nieuwerburgh, and Verdelhan (2010). Macroeconomic measures, calculating aggregate labor income as share of total economic size, usually yield values around 60% (Baxter and Jermann, 1997). In any scenario, human capital constitutes the most important determinant of the level of consumption that can be achieved. Hence, considering the non-tradable nature of labor income, investors that decide to hold financial assets such as equity, likely do so in order to hedge their human capital (Heaton and Lucas, 2000; Betermier, Jansson, Parlour, Walden, 2012). Conditional on observed negative co-movements between human capital and domestic equity returns, positive hedging demand arising from labor income risk may help rationalize the equity home bias.
Various studies reject significant correlations between equity and human capital returns based on empirical results (Fama and Schwert, 1977; Heaton and Lucas, 2000; Davis and Willen, 2000). Another prominent study documents strong positive correlations of labor and capital returns (Baxter and Jermann, 1997).2 However, Baxter and Jermann’s (1997) analysis
is based on a macroeconomic measure for capital returns and they reject the assumption that labor income growth rates are stochastic and hence the definition used in this paper. Contrary
2 Capital returns describe equity returns, although not exclusively. This paper uses the terms capital returns
and equity returns sometimes interchangeably. However, when the discussed relationship to human capital returns only applies to equity returns, this term is used.
to these findings, there are theories that suggest the return behavior of domestic equity, under certain circumstances, can be suitable as a hedge for human capital. Bottazzi et al. (1996) highlight the importance of redistributive shocks that affect human capital and equity returns asymmetrically. A redistributive shock, that is, a shock that reallocates income between labor and capital, makes returns less, potentially even negatively correlated. They observe negative correlations of real wages and equity returns between -0.39 and -0.43 across various OECD countries.
The argument for shocks to income distribution is reinforced when considering industry-specific human capital. Many redistributive shocks, for instance, productivity-enhancing innovations or a sudden increase in competition, occur on the industry-level. Moreover, Krueger and Summer (1988) point out that the risks associated with labor income can be very diverse across industries (Katz and Summers, 1989; Neal, 1995). Lastly, Fugazza, Giofre and Nicodano (2011) suggest that in contrast to aggregate wage levels, which tend to develop procyclical, industry-specific wages may move acyclical. Barsky and Solon (1989) explain potential acyclical movements on the industry-level with the fact that wage levels in some industries show strong seasonal (annual) movements. Based on these theories, any analysis on international portfolio choice should acknowledge the industry-specific dimensions of labor income risk and assess hedging demand due to human capital on the industry-level.
In order to assess portfolio choice implications due to industry-specific labor income risk, adjustment weights to optimal portfolio holdings that arise due to hedging demand are determined. This paper does not take a general equilibrium approach. Rather, it is assumed that investors already determined optimal portfolio weights based on a mean-variance framework. The adjustment weights are based on coefficients from OLS regressions of human capital returns on large equity index returns of different countries. The approach follows Eiling (2013), who develops a mean-variance asset pricing model that allows for non-tradable assets. The analysis is separated in two parts for two sample populations. The first part focused on the US, Germany and Finland and comprises data from 2000Q1 trough 2017Q4. The second part assesses a longer sample period in the US, starting in 1960Q1.
The study shows varied results. The first part of the analysis, based on the three OECD countries, yields findings similar to other research suggesting low co-movement of human capital and equity returns. Out of 27 industry-coefficients that describe portfolio adjustments to domestic equity, five are significant, indicating domestic equity is unsuitable to hedge labor income risk for the other industries in this sample. Three of the significant coefficients are
negative, indicating optimal portfolios should be adjusted towards smaller weights on domestic equity. The results of the second dataset appear different. For the quarterly analysis, all coefficients for domestic equity have a positive sign, and three out of nine are significant at the 5% level. Investors working in the construction-, service- and government-sector should adjust optimal portfolio weights between 6.4% and 16.44% towards domestic equity in order to hedge their industry-specific human capital.3 This increased demand offers some rationale for the
observed equity home bias in the US. However, the first analysis suggests that equity returns may in principal be unsuitable to hedge human capital. In any case, the results support the notions on industry-specific human capital of other studies like Fugazza, Giofre and Nicodano (2011). Optimal portfolio analysis in face of labor income risks should be conducted on industry-level labor income data, as based on the results in this analysis, hedging demands are heterogeneous across industries.
The rest of the paper is structured as follows. First, Section 2 surveys the relevant literature on equity home bias. In Section 3, data retrieval and the modelling of human capital returns and the home bias measure is explained. Section 4 shows descriptive statistics and discusses the suitability of equity returns as hedging instrument with simple correlations of human capital and equity returns. Section 5 presents an analysis of the hedging demand that arises due to industry-specific human capital. A discussion of the results and implications is imparted in Section 6. Section 7 concludes.
2. Related Literature
The existing literature has given two broad explanations for the presence of equity home bias. Much of the initial discussions focused on the first explanation: institutional constraints. Institutional constraints arise because international investment inevitably exposes one to the investment rules of the target country invested in. They comprise transaction costs due to financial market segmentation, capital controls such as withholding taxes and political risk, all of which lower expected returns from investing in foreign equity. Furthermore, one may be exposed to exchange rate risk and the political risk of changing regulations. In an early effort to incorporate these market imperfections, Black (1974) presented a model further developed by Stulz (1981) that introduces taxes as explicit barriers to foreign investment. Black (1974)
3 Taking into account the ratio of wealth tied up in human capital over financial wealth, up to which the
and Stulz (1981) challenge the ICAPM and show that it cannot explain the strong preference for domestic equity without explicit barriers between countries. Other efforts have similar approaches and suggest that the definition of taxes in these models represents various forms of transaction costs (Cooper and Lessard, 1981). Several studies point out the difficulties in quantifying the implications of institutional constraints (Cooper and Kaplanis, 1986; French and Poterba, 1991; Ahearne et al 2004). There is no widely accepted measure for the intensity of capital controls. Transaction costs, for instance, are estimated by Ahreane et al. (2004) using Elkins-McSherry Co.’s measure and range from 0.2% to 2.2% in 1997. Furthermore, political risk is practically impossible to measure consistently. Next to these observed difficulties, the economic significance of institutional constraints in international investment declined drastically. French and Poterba (1991) note that institutional constraints would have to be much higher in order to explain the observed level of international investment. Arguably, integration of financial markets and institutions is an ongoing process and will further decrease the role of institutional constraints in international investment decisions.
