Issuu

humanidades

tirant

humanidades

978-84-18802-42-3

9 788418 802423

Antonio Moreno-Sandoval (Coordinator)

FINANCIAL NARRATIVE PROCESSING IN SPANISH

Textual Analytics applied to Financial Narratives is a research field with multiple perspectives, as this volume shows. Economists and financial experts can base their economic analyses not only on quantitative disclosures but also on the communication tools employed by managers to persuade their investors. From the NLP perspective, financial texts provide a suitable domain to apply techniques successfully used in information extraction, summarising, classification, and sentiment analysis. Applied linguistics also benefits from the exponential availability of electronic texts to carry out terminological studies and, in general, to better understand financial language in its lexical, grammatical, and discursive aspects. This set of interdisciplinary studies are known as Financial Narrative Processing (FNP). Until recently, FNP was restricted to documents in English, the international language of finance. Financial Narrative Processing in Spanish is the first collection of essays devoted to Spanish Financial Language (SFL). It brings together six pieces that provide an account of the current research in the field from a multidisciplinary perspective.

tecnología, traducción y cultura

FINANCIAL NARRATIVE PROCESSING IN SPANISH Antonio Moreno-Sandoval (Coordinator)

tirant

humanidades tecnología, traducción y cultura

FINANCIAL NARRATIVE PROCESSING IN SPANISH

COMITÉ CIENTÍFICO DE LA EDITORIAL TIRANT HUMANIDADES Manuel Asensi Pérez

Catedrático de Teoría de la Literatura y de la Literatura Comparada Universitat de València

Ramón Cotarelo

Catedrático de Ciencia Política y de la Administración de la Facultad de Ciencias Políticas y Sociología de la Universidad Nacional de Educación a Distancia

Mª Teresa Echenique Elizondo Catedrática de Lengua Española Universitat de València

Juan Manuel Fernández Soria

Catedrático de Teoría e Historia de la Educación Universitat de València

Pablo Oñate Rubalcaba

Catedrático de Ciencia Política y de la Administración Universitat de València

Joan Romero

Catedrático de Geografía Humana Universitat de València

Juan José Tamayo

Director de la Cátedra de Teología y Ciencias de las Religiones Universidad Carlos III de Madrid

Procedimiento de selección de originales, ver página web: www.tirant.net/index.php/editorial/procedimiento-de-seleccion-de-originales

FINANCIAL NARRATIVE PROCESSING IN SPANISH

ANTONIO MORENO-SANDOVAL (Coordinator)

tirant humanidades Valencia, 2021

Copyright ® 2021 Todos los derechos reservados. Ni la totalidad ni parte de este libro puede reproducirse o transmitirse por ningún procedimiento electrónico o mecánico, incluyendo fotocopia, grabación magnética, o cualquier almacenamiento de información y sistema de recuperación sin permiso escrito de los autores y del editor. En caso de erratas y actualizaciones, la Editorial Tirant lo Blanch publicará la pertinente corrección en la página web www.tirant.com. Esta publicación es responsabilidad exclusiva de sus autores. La Comisión no es responsables del uso que pueda hacerse de la información aquí difundida.

Este libro ha sido co-editado y financiado con un proyecto de la convocatoria 2017 del programa PROYECTOS I+D+I RETOS (Ref. TIN2017-89351-R) Directores de la colección: MIGUEL ÁNGEL CANDEL MORA ANA BELÉN MARTÍNEZ LÓPEZ

© TIRANT HUMANIDADES EDITA: TIRANT HUMANIDADES C/ Artes Gráficas, 14 - 46010 - Valencia TELFS.: 96/361 00 48 - 50 FAX: 96/369 41 51 Email: tlb@tirant.com www.tirant.com Librería virtual: www.tirant.es ISBN: 978-84-18802-43-0 Si tiene alguna queja o sugerencia, envíenos un mail a: atencioncliente@tirant.com. En caso de no ser atendida su sugerencia, por favor, lea en www.tirant.net/index.php/empresa/politicas-de-empresa nuestro procedimiento de quejas. Responsabilidad Social Corporativa: http://www.tirant.net/Docs/RSCTirant.pdf

