Tax Loss Carryforward Disclosure

Abstract

This study examines firms’ voluntary disclosure of tax loss carryforward (TLCF) information. We measure the content and presentation of TLCF information by a disclosure score based on hand collected data from annual reports. Our identification strategy employs new proprietary data to control for overall reporting quality. We argue and find that uncertainty about the usability of TLCF is a key driver of voluntary TLCF disclosure. The disclosure score is on average 3.86 points higher for firms with a strong loss history vis-à-vis firms without such a loss history. This positive association is economically meaningful, since the sample mean for the disclosure score is 8.88 (median 7.5). We also find that the content and type of disclosure vary systematically with the signal of uncertainty. In instances of historic uncertainty, firms exhibit increased reporting on the reasons and mechanisms behind changes in TLCF. Conversely, in the case of future uncertainty, they increase disclosure related to valuation allowance information. Our study provides detailed and unique insights into TLCF disclosure, suggesting that managers enrich the information environment with voluntary disclosure that caters to expected investors’ needs.

1 Introduction

Information about tax loss carryforwards (TLCF) is complex and difficult to comprehend, but their considerable magnitude renders them important for estimating future tax savings. For example, the mean (median) TLCF for large German firms between 2005 and 2016 is 10.4 (3.7)% of total assets.¹ The COVID-19 pandemic and the resulting financial problems for certain businesses have likely increased the amount of TLCF for many firms and consequently, their economic relevance. To assess the effect of TLCF on firms’ future tax payments, financial statement users must understand how firms can use their TLCF for offsetting future profits (TLCF usability). We investigate whether firms voluntarily provide TLCF information when the TLCF usability is uncertain.

Research on TLCF-related balance sheet information, e.g., deferred tax assets or valuation allowances, finds that this information is a useful signal for future performance (Dhaliwal et al. 2013; Kumar and Visvanathan 2003), tax payments (Flagmeier 2022), and may indicate earnings management (Frank and Rego 2006; Herbohn et al. 2010, 2016; Schrand and Wong 2003). Accordingly, the literature on deferred taxes mainly provides evidence on the value relevance of these positions (e.g., Amir and Sougiannis 1999; Chang et al. 2009). Importantly, institutional circumstances and the firm’s economic environment may render TLCF-related balance sheet information less useful (Badenhorst and Ferreira 2016; Flagmeier 2022; Hanna et al. 2019). In particular, investors do not seem to rely on recognized deferred tax assets under International Financial Reporting Standards (IFRS) in a German setting (Chludek 2011; Flagmeier 2022) and the TLCF items seem not to be useful to predict future performance in such settings (Dreher et al. 2024). In such accounting regimes firms may complement financial statements with voluntary guidance in the tax footnote as an alternative to communicate TLCF usability.² Thus, we explore how firms voluntarily cater to investors’ TLCF information needs.

We predict that firms provide additional TLCF information when the future usability and hence the value of TLCF are uncertain. Investors perceive TLCF as valuable when they can be offset against profits in the near future (McGuire et al. 2016). Unusable TLCF indicate unfavorable earnings expectations or even further losses in the future (Amir and Sougiannis 1999). It is therefore essential for investors to assess whether TLCF can offset future taxable income. Uncertainty about the usability of TLCF can create (or increase) capital market pressures, for example affecting the liquidity of a firm’s shares and hence the cost of capital.³ To mitigate this threat, firms can voluntarily provide TLCF information. Anecdotal evidence from interviews with top managers of German corporations supports the notion that both recognition of deferred tax assets for TLCF and respective disclosures are a discussion point and a strategic choice in periods of high economic uncertainty. This also corresponds with interviews in Flagmeier et al. (2023) who document that tax disclosure decisions are typically made after business or recognition decisions.