Although there is a decline in the equity home bias over recent decades a strong preference for domestic equity prevails. This preference seems to be a result of investor behavior which is the focus of the second branch of explanations for equity home bias. Investor behavior is driven by investor beliefs about risk and return; if these beliefs are incorrect, this leads to non-optimal outcomes. In 1990, Shiller et al. (1990) surveyed portfolio managers from the US and Japan and found that in both countries, managers estimates about domestic returns were relatively more optimistic . The perception of risk also plays an important role. Tversky and Heath (1991) found evidence which suggests people favor familiar concepts when they are prompted to take gambles and that they treat unfamiliar gambles as riskier. Interpreting this finding in an investing context suggests that investors would rather invest in an equity market that is familiar to them and that they assess unknown equity markets as riskier. This observation, in combination with the presence of what is commonly recognized as information barriers, may be another cause for equity home bias.
Information barriers arise because it proves more difficult to acquire information in other regions or countries than in your home country. Individuals consume news-media like newspaper, radio and television primarily from regional providers, hence naturally becoming more informed about companies and stock markets in said region or country. Moreover, it can be costly in terms of time and effort to obtain information about markets in other countries, and investors are limited with regard to time and personal cognitive ability. It may be
impossible to comprehend information of all major equity investment opportunities in the world in a timely manner. Ahreane et al. (2004) provides a convincing quantitative measure for information barriers, which are generally not directly observable. Using high quality crossholdings data of US investors, Ahreane et al. (2004) measure information bias as a fraction of a countries companies that are listed on a US stock exchange. Providing financial information under the strict regulatory environment of the US reduces the information barriers for US investors. The authors find that countries who have a lower fraction of US listed companies are more heavily underweight compared to the ICAPM.
Moskovitz (1999) builds on previous findings, thus discovering that preference for equities in proximity persists inside a countries borders. Studying mutual fund managers and data on company headquarter locations in the US, Moskovitz (1999) finds that, on average, a manager invests in equities that are 160 to 184 kilometers closer to her than benchmark equity indices. The characteristics of the companies in proximity suggest that this preference is due to information barriers. Small and highly levered firms that primarily produce locally consumed goods are most likely to locally provide easy access to information. At the same time, information for these type of companies is most valuable. Even though information barriers have likely decreased drastically since the introduction of the internet and the increasing range of financial media, those barriers certainly persist and belong to the most convincing explanations for equity home bias.
Previous literature offers further explanations for the presence of (equity) home bias. Political risk is a more general term for several risks international diversification entails. It describes regulatory constraints such as withholdings tax and capital controls, but also the likelihood of unforeseen changes of those regulations due to corruption and lack of transparency. Smimou (2013) introduces such “instability risks” in mean-variance models and, based on portfolio adjustments due to this factor, argues these risks are more relevant for investors than mean-variance characteristics. Dahlquist et al. (2003) call attention to the fact that poor investor protection in many countries leads to a substantial fraction of shares being held by large shareholders. Those shareholders usually do not trade their holdings meaning that the investment opportunity in that country does not equal the full market capitalization. The actually available world investment opportunity-set, termed “world float portfolio”, hence differs strongly from the world market capitalization. A regression analysis of equity home bias in international portfolio weights shows significant evidence for explanatory power of adopting the constrained “world float portfolio” in analyses. Other explanations have been
rejected by the literature. Observing equity home bias due to inflation hedging is doubtful, as theory and empirical evidence suggest the link between equity returns and inflation is too weak (Cooper and Kaplanis, 1994). Moreover, choosing to invest domestically to hedge nominal exchange rates is economically not sensible, given that bond investments provide excellent protection against this risk. The analysis of this paper tries to explain the preference for domestic equity by a hedging motive. The role of (industry-specific) human capital in this issue is discussed in the next section.
2.1. The Case for Human Capital
When an investor decides to hold financial assets such as equity, does she try to maximize her equity returns, or does she aim to maximize her level of consumption over time that can be achieved considering all her wealth resources? Next to financial assets, household wealth consists of several other components. Pensions, real estate, non-corporate businesses and durable consumption goods can all be part of household wealth, however, human capital makes up the largest share (Baxter and Jermann, 1997; Heaton and Lucas, 2000; Lustig, van Nieuwerburgh, and Verdelhan, 2010). As human capital is such a large determinant of the level of consumption, it stands to reason that investors want to hedge this asset. Prospect theory, as introduced by Kahneman and Tversky (1979), suggests that investors look at individual gambles separately rather than at their wealth as a whole. However, human capital is non-tradable, which implies that investors who try to maximize their level of consumption want to hedge this income stream.
Several studies try to assess the link of human capital returns and profit rates within a country to explain observed international diversification (Bottazzi et al., 1996; Baxter and Jermann, 1997; Davis and Willen, 2000b). Profit rates is a broad term describing the different macroeconomic and financial return measures that have been used in the literature. Next to financial measures such as equity and money market returns, human capital’s relation to fundamentals and national account data has been analyzed to explain the presence of home bias. Diverse definitions and assumptions possibly lead to the inconsistent conclusions of this literature. Generally, returns to non-traded human capital are less flexible than prices of equity, hence, shocks should affect equity and human capital returns asymmetrically and make equity
potentially suitable as hedge instrument for human capital.4
Baxter and Jermann (1997) provide one of the most referenced studies on this topic. Their analysis proposes a general equilibrium two-country model that is driven by productivity shocks. In the model, each country j uses two inputs: labor (Ljt) and capital (Kjt) to produce a
single good. The unexpected component of both labor and capital returns between two periods t and t+1 is derived as
𝑟#,#%&';) − 𝐸,𝑟#,#%&';) - = (𝐸#%&− 𝐸#)(1 𝜌3∆𝑑
';),#%&%6) 7
389
, (1)
where ∆𝑑';) is defined as the percentage change of labor and capital income between two
periods. The difference between expected and observed values of labor and capital returns strongly depends on rJ, the correlation between the two returns within a country. High observed
values for rJ in the data lead to very strong co-movements labor and capital. Consequently,
they propose optimal international diversification should include short positions in domestic equity, hence worsen the puzzle of international diversification.
A slightly different approach is offered by Dahlquist et al. (2004). An extended version the model of Baxter and Jermann (1997) allows for bond investments. In the model, investors always choose to hedge exchange rate risk with bond investments. They go on and argue that equity home bias depends on correlations of human capital (non-financial) risks and returns on equity investments they choose conditional on the bonds returns investors incur through exchange rate hedging. Applying the model to data on G-7 countries, Dahlquist et al. (2004) find negative and economically significant conditional correlations of equity returns and non-financial returns.