Listado de autores Ahmed AbuRaed Najah-Imane Bentabet Blanca Carbajo-Coronado Mahmoud El-Haj Ismail El Maarouf Ana García Toro George Giannakopoulos Ana Gisbert Clemente José María Guirao Miras José Antonio Jiménez Millán Estelle Labidurie Marina Litvak Dominique Mariko Antonio Moreno-Sandoval Paul Rayson Chelo Vargas-Sierra Nadhem Zmandar

Index Preface............................................................................................... 11 Antonio Moreno-Sandoval

Financial Narratives.......................................................................... 15 Ana Gisbert Clemente

Review of the State of the Art in Financial Narrative Processing....... 51 Mahmoud El-Haj Paul Rayson Ismail El Maarouf Najah-Imane Bentabet Dominique Mariko Estelle Labidurie Marina Litvak George Giannakopoulos Ahmed AbuRaed Nadhem Zmandar

Anglicisms in a Financial Corpus: Exploiting Resources for Terminological Retrieval and Analysis............................................................ 99 Chelo Vargas-Sierra Blanca Carbajo-Coronado

Discourse Markers in Financial Narrative: the Case of the Annual Reports and Letters to Shareholders.................................................. 135 Ana García Toro Antonio Moreno-Sandoval

Machine Learning Models for Classifying Spanish Beaters and NonBeaters Financial Reports.................................................................. 179 Mahmoud El-Haj Antonio Moreno-Sandoval José A. Jiménez Millán

Tools for Processing FINT-ESP Resources.......................................... 199 Antonio Moreno-Sandoval José María Guirao Miras José Antonio Jiménez Millán

References.......................................................................................... 211

PREFACE ANTONIO MORENO-SANDOVAL Universidad Autónoma de Madrid

Textual analytics applied to financial narratives is a research field with multiple perspectives, as seen in this volume. On the one hand, economists and financial experts can base their analyses not only on quantitative disclosures but also on the communication instruments employed by managers with their investors. From the NLP field, financial texts provide a domain in which to apply techniques successfully used in information extraction, summarising, classification and sentiment analysis. Applied linguistics also benefits from the exponential increase of texts in electronic format for terminological studies and, in general, to better understand financial language in its lexical, grammatical and discursive aspects. We call this set of interdisciplinary studies Financial Narrative Processing (FNP). Until recently, FNP was restricted to documents in English, the international language of finance. In the last five years, things have changed dramatically. French, Spanish, Portuguese, Chinese, Japanese, Korean or Arabic are some of the languages that have received computational treatment in this domain. It is easy to predict that most of the languages in which financial texts are produced will be incorporated into this list in the coming years. AI and especially Machine Learning based on deep neural networks have been a revolution in NLP. Pre-trained language models are an avenue of research that has just started in NLP and will surely give such remarkable results as we see in Machine Translation or Speech Processing. The papers published in this volume are a selection of results from the FinT-esp project (Financial text analytics in Spanish: Tools and language resources) funded by the Spanish Ministry of Economy, Industry and Competitiveness (Ref: TIN2017-89351-R) and developed between January 2018 and June 2021. Researchers from the Universidad Autónoma de Madrid (coordinating centre),

Antonio Moreno-Sandoval

the Universities of Lancaster, Alicante, Granada and Cádiz, and the Instituto de Ingeniería del Conocimiento (Madrid) have participated. Chapter 1 presents the perspective of the finance and accounting researcher. Gisbert defends the multidisciplinary nature of FNP to achieve the best results. After reviewing the typology of financial narratives, she describes the relevant topics in Annual Reports, the instrument par excellence of financial communication. The different theories in narrative reporting introduce us to impression management, i.e., the various strategies that managers use when communicating with their stakeholders. Gisbert describes with real examples in Spanish seven different strategies, ranging from rhetorical, thematic and visual manipulation to the attribution of performance and the selective use of earnings indicators. Gisbert then sets out the possible factors that explain these narrative strategies for masking poor performance. She ends the chapter with an account of opportunities for using NLP as new tools for understanding financial narratives. Accounting academics need the support of computer scientists and linguists to measure the quality of financial reports further. The ten authors of Chapter 2 present an overview of FNP. They are responsible for the different editions of the Financial Narrative Processing - Financial Narrative Summarisation Workshop Series since 2018. These workshops show a growing interest in applying computational approaches for extracting, summarising, and analysing both qualitative and quantitative financial data. El-Haj et al. describe datasets and language resources that are publicly available for conducting FNP research. The chapter illustrates in detail the competitions and results of dozens of teams and systems applied to tasks such as automatic extraction of structure from financial documents (FinTOC), causality detection in financial texts (FinCausal) and the financial narrative summarisation from annual reports (FNS). Chapter 3 and Chapter 4 are devoted to linguistic analyses of the FinT-esp corpus of financial documents. Vargas and Carbajo present a corpus-based methodology to extract financial terminology in Spanish, specifically neologisms and Anglicisms. The authors provide a classification of three types of terminological Anglicisms: pure, hybrid combinations, and calques. This paper describes the new lexical structures created in Spanish based on hybrid combinations or

