Digital agendas in the insurance industry: the importance of comprehensive approaches^†

Abstract

With a growing awareness of the potential of innovation provided by digital technology, insurance companies have increasingly adopted digital agendas in their business activities. Our paper studies the relationship between the expression of a digital agenda in annual reports and the business performance of 41 publicly-traded European insurance companies for the time period from 2007 to 2017. Our findings show a positive relationship, which is particularly strong in cases where companies take a comprehensive approach by addressing digital technology both in the context of internal activities within their own organisation and external activities in connection with customers and business partners.

Appendix

A.1: Treatment-effects model

The treatment-effects model is given by the following two regression equations that are simultaneously estimated via maximum likelihood. We assume $d_{i,t}^{k} = d_{i,t}^{c20,ei,binary}$ in our base case.⁹ The regression equation (“Q Equation”) is given by

$$ Q_{i,t} = x_{i,t} \beta + d_{i,t}^{k} \delta + \varepsilon_{i,t} $$

(3)

and the selection equation (“Digital Equation”) is defined as

$$ {}^{*}d_{i,t}^{k} = z_{i,t} \gamma + u_{i,t} , $$

(4)

where

$$ d_{i,t}^{k} = \left\{ {\begin{array}{*{20}c} 1 & {{\text{if }}{}^{*}d_{i,t}^{k} > 0} \\ 0 & {\text{otherwise}} \\ \end{array} } \right. $$

and error terms $\varepsilon_{i,t}$ and u_i,t that are assumed to be normally distributed with a mean vector of zero, variances of $\sigma_{\varepsilon }$ and 1, and a covariance of $\rho$ (see, e.g. Maddala 1983; Guo and Fraser 2010; Hoyt and Liebenberg 2011; Bohnert et al. 2018).

A.2: Pseudocode

Repeat for each PDF document, i.e. annual report

• Rename and assign unique identifier (ID) to each PDF document (ID.pdf)

• Prepare document for text extraction, i.e. remove access restrictions (via Ghostscript)

• Extract plain text (via Xpdf) and create a text document with identical ID as the PDF document (ID.txt)

Repeat for each plain text document (for all files ID.txt)

• Create corresponding new (empty) text document for concordances (ID_conc.txt)

• Translate all characters to lower case characters

• Wrap all characters that are not alphabetic characters or digits with one space character before and after

• Replace (multiple) white space characters (including tab keys) and newline with one space character

• Identify occurrences of words containing the strings “digita” or “digiti”

• Repeat for each occurrence

  • Extract corresponding keyword string and certain number of words (string wrapped by space characters) before and after the corresponding keyword, i.e. extract concordance of a given length, e.g. 20 words before and 20 words after the keyword in case of c20

  • Add concordance line to ID_conc.txt

  • Repeat for each concordance document (for all files ID_conc.txt)

    • Remove all characters besides alphabetic characters

    • Remove common words from common.list¹⁰

    • Transform all words into word stems (via Porter’s word stemming algorithm)

    • Set variable d_(i,t)^(c20,e,binary) to 1, if at least one of the following word stems can be found: “channel”, “client”, “custom”, “distribut”, “market”, “online”, “product”, “sale”, “service”

    • set variable to 0 otherwise

    • Set variable d_(i,t)^(c20,i,binary) to 1, if at least one of the following word stems can be found: “board”, “employe”, “group”, “manag”, “model”

    • set variable to 0 otherwise

    • Set variable d_(i,t)^(c20,ei,binary) to 1, if d_(i,t)^(c20,e,binary) == 1 and d_(i,t)^(c20,i,binary) == 1

    • set variable to 0 otherwise