CONCEPTO URBANISTICO

This subsection presents the methodology used in the industry relatedness analysis. First, the calculation of firm-, industry- and market-level return on assets is presented. Then the fixed effects panel regression used to study correlation of fundamentals along the supply chain is presented. The methodology used is consistent with Menzly and Ozbas (2010).

As mentioned earlier, two important assumptions are required to obtain return cross-predictability in a limited-information model: (i) firms in different industries or market segments have correlated fundamentals and (ii) markets are informationally segmented as informed investors, to some degree, specialize along these boundaries in their information-gathering activities (Menzly and Ozbas, 2010). Evidence on the latter assumption is provided in the literature review in subsection 2.2.2.1 and providing further proof of this assumption is outside the scope of this study. However,

empirical evidence on assumption (i) is provided in order to verify the empirical design of this paper. More specifically, this is confirmation is performed to ensure the validity of the consolidated Eurostat input-output tables in describing industry relatedness and to address the possibility that border effects may cause industry fundamentals to be insufficiently correlated (see subsection 3.1). In the particular empirical setting used in this study, the correlated fundamentals assumption means that firms in a given industry need to have correlated fundamentals with firms in their supplier and customer industries. To test whether firms along the supply chain have correlated fundamentals, firm-, industry- and market-level measures of profitability are constructed and used in a fixed- effects panel regression. The calculation of firm-level return on assets is shown in Equation (4) below.

, = , _, , (4)

where, _, is the return on assets of firm j in year t, _, is the operating income of firm j in year

t, _, is the depreciation, depletion and amortization of firm j in year t and _, is the total assets

of firm j in year t. Industry- and market-level ROA are calculated by aggregating the above firm- level ROAs with a portfolio approach using firm assets as portfolio weights. The industry- and market-level profitability calculation are shown below in Equations (5) and (6) for both market level and industry level ROAs, respectively.

= ,

, , (5)

where, is the aggregated market-level ROA for year t, nt is the total number of firms in

year t, _, is the total assets of firm j in year t and _, is the firm-level return on assets for firm

j from equation (4).

, = ,

,

, ,

, ₍₆₎

where, _, is the aggregated industry-level ROA in year t for industry i, ni,t is the number of

firms in industry i in year t, _, is the total assets for firm j belonging to industry i in year t and

, is the firm-level return on assets for firm j from Equation (4). Industry-level ROAs are used

to further calculate supplier and customer industry ROAs for each industry i. This calculation is done by weighting the industry-level ROAs of supplier and customer industries with the flow of goods and services to and from the industries in question. This approach is similar to the calculation

of supplier and customer industries returns in Equations (2) and (3). The ROA calculation for supplier and customer industries is shown in Equation (7) below.

, ( ) = , _, , (7)

where, _, ( ) is the aggregated return to assets in year t for industry i’s supplier

(customer) industries weighted by the flow of goods and services into (out of) industry i, is the

number of industry i’s supplier (customer) industries excluding industry i, _, is the amount of

industry i’s purchases from (sales to) industry k and _, is the return on assets of industry k in

year t from Equation (6). The ROA figures received from Equations (4) to (7) are used in a fixed effects panel regression that is presented in Equation (8) below. Two specifications of Equation (8) are performed, one with firm-level ROA as dependent variable and another with industry-level ROA as dependent variable.

( ), = + + _{( ),} + ( ), (8)

+ ( ),

where, the dependent variable _{( ),} is the return on assets of firm j (industry i) in year t

depending on the specification, is the coefficient on market-level ROA, is the

contemporaneous market-level ROA from equation (5), is the coefficient on supplier-

industry ROA for firm j (industry i), _{( ),} is the supplier-industry ROA for firm j (industry

i) in year t from equation (7), is the coefficient on customer-industry ROA for firm j

(industry i), _{( ),} is the customer-industry ROA for firm j (industry i) in year t from

Equation (7) and _{( ),} is the error term from the regression in year t.

As mentioned in section 4.4, there is considerable variation in the annual ROA figures across sample countries. Therefore, a possible concern is that country-specific differences in average profitability are driving the results, particularly in the firm-level ROA panel regressions where the independent variables are country-specific. In order to address the possibility that country-specific factors such as varying economic conditions are confounding the regression results, the firm-level ROAs are adjusted for country-specific differences in overall profitability in the firm-level specification of Equation (8). The indexation formula used in this study is shown in Equation (9).

where, _, is the indexed return on assets of firm j in year t, _, is the return on assets

of firm j in year t and _, is the aggregated market-level ROA in year t for country c in

which firm j is listed. Appendix 3 contains the annual country-level return on assets used in Equation (9).

In panel data sets, such as the one studied here, the residuals may be correlated across observations which can cause OLS standard errors to be biased (Petersen, 2009). More specifically, the firm effect and/or time effect may introduce bias in the standard errors which, in turn, may lead to

incorrect t-statistics and unjustified rejection (acceptance) of the null hypothesis.52 Following

Petersen (2009), the regression standard errors are adjusted for both firm and time effects, to address the possibility that correlated residuals are biasing the standard errors and thus, the t- statistics. Firm effect is taken into account parametrically by using fixed effects regression with fixed firm effects. The time effect is addressed by adjusting the standard errors for clustering by year.53