Preface

English calques. Multiple examples illustrate this lexical richness. The inventory of terms is an initial step for annotating key concepts and creating an automatic extractor and classifier of financial terms. Chapter 4 by García-Toro and Moreno-Sandoval addresses the analysis and distribution of Discourse Markers (DM) in a financial corpus composed of company presidents’ letters to shareholders. A quantitative methodology is applied, based on the standardised frequency of DMs to perform a contrastive analysis against the general register represented by CORPES XXI. The authors follow the classification of Portolés (2011) to group the different types of DM, consisting of a total of 33 discursive particles. Connectors are the most frequent in the studied domain, at 60% of the total, among which stand out además (also) and así como (as well as). An analysis of the distribution of DMs in the argumentation of companies experiencing profit and loss is also included. The only significant difference is the frequent use of counter-argumentative connectors in the reports of companies whose financial results have been lacking. Chapter 5 and Chapter 6 are devoted to computational tools applied to the FinT-esp corpus. El-Haj, Moreno-Sandoval and JiménezMillán address different methods to classify the narratives of company managers into two groups: Beaters and Non-beaters. Due to the small size of the corpus (about 400 texts with an average length of 1100 words) and, above all, the imbalance between the two classes (Gain 85% vs. Loss 15%), the classification task has focused on two strategies. The first one consists of using classical models and deep learning models without Transformers. To overcome the imbalance in the classes, the authors have applied lemmatisation and stop-word elimination (to all documents) and oversampling to the texts of loss-making companies. The second approach is the use of Transformers models in Spanish (BERT and spaCy) with the idea of testing the most advanced DL models to date. The more classical models have obtained better results, but thanks to oversampling. The Transformers models have remained below 0.90 in accuracy. The authors also explore possible explanations for error in automatic classification. In particular, the use of positive words to mitigate the impact of negative financial results.

Antonio Moreno-Sandoval

In the final chapter, Moreno-Sandoval, Guirao, and Jiménez-Millán describe two tools developed within the FinT-esp project, mainly for terminological purposes. The first of these is used to access the compiled corpora contents. The second tool, WikiCorporaComposer, has been designed to enable the creation of corpora tailored to users’ needs. The authors have used this tool to generate more sources to elaborate draw up lists of financial terms and their equivalents in Spanish and English. We hope that these papers will attract other researchers interested in the opportunities and challenges in this exciting field of study.

FINANCIAL NARRATIVES ANA GISBERT CLEMENTE ana.gisbert@uam.es Universidad Autónoma de Madrid

1. THE NEED FOR ACCOUNTING INFORMATION AND THE EVOLUTION OF FINANCIAL NARRATIVES Language plays a major role in understanding people’s perceptions of the world and the environment they live in. “Language mirrors the mental processes” (Chomsky, 1972). Therefore, if we can analyse their language, we can gain significant insights into how people perceive and understand their world (Kabanoff, 1996). The language of accounting and, more specifically, narratives that support accounting figures, can be regarded as the language used within business organisations and in which managers communicate with users. Gaining insights on this type of language helps us to understand how a company is performing, its future perspectives and the managers’ professional opinion on the overall business outcomes. The concept of financial narratives embraces all the different types of documents prepared by managers to provide users with reliable information relevant to economic decision-making. Accounting and financial information are, above all, an information system representing real business scenarios that aim to help users assess company performance. Together with traditional quantitative accounting figures, in recent decades there has been an exponential increase in the set of corporate narrative disclosures containing additional detailed explanations for the standard numeric financial data that accounting researchers have been studying for many years (Li, 2010). As Jensen and Mecklin (1976) explain, the ownership of most corporations is structured within the ‘agency relationship’ framework, where “the principals (shareholders) engage with another person (the agent) to perform a given service on their behalf, delegating decision-making authority in the agent”. This agency relationship is formalized in contracts between the parties involved (managers and shareholders), leading to assumption of a set of costs known as ‘agen-