Bottazzi et al. (1996) challenge the conclusions of Baxter and Jermann (1997), who point out the importance of redistributive income shocks. For instance, electoral outcomes that entail a new labor market agenda promising to improve the bargaining power of labor unions. Another example constitute international price shocks of commodities such as oil. Those events are expected to affect wages and prices asymmetrically, leading to less and potentially negative co-movement between labor income and capital returns. The authors considered sixteen OECD
4 There are various shocks that describe different unexpected scenarios. Productivity shocks as in Baxter and
Jerman’s (1997) model describe unexpected changes in the productivity of labor or capital. Bottazzi et al. (1996) discuss redistributive income shocks. Some mentioned examples include positive labor demand shocks, trade shocks and shocks to the political business cycle. All of those are expected to affect the returns of capital and labor asymmetrically.
countries from 1970-1992 and did not find empirical evidence that returns to human capital and financial capital are highly correlated. Instead, the results suggest that redistributive shocks help explain the preference for domestic equity as they observe moderately negative correlation of human capital returns and equity returns. Various other papers stand in contrast to both aforementioned studies, documenting low correlations of human capital returns and equity returns (Fama and Schwert, 1977; Heaton and Lucas, 2000; Davis and Willen 2000b; Lustig and van Nieuwerburgh, 2010)
When the distinct characteristics of human capital and equity lead their returns to exhibit low co-movement during shocks, these findings are intensified considering the industry-specific nature of human capital. For instance, a positive supply shock in copper from newly found reserves and increased production in Chile could lead to a large price drop, which hampers growth in the Finnish mining industry while leaving other industries unaffected. Similarly, a change in sentiment on bonus payments at the end of a business cycle may have a much larger effect on wages in the finance industry.
Next to industry-level shocks affecting labor income, there is occupational-level income risk. Human capital is mostly developed and specialized in a single industry, which exposes one to its particular risks and opportunities (Katz and Summers, 1989). Compare, for instance, an accountant to a research chemist. While an accountant may face great uncertainty in job security in the long run, a research chemist’s performance may solely depend on who obtains the patent for a new drug first. Additionally, there are different degrees of specialization; these will determine how easy it is to transfer to another industry (Neal, 1995). The value of highly specialized skill sets like that of a surgeon, for example, is at risk in case technological advancements make human intervention in surgery redundant. It is important to understand that industries are exposed different dimensions of labor income. Hence, labor income in some industries may be stronger related with equity returns than others. Katz and Summers (1989) record substantial differences between industry wage levels in the US from 1960-1985. Davis and Willen (2000b) determine optimal portfolio weights based occupational-level labor income data. They find significant differences is in optimal portfolios between industries. Furthermore, Eiling (2013) shows evidence that suggests industry-specific idiosyncratic labor income risk is priced in the cross-section of stock returns. In contrast, aggregate labor income growth coefficients are insignificant, implying industry-level risk characteristics predominate aggregate risk factors in the general perception of labor income risk.
Transcending the debate whether inclusion of labor income worsens the observed home bias or not, this paper proposes that non-traded, industry-specific human capital has heterogeneous risk and return origins very different from equities, which can make equity investments suitable as hedge for human capital. Conditional on observed co-movements, the hedging demand created through industry-specific human capital may help rationalize the equity home bias.
3. Data and Methodology
3.1. Data
Determining the hedging demand that emerges when industry-specific human capital is treated as non-tradable component of the wealth portfolio requires equity returns that mimic the world investment-opportunity set and data on industry-level labor income returns. Additionally, data on foreign equity holdings is obtained in order to create a home bias measure and show the magnitude of under-diversification according to the ICAPM. The analysis of consists of two parts. First, due to limited data availability on quarterly industry-level wage data, three OECD countries, the US, Germany and Finland, are selected and a quarterly analysis of portfolio adjustments is conducted from 2000Q1 to 2017Q4. In the second part, the US Bureau of Economic Analysis (BEA) provides quarterly labor income data for nine industries starting 1960Q1. A longer sample period enables a more informative analysis of the portfolio adjustments that should be in face of labor income risk. Additionally, an analysis on annual returns is conducted.
For the analysis, large country indexes serve to represent a countries investment opportunity-set. While the actual opportunity set may be smaller due to majority shareholders or capital controls, it is a reasonable proxy for a countries market return. Quarterly data from 2000 to 2017 of the total return indices of the three OECD countries, France, Japan and China has been retrieved from DATASTREAM. Moreover, DATASTREAM provides index returns of Canada, which enables a regression analysis of six country index returns on US industry-level labor income returns from 1986Q1. Quarterly returns are compounded to arrive at returns for an analysis of annual returns. The chosen economies make up a share of 70% (61% for the second dataset starting 1986) of the total world market capitalization, which is a reasonable
share to represent the world investment-opportunity set.5
Some papers note a robustness to weather the equity returns are denoted in countries currency or in local currency (Bottazzi et al., 1996; Baxter Jerman, 1997). However, to take into account the exchange rate risk that arises with international diversification, index returns are translated into the local currency of the four countries. Data on quarterly exchange rates is obtained from the Federal Reserve Economic Data (FRED) website.6 Further, the index returns
are turned into excess returns to represent an investment environment where a person could always obtain the risk free return. 3-month Treasury Bill rates for each country are also retrieved from the FRED website.
Quarterly data on monthly per-person labor income on the industry-level from 2000Q1 to 2017Q4 is obtained from the statistical office websites of Germany, the UK and Finland.7
For the US, the State Quarterly Survey from the BEA provides absolute numbers on labor income since 1960Q1 through 2017Q4. These values are then scaled by the number of employees working in the respective industry, on which data is provided by the Current Employment Statistics (CES) of the Bureau of Labor Statistics.8
Nine sectors have been defined according to their SIC (NAICS for the US after 2001) codes: Construction, Finance, Manufacturing, Mining, Retail Trade, Service, Transportation, Wholesale Trade and the Government.9 Data for labor income in the finance industry starts at
1995, and data for labor income in the transportation industry starts 1976. A detailed description of the industry compositions is given in the appendix. This definition follows Eiling (2013), however, for the first dataset on three OECD countries it excludes the government sector due to data unavailability in Germany. Note that while some definitions include other branches (transportation, for example, includes the communications and utilities sectors), the industries do not span the whole economy (most notably, the agricultural sector is excluded). A more detailed explanation on industry definitions can be found in the appendix. The industry definitions in the State Quarterly Survey have changed from SIC to NAICS in 2001. Although
5 This is calculated with data on nominal GDP from the IMF’s World Economic Outlook 2018. 6 www.fred.stlouisfed.org
7https://www.destatis.de/DE/Publikationen/Thematisch/VerdiensteArbeitskosten/ThemaVerdiensteArbeitsko
sten.html; https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/earningsandworkinghours; http://pxnet2.stat.fi/PXWeb/pxweb/en/StatFin/
8 https://www.bls.gov/data/ provides data on employment levels. https://www.bea.gov/index.htm provides
data on industry-level wages.