The mechanism of using voluntary disclosure to reduce information asymmetry is well established in the accounting literature (e.g., Guay et al. 2016; Verrecchia 2001). However, TLCF is an economically important tax item with some unique features. First, unlike many financial accounting items, TLCF provides not only information about past events but also forward-looking information. Information in TLCF disclosure can go beyond management’s earnings forecasts as, for example, details about TLCF usability reveal expectations about future taxable instead of financial income. This information aids in assessing (potential) future tax savings and can be particularly informative for investors. Second, the audience for tax information is generally very broad. In addition to outside equity investors, analysts, tax authorities, legislators and the general public are interested in a firm’s tax details, potentially creating additional disclosure costs (e.g., Bozanic et al. 2017; Flagmeier et al. 2023; Müller et al. 2020; Robinson and Schmidt 2013). Third, gathering and editing the relevant TLCF information is costly, as it has to be derived from single entity tax statements (e.g., PricewaterhouseCoopers [PwC] 2012; Smith Raedy et al. 2011). Fourth, due to the complex nature of TLCF information, high processing costs occur, which may distort the investor perception of such information (Blankespoor et al. 2020). In sum, firms have incentives to disclose more TLCF information in the case of uncertain usability, but they also face disclosure costs.

To examine whether firms disclose more TLCF information if its usability is uncertain, we derive two signals of usability uncertainty from IAS 12: historic uncertainty resulting from a firm’s loss history and future uncertainty based on expected future earnings (and hence loss-offsetting options). We create a disclosure score to measure TLCF disclosures. The score captures all information about TLCF that is disclosed voluntarily in the tax footnotes, i.e., beyond the mandatory requirements under IAS 12. Each TLCF disclosure is evaluated regarding the information content and the type of information disclosed (e.g., qualitative vs. quantitative), including the way in which it is presented. This comprehensive approach allows us to distinguish between different disclosure properties and to draw granular inferences about firms’ disclosure behavior. We apply the score to a sample of large German public firms between 2005 (after mandatory adoption of IFRS) and 2016, and we mainly hand-collect data from their annual reports’ notes on income taxes. The institutional benefits of the German setting are reinforced by an innovative design feature of our study. We control for the general disclosure behavior of a firm based on proprietary data of the German yearly annual report competition of ‘Manager Magazin’. This allows us to analyze the incremental tax reporting disclosure choice over the general disclosure policy of a firm more precisely than past studies.

We find that greater TLCF usability uncertainty is related to the voluntary disclosure of TLCF information. We find, for example, that the disclosure score is on average 3.86 points higher for firms with a strong loss history vis-à-vis firms without such a loss history. This positive association is economically meaningful, since the sample mean for the disclosure score is 8.88 (median 7.5). The finding qualitatively holds for both signals of uncertainty, historic and future, and suggests that firms use their TLCF disclosure to guide users with TLCF information. A placebo test in which we use the annual report disclosure quality instead of the TLCF disclosure score yields mainly insignificant results, suggesting that firms’ disclosure choices are TLCF specific. Our findings are further corroborated in tests with an alternative measure for uncertainty based on tax planning opportunities, derived from McGuire et al. (2016). Moreover, the results are robust to a battery of sensitivity tests, including a two-stage estimation to address endogeneity concerns and an exogenous uncertainty shock.

In additional tests, we break down the disclosure score into subcategories to explore the characteristics (content and type) of disclosure behavior. First, our findings reveal a systematic pattern in disclosure behavior related to content, contingent on the uncertainty signal. In instances of historic uncertainty, firms exhibit increased reporting on the reasons and mechanisms (e.g. effect on current year’s income) behind changes in TLCF. Conversely, in the case of future uncertainty, they increase disclosure related to valuation allowance information.⁴ Further, firms provide additional unsystematic information under both uncertainty signals. Second, firms seem to apply specific disclosure types under uncertainty. Firms provide more detailed information on TLCF items and enhance visibility through the use of tables under both uncertainty signals. In summary, our results suggest that firms adjust their TLCF disclosure based on the prevailing situation, encompassing considerations of both content and ease of processing.