Ana Gisbert Clemente

cy costs’ arising from the requirement for establishing controls to ensure that the interests of the “agent” and the “principal” are aligned and, therefore, that the “agent” acts in the best interest of the “principal”. A clear example of these controls are bonus contracts, incentive compensation contractual agreements between managers and shareholders in which the latter will receive an economic incentive (bonus) if the performance of the company reaches a specific and pre-established target (i.e. achieving a specific earnings increase with respect to the previous year). Monitoring controls of this type are based on financial performance metrics, requiring the establishment of an information system that enables managers to provide shareholders with periodic financial information. This information must be relevant and reliable to be useful in decision making, requiring the company to incorporate additional controls, such as financial statement audits. The aim of these controls in the agency relationship is twofold. On the one hand, to align the interests of managers and shareholders, avoiding any potential managerial malpractice. On the other hand, to guarantee the reliability and quality of the financial information that managers periodically provide to shareholders. As key capital providers, shareholders need to assess performance regularly to gauge the profitability of the business they have invested in. Periodically reporting financial information contributes to reducing the asymmetry of information existing between managers and capital providers, thus lowering agency relationship costs. However, the financial information system is not only useful to shareholders. Other capital providers such as financial institutions need to periodically assess the risks and growth potential of the business to take decisions relating to the maintenance of the firm’s financing resources. Additionally, beyond shareholders and creditors, other economic agents show increasing interest in companies’ financial information to assess overall financial and non-financial performance and take relevant economic decisions accordingly. Hence, financial information must be publicly available so that any economic agent can access and use it in assessing the company’s financial performance. Table 1 provides a list of the different economic agents, known as “stakeholders”. The analysis perspective and the type of information they demand are different across users, requiring an increasing variety

Financial Narratives

of information disclosure. Analysts, industry peers, investors or suppliers may be more interested in the financial performance, growth potential or the multiple risks of the company, while other users such as customers, government, and society at large may focus on non-financial performance issues such as gender equality policies or social and environmental performance. Type of user

Interest and use of financial narratives

Shareholders

They use the financial information to take their investment decisions.

Customers

They use the financial information to take consumer decisions (i.e. interest in products from companies with a high social and environmental performance).

Industry competitors

They use the financial information to assess the performance of their competitors and take decisions regarding current and future strategy.

Suppliers

They use the financial information to assess the company’s operating risks and decide on financing schemes for new acquisitions of merchandise or raw materials.

Creditors

They use the financial information to assess the company’s risks and decide on providing new financing resources.

Financial (Sell -Side)

Analysts

They use the financial information to assess the performance of the company, prepare their reports and release their investment recommendations.

Financial (Buy -Side)

Analysts

They use the different sources of financial information to assess performance and take decisions about the composition of their investment funds.

Governments & Regulatory Bodies

They use the different sources of financial information to assess the performance of the company and take regulatory actions in issues such as new regulatory tariffs, taxes or a potential financial rescue.

Labour Unions

They use the different sources of financial information to assess the performance of the company and protect the overall interests of employees.

Society, Media.

They assess the different sources of financial information to assess financial and non-financial performance and therefore construct an image of the company.

Table 1: Users of the company’s different sources of financial information

Ana Gisbert Clemente

This variety of interests has led to a significant change in the way companies have shaped their publicly available financial information. Regardless of the significant preparation costs of this information, financial reporting has gone beyond the traditional financial statements and has gained complexity over the last decades. The demand for additional accountability among economic agents has gained significance. Responding to users’ multiple and diverse demands, financial narratives have experienced an upsurge during the last two decades, becoming a central component of the company’s overall financial reporting package (Beattie, 2014). In recent decades, financial communication has revolved around a complex strategy planned to shape stakeholders’ perceptions of the organization (Laskin, 2018). The financial reporting filings traditionally required by the securities market regulators (i.e. annual financial statements, 10-K, 10-Q) have given way to other reporting layouts, extensive documents with appealing narratives, figures and pictures where financial performance analyses coexist with other relevant non-financial matters. As explained in Beattie (2014), Anthony Hopwood1 in 1996 already highlighted the transformation of annual reports from minimalist to “flamboyant” documents making creative use of text and images. No matter in which country the company is listed, the extensive use of these attractive documents and the increase in financial narratives to support and explain the “accounting numbers” is a widespread phenomenon throughout multinationals. In fact, some authors claim to be facing an information overload context (Stolowy and Paugam, 2018). However, “Narrative Financial Reporting” goes beyond the simple idea of qualitative explanations of financial data. As explained by ElHaj et al. (2021) in chapter 2, it implies a “variety of data sources,