9 Mining (SIC 1000-1499; NAICS 21), Construction (SIC 1500-1799; NAICS 23), Manufacturing (SIC 2000-3999;
NAICS 31-33), Transportation (SIC 4100-4999; NAICS 22, 48-49), Wholesale Trade (SIC 5000-5199; NAICS 42), Retail Trade (SIC 5200-5999; NAICS 44-45), Finance (SIC 6000-6799; NAICS 52-53), Services (SIC 7000-8999; NAICS 81) and Government (SIC 9100-9999; NAICS 90-99)
it is possible to map the SIC to NAICS code, for the second part of the analysis, per person labor income (Lt) for each industry k and quarter t has been calculated as
𝑅<,#=> = 𝐿<,#+ 𝐿<,#A&
𝐿<,#A&+ 𝐿<,#AB − 1, (2)
in order to smooth out artificial fluctuations. For country wide returns, data on aggregate per person labor income has been retrieved from the same statistical office websites. The resulting returns describe the quarterly change of monthly per person labor income.
3.2. Equity Home Bias Measure
Before 2001, data on foreign asset holdings was not provided on a regular basis. In 1997, the first Coordinated Portfolio Investment Survey (CPIS) was conducted by the IMF, which included 29 countries. The collected data includes foreign equity holdings as well as long- and short-term debt. In a comprehensive analysis of home bias measures, Mishra (2015) notes that the CPIS has some drawbacks. The way data is collected varies by country, such as how some countries enforce mandatory participation and the level the survey is conducted at varies between aggregate and stock-by-stock. Moreover, the survey does not address domestic portfolio holdings and it does not discuss a currency breakdown for the in US Dollar denoted holdings. However, up to today the CPIS provides the highest quality data on foreign equity holdings available.
After 2000 the IMF started conducting an annual CPIS which enables an empirical analysis of the equity home bias. Data on foreign portfolio holdings is necessary to understand how much investors invest internationally compared to domestic holdings. The ICAPM suggests that optimal diversification means weighting a country in a portfolio according to its relative economic size to the world economy. Accordingly, a common measure of the home bias is
𝐻𝑜𝑚𝑒 𝐵𝑖𝑎𝑠 = 1 − 𝑆ℎ𝑎𝑟𝑒 𝑜𝑓 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 𝐸𝑞𝑢𝑖𝑡𝑖𝑒𝑠 𝑖𝑛 𝐶𝑜𝑢𝑛𝑡𝑟𝑦W𝑠 𝑃𝑜𝑟𝑡𝑓𝑜𝑙𝑖𝑜 (𝐴𝑐𝑡𝑢𝑎𝑙)
𝑆ℎ𝑎𝑟𝑒 𝑜𝑓 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 𝐸𝑞𝑢𝑖𝑡𝑖𝑒𝑠 𝑖𝑛 𝑊𝑜𝑟𝑙𝑑 𝑃𝑜𝑟𝑡𝑓𝑜𝑙𝑖𝑜 (𝑂𝑝𝑡𝑖𝑚𝑎𝑙) . (3) According to the ICAPM, a value of 0 implies there is no home bias because the actual level of investment corresponds to the optimal level. A value of 1 means there is no international diversification at all. Optimal shares of foreign equities can be easily calculated with data on
market capitalizations. For this purpose, annual data on countries market capitalization of domestic firms is obtained from the World Bank’s World Federation of Exchanges Database.10
For the actual foreign equity holdings, the CPIS provides annual data since 2001. In order to derive the share of foreign holdings over total holdings, this paper adapts the definition provided by Mishra (2015). The total holdings are derived as foreign equity holdings and market capitalization of domestic countries less foreign liabilities.
𝐴𝑐𝑡𝑢𝑎𝑙 = 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 𝐸𝑞𝑢𝑖𝑡𝑦 𝐴𝑠𝑠𝑒𝑡
𝐹𝑜𝑟𝑒𝑖𝑔𝑛 𝐸𝑞𝑢𝑖𝑡𝑦 𝐴𝑠𝑠𝑒𝑡 + 𝑀𝑎𝑟𝑘𝑒𝑡 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 − 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 𝐸𝑞𝑢𝑖𝑡𝑦 𝐿𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑦 (4) The resulting home bias measure is shown in Graph 1 in the appendix. The graph includes data on nine OECD countries and India. For eight of the nine countries, there is a clear downward trend over the observed time period. After 2008, there is a slight rise in home bias for the Netherlands, Norway, Germany and the UK. This may be explained by the rise in uncertainty due to the recession, which increased the perceived risk associated with international investment. Next to the potential hedging motive due to labor income risk, there may be country specific conditions leading to equity home bias, especially in the countries with the highest observed levels, Finland, Japan and India.
For Finland, data on market capitalization was provided by the World Bank’s database only until 2004. Data on market capitalization as percentage of GDP and absolute GDP has been retrieved from the FRED website an multiplied, in order to be able to display the home bias measure for Finland until 2011.11 Except for Japan and India, Finland shows the strongest
home bias among all ten countries, with a large jump from roughly 0.5 to 0.9 on the home bias scale in 2001. The jump in the observed home bias can be rationalized by the strong decline in Finland’s market capitalization, leaving any observed level of foreign holdings further from its optimum. A reason for the decline was possibly slowing industrial output growth as well as a strong drop in Nokia’s share price during that time period.12 After 2001, there persists a high
level of home bias compared to the other seven countries showing a downward trend. According to a working paper by the European Commission, some European countries, including Finland, implemented substantial restrictions on foreign investment for institutional
10 https://data.worldbank.org/
11 The year the data on market capitalization for Finland on the FRED website stops.
12 The industrial output dropped below 3% in 2000 after 75 years of being consistently above 6%.
investors in 2001, making it hard to adjust their international diversification to a drop of the domestic market capitalization.
Japan and India show almost no international diversification at all and seemingly no downward trend. For India, this may be explained by the fact that it is a developing economy. Opportunities to invest internationally are constrained due to the low integration of capital markets. Investors can choose either to invest in an international equity mutual fund or a direct foreign investment via a broker; however direct foreign investment is legally restricted to $250,000 under the Liberalized Remittance Scheme (LRS) introduced by the Reserve Bank of India. The heavy focus on domestic stocks is a more difficult puzzle in Japan, as capital markets are well developed and no foreign investment constraints as in India exist. Japans economy offers attractive investment opportunities, however, this does not explain why investors relinquish the potential diversification benefits of even small levels of international diversification. A strong preference for domestic equity is indicated. A possible explanation may be strong optimism about domestic companies, as proposed by French and Poterba (1991).
Overall, there is a trend towards less equity home bias. The reason for this decline may be availability of higher quality information about stock markets, making it easier to attain and process information, hence facilitating international investment. Moreover, ongoing integration of capital markets may provide more diverse opportunities to invest internationally. The literature, however, has failed so far to explain remaining difference between optimal and actual diversification as either non-optimal behavior by investors or by market frictions that lead to a departure from the diversification the ICAPM proposes.
3.3. Methodology
When domestic equity investments are suitable to hedge the labor income risk of non-tradable human capital, adjustments to optimal portfolio weights towards domestic equity may help rationalize the home bias. This paper adopts Eiling’s (2013) methodology to derive adjustments to optimal portfolio weights in a mean-variance framework that allows for non-tradable assets.
It is important to recognize that the variation in results about the effect of labor income risk on home bias may be due to the definition of human capital returns. For instance, Baxter and Jermann (1997) use per person employee compensation as a measure for labor income (importantly, they were assessing the relation to capital returns, defined as GDP at factor cost minus employee compensation, instead of equity returns). They find strong positive correlations of up to 0.997 of labor income returns and capital returns, however, they note that
the growth rate of labor income exhibits much lower correlation with equity returns. Botazzi et al. (1995) develop a more sophisticated measure. Using total employee compensation, divided by the private consumption price deflator, and then normalized for the number of employees, they arrive at discrete conclusions to Baxter and Jermann (1997).
The difficulty of determining human capital returns is that the discount rate is not observed, which makes it problematic to derive the value of all future human capital. Following the approach of Jagannathan and Wang (1996), this paper uses the growth rate of per person labor income. Assume that r, the expected rate of return on human capital is constant, and that per-person labor income Lt follows an autoregressive process
𝐿# = (1 + 𝑔)𝐿#A&+ 𝜀#. (5)
The expected growth rate of labor income is denoted by g, and 𝜀t is an error term with 𝐸[𝜀#] =
0 and independent distribution over time. Human capital wealth is then defined in a simple Gordon Growth Model as
𝑊#=> = 𝐿#
𝑟 − 𝑔. (6) Moreover, under these assumption the return to human capital is given by
𝑅#=> = 𝐿#− 𝐿#A&
𝐿#A& . (7)
This approach allows for heterogeneity across industries. This is important, as one of the main points of this paper is that human capital has industry-specific risk and return characteristics, meaning that it would be incorrect to apply the same discount rate for each industry. An appropriate benchmark against the industry-level returns is aggregate per person labor income returns. Previous analysis of home bias and human capital focused on the relation of aggregate labor income and equity returns; most studies concluded that human capital does not explain, or even worsens, home bias. Looking at individual industries, however, suggests growth rates may be less influenced by country level events and more idiosyncratic.
Other studies on optimal portfolio implications of human capital propose to look at lagged labor income data. Jagannathan and Wang (1996), for instance, say that labor income is published with a delay and hence use 1-month lagged income data in their analysis. Baxter and Jermann (1997) even use up to 3 lags on annual data, however, their results are insensitive to lag lengths. More importantly, Heaton and Lucas (2000) argue that people have
instantaneous knowledge about the labor income they receive, connoting that they observe labor income and equity returns simultaneously. As this paper observes quarterly averages of monthly labor income returns, contemporaneous labor income returns are used in the main portfolio allocation analysis. Bottazzi et al. (1996) propose a potential inverse relationship, arguing that increased capital returns may lead to an increase in future wage levels. They test this relationship in AR regressions, however, reject the theory based on the results.
Eiling’s (2013) approach includes regression analyses of human capital returns of one specific industry on several defined industry equity portfolios within the US. In that, she determines how industry portfolios weights should be adjusted for a person working in one of those industries. For an analysis of the implications of industry-specific human capital risk on equity home bias, the focus should shift onto international investment opportunities.
Hence, an OLS analysis of industry-level human capital returns on a constant and equity index returns of the seven countries is conducted. Standard errors are based on Newey and West (1987) with two lags, to account for autocorrelation. Considering most OECD countries report positive economic growth each year, there is a strong reason to assume autocorrelation of labor income in the data. The equity returns of major indices of the included countries should represent the world diversification opportunities. Hedging demand is given by
− ∑ ∑A& #o,po𝑞n
#o , (8)
where the term ∑A&=
#o 𝑉𝑎𝑟[𝑟#o]A&, the inverse of the variance of the expected return of the
tradable (tr) component of the portfolio (equity). The second term, åtr,nt, is a N ´ K matrix of
the covariances of equity returns and non-tradable (nt) human capital returns. The negative sign indicates that the whole term denotes a hedging demand. Estimated regression coefficients are therefore multiplied by -1, and should be multiplied by qi, which is the ratio of the value of
an investor’s human capital over her financial wealth. If an investor’s human capital value is greater than the value of financial wealth, which is observed in most studies on the subject, a stronger adjustment than what the estimation coefficient’s suggest should be made.
The next three sections provide a throughout analysis of the co-movement of industry-level human capital returns and index returns and the hedging demand for domestic equity that arises through it.
4. Results
4.1. Descriptive Statistics
Table 2 shows the average quarterly change and standard deviation of monthly human capital returns for the eight industries. The first dataset of (1) trough (3) starts at 2000Q1 and has 72 observations for each country and industry until 2017Q4. The US data starts at 1960Q1 and contains 228 observations.
For the first dataset, different countries and industries exhibit different growth rates and variability in growth rates. Finland’s construction industry has a quarterly mean return 4-times as high as in manufacturing. In Germany, mean human capital returns are less dispersed across industries, however variability differs. Whereas the retail trade and service industry offer very similar returns, returns in retail trade have a standard deviation almost twice as high (1.9% and 0.9%, respectively). Comparing the US and Finland, returns seem quite similar, while returns in the respective industry in Finland are mostly higher and less volatile than in the US. This makes Finnish labor income seem more attractive than that of in the US; however human capital is not tradable and it is problematic to see these returns as interchangeable. Moreover, the returns are based on wage levels in domestic currencies and do not reflect the change in purchasing power parity.
For the second dataset concerning US, the values are less dispersed, however there are still substantial industry differences. The highest observed means are 1.511% for the finance industry and 1.236% for the mining industry. They also provide the highest standard deviations with 2.04% and 1.75% respectively. The lowest mean is observed in the retail trade industry with 0.81% and a standard deviation of 0.97%. The lowest standard deviation of 0.58% is given by the aggregate labor income return, which stands to reason as potential industry-level variation is smoothed out on the country-level.