Anthony Hopwood is one of the most outstanding academics in the history of accounting discipline. His research was highly influential, considering accounting not just as a technical discipline, but as an organizational and social practice with high impact on economics and organizations. As the founder of the European Accounting Association, he contributed to the creation and development of the European Institute of Advanced Studies in Management and the leading European accounting journal European Accounting Review (Carmona, 2010).

Financial Narratives

modalities and formats”. Erkens et al. (2015) define it as “the disclosure provided to outsiders of the organization on dimensions of performance other than the traditional assessment of financial performance from shareholders and debt-holders’ viewpoint”. This definition includes items related to social and environmental reporting —CSR— and intellectual capital disclosed outside the traditional financial statements. Taking Spain as an example, listed companies are required to prepare a regulated set of corporate information on the preceding year, known as “Audited Financial Statements”. Spanish Corporate Law requires all listed companies to submit this document to the Spanish National Securities Market Commission2 (CNMV) and post it on corporate websites to be publicly available for all users. These Financial Statements are a regulated document with a fixed structured format comprising two clear-cut sections. The first of these is fully quantitative and incorporates the following documents: the balance sheet or statement of financial position, the income statement, the comprehensive income statement, the cash flow statement and the statement of changes in equity. The second section, known as the “notes to the financial statements”, is mainly qualitative and offers detailed explanations for the quantitative financial data gathered in the first part. Despite their qualitative nature, these “notes” are highly structured and standardized, requiring the use of technical accounting narratives that limit room for discretion in the choice of discourse. While standardization lessens flexibility regarding the type of information to be included in the Financial Statements, it enhances comparability of information across companies, facilitating the analytical role of international investors and other users. Such standardization enables the implementation of automatic digital reporting tools such as XBRL (eXtensible Business Reporting Language)3 that can be used to com

The Spanish Securities Market Regulator is the CNMV (Comisión Nacional del Mercado de Valores). XBRL is an international standard to facilitate digital automatic reporting that applies tags to data. It has been developed and refined in the last decade and is currently accepted as an automatic reporting mechanism in more than 60 countries. Such is the case of the US, where the SEC requires the financial statements of listed companies to be filed in XBRL format. Similarly, in the EU the European Single Electronic Format (ESEF) Regulation requires listed companies to mark

Ana Gisbert Clemente

ply easily and automatically with the regulatory information requirements of securities market regulators and tax authorities. Additional reports addressing company ownership and governance structure (Corporate Governance Report) or the remuneration of the board of directors (Directors’ Remuneration Report) have also become compulsory in most countries in the last decade, providing information in a standardized format on the company’s remuneration policy and the contractual conditions of its executive directors. Together with these regulated and structured reports, most companies in Europe prepare a much more extensive, appealing, detailed and comprehensive document known as the Annual Report, Global Report or, more recently, the Integrated Report. In countries such as Spain, this is not required by Corporate Law. Nevertheless, following the international trend, most Spanish listed companies prepare these extended reports providing both quantitative and qualitative disclosures, with attractive graphs and figures that contain information about the financial and non-financial performance of the company. In the last few years, around 60% of listed companies in the Madrid Stock Exchange have prepared an Annual, Global or Integrated Report. These reports are mainly prepared by big multinationals that can exploit their existing information systems and communication departments to prepare these non-standardized documents at lower marginal costs compared to smaller companies. These Annual Reports are not comparable across firms. Formats and content vary in different countries, and they even present significant differences among companies within the same country. There is a wide discretion on the choice of topics and scope of data analysis. This renders these reports an essential source of financial narratives. Despite the variability in content, there is a set of core topics that are always reported in response to stakeholders’ demands, namely corporate social responsibility, corporate governance, financial per-

the primary financial statements with XBRL tags from 2020 onwards. XBRL reporting is a structured reporting format that facilitates the comparability of financial information across companies. More information can be obtained from: www.xbrl.org.