It seems that there are some different characteristics of human capital between countries, however the within country differences are more pronounced. There are substantial enough differences in returns that one can assume hedging demand is heterogeneous for investors working in different industries. The results support the assumption that in order to assess the implications of labor income risk for optimal portfolio weights correctly, the analysis should focus on industry specific human capital.
Table 2.
Mean and Standard Deviation of Human Capital Returns
(1) (2) (3) (4)
Sector Germany US Finland US (1960-2017)
Construction 2.142% 0.714% 1.369% 1.072% (0.015) (0.023) (0.015) (0.0087) Finance 2.713% 0.840% 0.878% 1.511% (0.009) (0.031) (0.012) (0.0204) Manufacturing 2.256% 0.593% 0.347% 1.066% (0.011) (0.019) (0.22) (0.0109) Mining 2.174% 0.890% 1.306% 1.236% (0.017) (0.047) (0.024) (0.0175) Retail Trade 2.09% 0.51% 0.91% 0.810% (0.019) (0.012) (0.008) (0.0097) Service 2.142% 0.692% 1.253% 1.103% (0.009) (0.01) (0.007) (0.0101) Transport 2.31% 0.68% 0.844% 1.046% (0.011) (0.012) (0.01) (0.0144) Wholesale Trade 2.436% 0.670% 0.715% 1.094% (0.015) (0.015) (0.011) (0.0105) Government 0.994% (0.011) Country Average 2.294% 0.699% 0.820% 1.070% (0.008) (0.015) (0.008) (0.0058)
This table reports the mean and standard deviations for quarterly growth rates in monthly per person labor income . Column (1) through (3) is based on quarterly data over the sample period 2000Q1 to 2017Q4. Column (4) is based on a sample period from 1960Q1 to 2017Q4. Data on quarterly labor income is obtained from the respective statistical office databases. Standard errors are reported in the parentheses.
4.2. Correlations of Human Capital and Equity Returns
A prerequisite for equity returns to be suitable as a hedge for human capital returns, in the sense that it rationalizes observed home bias, is to observe negative or at least low correlation between the two. When two asset values have a positive correlation, then both tend to yield low returns at the same time. It is not common to hedge largely with short positions, unless assets are very highly correlated. Shorting exposes to higher transaction costs and margin accounts, as well as a potentially unlimited downside. Ideally, one asset should give higher
returns exactly during the time the other asset gives lower returns. Negative correlation between industry-level human capital and domestic equity returns would suggest that in a hedging portfolio for human capital returns, positive weights in domestic equity should be held. This observation can help to rationalize the home bias.
Table 3 shows the correlation coefficients of industry-level human capital returns and the returns of a large domestic equity index for the two datasets. In panel A, the correlations are
Table 3.
Panel A Panel B
(1) (3) (4) (1) (2)
Germany US Finland Quarterly
Returns Annual Returns Construction -0.1261 -0.0212 0.1228 -0.2605* -0.4442* Finance 0.1027 0.1478 0.3097* 0.1067 0.2475 Manufacturing -0.0151 0.0965 0.0701 -0.056 -0.0592 Mining -0.1695 0.0705 -0.0493 0.0088 -0.1985 Retail Trade 0.0913 0.1446 0.0582 0.0532 0.1394 Service -0.0474 0.1176 -0.0264 -0.1304* -0.3508* Transport -0.1394 0.0300 -0.0544 -0.1514* -0.3695* Wholesale Trade -0.1869 0.1782 -0.0335 -0.0574 -0.074 Government -0.1457* -0.3086* Aggregate -0.1172 0.1595 0.034 -0.1432* -0.2650*
This table reports the correlations of quarterly growth rates in monthly per person labor income of and quarterly returns of a domestic equity index. The correlations are for labor income of nine different industries and one aggregate measure. Correlations in Panel A are based on sample data from 2000Q1 to 2017Q4. Correlations in Panel B column (1) are based on a sample data from 1960Q1 to 2017Q1. Column (2) shows correlations based on annual data from 1960 to 2017. ret human capital returns and the domestic equity index returns over the period 2000Q1-2017Q4. * denotes significance at the 5% level.
based on quarterly data from 2000 to 2017. All four countries exhibit only a weak correlation of aggregate human capital returns and equity. The only significant coefficient is for the Finnish labor income finance industry and domestic equity is 0.31 and significant, implying a moderately positive correlation. A possible explanation for this observation can be the higher share of bonus payments in total compensation in the financial service industry. Bonus payments like stock options in turn depend on returns in equity markets. The remaining coefficients in panel A exhibit only weak and insignificant values, however, the signs vary between countries and industries. The insignificance may be either because the estimates contain noise, or, consistent with many studies in this field, there is actually low co-movement. Possible noise could be introduced by two crisis that occurred during the 18-year sample period. The dot-com bubble as well as the 2008 financial crisis had a strong effect on equity prices around the world.
Panel B exhibits correlations for quarterly and annual returns within the US. With the sample data starting 1960, this analysis may offer more consistent coefficients. For quarterly data (column 1), seven out of ten show a negative sign for the estimate, although their magnitude is rather low. Remarkably, five out of those are deemed significant. The construction sector shows the strongest correlation with -0.26. The aggregate, transportation-, service-, government-sector have significant correlations between -0.13 and -0.16. The strongest positive coefficient is observed for the finance industry with 0.11. Whereas the estimates still describe a moderate co-movement at most, the results point towards an overall negative relationship of human capital returns and equity returns in the US.
This observation is amplified looking at the annual data. The correlations of annual equity and human capital returns is interesting for investors who reallocate the capital of their portfolio less frequently. When an investor plans to readjust her portfolio only every year, she is more interested in less frequent, annual correlations. Column 2 in panel B exhibits stronger evidence for negative co-movement. The same industries from the quarterly data are significant and have increased magnitude. In contrast, the coefficients for retail trade and finance industry increased, with finance showing an estimate of 0.25. Four industries have a coefficient above -0.3, with the construction being the strongest at -0.44. Contrary to the quarterly data, the magnitude of coefficients of the transport and construction industries increased the most. Especially for the construction industry this may suggest income is more seasonally dependent, meaning it provides different wage levels during a year. Overall, there seems to be a country wide trend towards negative co-movement, however, inter-industry differences persist. Two industries
offer a positive coefficient, and the difference between the highest and lowest estimate in column (2) is +0.69 (although the coefficient for the finance industry is insignificant at the 5% level).
It is important to highlight that a correlation coefficient of time series data assumes a constant relation between labor income growth and equity returns over the entire sample period. This implies that the relationship between the two returns did not change between 1960 and 2017. This is a strong assumption which is questionable here. Hence, the discussion in Section 6 provides a review of 60-month rolling window correlations of the human capital and equity returns in the US.
According to classic economic and asset pricing theories, the findings in Table 2 may appear puzzling. Equilibrium models that establish the relationship of capital, labor and output with a Cobb-Douglas production function suggest a strong positive relationship of aggregate equity returns and the value of (aggregate) human capital (Davis and Willen, 2000b). There are some explanations for why the observed correlations in panel B deviate from what is proposed by these models. Firstly, equity returns are used in this analysis instead of macroeconomic return measures for capital like GDP growth, for instance. Equities are tradable assets for which a market price is determined based on its fundamental value but also by supply and demand, which in turn largely depends on investor sentiment. Although equity indices take away most of the idiosyncratic factors influencing stock prices, it still depends on market conditions. Human capital, on the other hand, is non-tradable and its value is not determined by a market. On the aggregate level, labor income is determined by factors like productivity, inflation and regulations.
Moreover, there may be redistribute shocks, which do not affect equity and labor prices symmetrically (Bottazzi et al., 1996). On the country-level, this may be, for instance, a raise in minimum wage. On the industry-level, possible influences include labor income innovation, competition and price fluctuations for industries which profits depend on commodities (Davis and Willen, 2000b). Even though these factors are more likely to affect equity prices as well, equity indices are well diversified and likely filter them out.
Furthermore, the index returns are excess returns, meaning the risk free rate is deducted. A rise in the risk free rate therefore leads directly to lower excess returns of equity. The relation on labor income is less straightforward, but moving in the opposite direction. Lower interest rates raise inflation, which in turn lowers unemployment. Lower unemployment increases wage inflation and therefore, ceteris paribus, labor income. It is noteworthy that recent analysis
from the OECD and IMF show the relationship between unemployment and inflation, captured in the so-called Phillips curve, suggests the relationship has weakened over the last decades (Dotsey, Fujita and Stark, 2017). This would imply an increase in the risk free rate lowers index returns, while simultaneously leaving labor income largely unaffected.
Several papers document low co-movements of human capital and equity returns similar to the observation in panel A. In an often cited paper, Fama and Schwert (1977) find weak correlations of aggregate human capital and equity returns for the US, applying a similar assumption as this paper about labor income growth. Further, Davis and Willen (2000b) assess correlations of labor income shocks and equity returns, and find low observed estimates between -0.1 and 0.2. While panel A seem to be in line with those previous findings, the correlations in panel B, especially for the annual data, show evidence for a negative relationship for some industries (and aggregate) human capital and equity returns.
One of the most important differences for all studies on this subject seem to be the definition of human capital returns. Closest to the analysis conducted in this paper comes Fama and Schwert’s (1977) work. In a different framework of similar assumptions, they also derive returns to human capital as the growth rate of labor income. Aggregate equity returns are represented by value-weighted portfolios of NYSE stocks and the analysis contains regressions for monthly and annual data. Nominal returns show weak correlations with positive signs on the monthly data, but a negative significant estimate of -0.11 for annual data. Similar to Table 2 panel B, the correlation becomes increasingly negative for longer horizon. The magnitude of the coefficients, however, is greater for some industries than in Fama and Schwert (1977). A possible explanation is a change in the return behavior of the risk-free rate. Graph 2 in the appendix shows the historic US risk-free rate from the FRED website which are used in this paper to calculate excess returns for an US investor. Over the period of 1953-1972, the risk-free rate was arguably more stable and had a lower mean. Under the assumption the relationship of the Phillips Curve holds over much of the sample period, a generally higher and more fluctuating risk-free rate possibly facilitating negative correlations.
The results observed in panel B are closest in line with the findings of Bottazzi et al. (1996). The authors assume stochastic labor incomes and model real wages and profit rates in a continuous-time VAR model. For financials they test an aggregate measure composed of stocks, bonds and short-term deposit rates, as well as leveraged stock returns. The average observed correlations range between -0.39 and -0.43, which is similar to the estimates for annual data in panel B. They are able to explain around 30% of the observed equity home bias.
Although results are robust across country and for different measures of capital returns, there are some differences. Most notably, the financial measures in the US are in line with an observation of home bias while the fundamental approach yields positive correlations with wage rates. Bottazzi et al. (1996) argue that shocks to income distribution between labor and capital within a country can be responsible for observed negative correlations. Conducting any analysis on industry-specific human capital returns should then reinforce this argument, as some shocks to income distribution occur only on an industry-level. The next section will describe an analysis of portfolio weight adjustments based on observed co-movements of industry-level human capital and equity returns.
4.3. Hedging Demand due to Industry-Specific Human Capital
According to traditional asset pricing theory, investors try to maximize the utility of their expected consumption by choice of their asset portfolio. Human capital is the largest component of individual and household wealth, and its non-tradable nature leads to the assumption that an investor would like to hedge the returns of human capital. Next to other potential investments like bonds, well-developed and active markets present stock investments as the most apparent choice as hedge for human capital. When a person decides to invest in equity, it may well be in anticipation to unsatisfying income streams or because current income levels are perceived to be too low. In order to maximize possible consumption, an investor should adjust their equity portfolio according to the labor income risk they face.
The following analysis aims to determine adjustments in optimal portfolios weights that arise due to the hedging demand of human capital. Optimal portfolio weights depend on desired levels of risk and return and will be discussed in this paper only theoretically in a mean-variance framework following Eiling (2013), who derives an one-period asset-pricing model that allows for non-tradable assets. In a portfolio optimization problem which is further described in the appendix, she shows that optimal portfolio weights 𝑥n in tradable assets for investor 𝑖 is given by
𝑥n = 𝛾nA&∑ #o A&𝜇
#o− ∑#oA&∑#o,p#𝑞n. (9)
The first term on the right hand side is the same as described in the CAPM. It implies the level of investment in tradable assets inversely depends on the risk aversion and the variance of the return of a tradable asset. The introduction of non-tradable assets leads to the second term,
which is essentially the hedging demand due to human capital (as indicated by the negative sign). It depends inversely on the variance of tradable assets, and positively on the ratio of human capital wealth over financial wealth and the covariance between tradable and non-tradable assets.
The higher the covariance of returns of one particular tradable asset and the industry-specific human capital of investor 𝑖, the smaller should be the amount 𝑥n invested in that asset. Treating equity returns as exogenous, simple OLS regressions of human capital returns on the returns of large equity indices will help determine this relation and show if the introduction of (industry-specific) human capital suggests portfolio adjustment towards domestic equity, hence rationalizing the equity home bias.
5. Analysis of portfolio adjustments
5.1. Dataset 1
For the first part of the analysis, quarterly data on industry-level, per person labor income is available from 2000 for three OECD countries. Tables 4 through 6 show correlation coefficients for per person labor income returns and equity returns of indices in seven different countries. The sign of the estimated coefficients has been inverted to describe the hedging demand, and they should be multiplied by 𝑞n to account for the value of human capital over financial wealth. The bold coefficients denote the adjustments that should be made to domestic equity.
Each row in the tables represents the portfolio adjustments in each of the seven countries for a person working in one industry. A first look at the tables shows that, based on coefficients significant at 5%, the portfolio adjustments due to hedging demand are heterogeneous across industries. For instance, assuming a q of 4 (80% of household wealth is due to human capital), an international investor working in the German manufacturing industry should underweight US equity investment by 29.08% (7.27%´4). By contrast, a person working in the German transportation industry should overweight French equity by 46.8% and underweight German equity by 31.4%. Considering that people working in different industries might have different ratios of 𝑞, this observation is amplified. Someone working in the finance industry might hold a substantially larger share in financial assets than someone working in the retail trade industry. Following, a Finnish investor in working in finance with a q of 1.5, should overweight Japanese equity by 7.59% and underweight Finnish equity by 4.88%. In contrast,
Table 4.
Hedging Demand of Industry-Specific Human Capital - US
(1) (2) (3) (4) (5) (6) (7)
Sector S&P500 (US) (Germany) DAX FTSE100 (UK) (Finland) OMXH (France) CAC40 Nikkei225 (Japan)
SSE Composite (China) Construction 0.0752 -0.138* -0.0264 0.0019 0.171* -0.0291 0.0256 (0.132) (0.072) (0.613) (0.932) (0.068) (0.251) (0.375) Finance 0.0326 -0.162** 0.1070 0.0151 0.1000 -0.0924** 0.0383 (0.611) (0.018) (0.180) (0.651) (0.373) (0.046) (0.322) Manufacturing -0.0021 -0.142*** 0.0420 0.0491** 0.1100 -0.0209 0.0157 (0.955) (0.003) (0.296) (0.023) (0.102) (0.325) (0.604) Mining 0.0215 -0.264* 0.1620 0.0960** 0.1180 -0.0146 0.0531 (0.813) (0.061) (0.364) (0.036) (0.451) (0.798) (0.371) Retail Trade -0.0104 -0.0559 0.0322 0.0107 0.0489 -0.0169 0.0128 (0.657) (0.125) (0.344) (0.573) (0.315) (0.354) (0.429) Service -0.0258 -0.0556* 0.0344 0.0003 0.0785* -0.0113 0.0182 (0.129) (0.070) (0.251) (0.979) (0.076) (0.281) (0.177) Transport 0.0542** -0.0658* -0.0358 0.0163 0.0595 -0.0119 0.0021 (0.036) (0.078) (0.429) (0.241) (0.160) (0.517) (0.785) Wholesale Trade 0.0078 -0.0747 0.0540 0.0200 0.0411 -0.0154 0.0047 (0.780) (0.100) (0.189) (0.205) (0.513) (0.362) (0.837) Total 0.0234 -0.134*** 0.0591 0.0317** 0.0903 -0.0266 0.0242 (0.488) (0.008) (0.230) (0.041) (0.174) (0.217) (0.412)
This table reports the estimated coefficients of OLS regressions of quarterly growth rates in monthly per person labor income on equity returns of large country indices and a constant. The coefficients represent adjustment weights to optimal portfolios for an investor working in the respective industry. Data on equity indices returns is retrieved from DATASTREAM. Equity returns are calculated as quarterly compounded returns from 2000Q1 to 2017Q4. Human capital returns are calculated as the quarterly growth rate of monthly per-worker labor income. These adjustment weights are multiplied by q to adjust for the value of human capital over wealth invested in financial assets. Newey-West (1989) standard errors are reported in the parentheses. *,** and *** denote significance at 10%, 5% and 1% respectively.
Table 5. Hedging Demand of Industry-Specific Human Capital - Germany
(1) (2) (3) (4) (5) (6) (7) Sector S&P500 (US) DAX (Germany) FTSE100 (UK) OMXH (Finland) CAC40 (France) Nikkei225 (Japan) SSE Composite (China) Construction -0.0693* -0.0494 0.0304 0.0112 0.0953* -0.0079 0.0139 (0.0380) (0.0448) (0.0310) (0.0153) (0.0522) (0.0223) (0.00949) Finance -0.0322 -0.0210 0.0219 -0.0087 0.0249 0.0201 0.0008 (0.0215) (0.0264) (0.0228) (0.00612) (0.0314) (0.0134) (0.00695) Manufacturing -0.0727** -0.0105 0.0120 -0.0033 0.0604 -0.0015 0.0109 (0.0299) (0.0308) (0.0346) (0.0125) (0.0461) (0.0186) (0.00909) Mining -0.0165 -0.0259 -0.0029 0.0038 0.0797 -0.0119 -0.0080 (0.0345) (0.0469) (0.0369) (0.0163) (0.0603) (0.0274) (0.00968) Retail Trade -0.105* -0.0104 0.0025 0.0292 0.0114 0.0230 0.0095 (0.0571) (0.0549) (0.0430) (0.0284) (0.0716) (0.0356) (0.0136) Service -0.0268 -0.0013 0.0182 -0.0042 0.0029 0.0264 -0.0028 (0.0264) (0.0264) (0.0209) (0.0179) (0.0302) (0.0169) (0.00511) Transport -0.0301 -0.0785*** 0.0088 -0.0072 0.117** 0.0247 0.0097 (0.0367) (0.0284) (0.0238) (0.0111) (0.0365) (0.0214) (0.00661) Wholesale Trade -0.0300 -0.0655 0.0353 -0.0272 0.122*** 0.0326 0.0025 (0.0586) (0.0443) (0.0332) (0.0177) (0.0557) (0.0329) (0.0101) Total -0.0478** -0.0328 0.0158 -0.0008 0.0642** 0.0132 0.0046 (0.0211) (0.0207) (0.0176) (0.00566) (0.0289) (0.0145) (0.00499)
This table reports the estimated coefficients of OLS regressions of quarterly growth rates in monthly per person labor income on equity returns of large country indices and a constant. The coefficients represent adjustment weights to optimal portfolios for an investor working in the respective industry. Data on equity indices returns is retrieved from DATASTREAM. Equity returns are calculated as quarterly compounded returns from 2000Q1 to 2017Q4. Human capital returns are calculated as the quarterly growth rate of monthly per-worker labor income. These adjustment weights are multiplied by q to adjust for the value of human capital over wealth invested in financial assets. Newey-West (1989) standard errors are reported in the parentheses. *,** and *** denote significance at 10%, 5% and 1% respectively.