We contribute to the literature in several ways. First, we advance the understanding of firms’ strategic tax disclosure by providing insights into firms’ disclosure behavior for TLCF, an economically important tax item with the unique nature of complex forward-looking information elements. Our analyses support and extend findings of increased voluntary tax-related disclosure in the case of high information asymmetry (Balakrishnan et al. 2019; Chen et al. 2023) and, in a broader context, voluntary disclosure as a response to financial statement complexity (Guay et al. 2016). Second, we add to the research on TLCF-related items. Extending evidence that mandatory items provide signals about future tax payments (Flagmeier 2022), we document that firms complement these mandatory items with additional voluntary guidance. These findings advance the understanding of tax disclosure decisions and tax footnote heterogeneity (e.g., Kvaal and Nobes 2013; Smith Raedy et al. 2011). We document how firms cater to the expected shareholders’ information needs about TLCF, contingent on the uncertainty signal, and thus we connect with Dreher et al. (2024) who investigate the predictive ability of certain disclosed TLCF information. However, our study does not delve into whether accounting for the nuanced heterogeneity in the content and type of voluntary disclosure could improve the prediction of earnings or cash flows. Third, we extend the emerging literature that performs textual analyses of tax information (e.g., Hutchens 2017; Inger et al. 2018) by introducing a disclosure score that captures the content and type of TLCF information. Our disclosure measure allows us to model disclosure decisions more precisely and to distinguish between the content and type of disclosure to overcome limitations of current natural language processing techniques. The complexity of accounting for TLCF provides ample opportunities for voluntary disclosures. At the same time, such information is not easy to collect, present or comprehend, which provides a unique setting with an unclear cost–benefit balance. Hutchens (2017) identifies disclosure characteristics that improve analysts’ understanding of the tax footnote and emphasizes the need for more research examining how disclosure characteristics affect users’ ability to process tax information. Our findings improve the understanding of the content and type of voluntary disclosure (e.g., Chen et al. 2017; Li 2010; Loughran and McDonald 2016). In particular, it appears that firms tailor the substance of TLCF-related voluntary disclosure to meet the anticipated demand for information by providing content that is more readily understandable, such as presenting it in a tabular format. Our findings could offer valuable insights for investors in different jurisdictions that adopt IFRS, particularly those where deferred tax assets are similarly regarded as less value relevant.

2 Setting and Hypothesis Development

2.1 Setting

Firms can communicate private information about TLCF value via the recognition of deferred taxes for TLCF. For the recognition of deferred taxes, both IAS 12 and the US-GAAP Accounting Standards Codification (ASC) 740 require taxable profit in the future.⁵ Hence, the amount of recognized deferred tax assets for TLCF is a signal of management’s earnings expectations. However, Chludek (2011) and Flagmeier (2022) examine German firms and do not find value relevance of deferred tax assets for TLCF. In a similar setting Dreher et al. (2024) show that mandatory TLCF items do not improve predictive ability regarding future performance. These findings suggest that deferred tax assets for TLCF are not informative in certain settings and raise the question whether firms anticipate a stakeholder demand for information and use an alternative channel—the voluntary disclosure of additional, yet complex information in the notes—to reduce information asymmetries. While our setting allows an improved identification due to the potentially high demand for information, mandatory TLCF disclosure requirements may not fully address investors information needs and leave room for firms’ voluntary disclosure also in other settings. The FASB’s Proposed Accounting Standards Update on Income Taxes (FASB 2019), for example, discusses additional disclosure requirements for TLCF to enhance the information environment on this important topic.

2.2 Hypothesis

Theoretical studies that analyze the value relevance of TLCF include De Waegenaere et al. (2003) and Sarkar (2014), both highlighting the ability of TLCF to reduce future tax payments. In line with this notion, empirical studies on US and Australian data mostly find a positive association between newly generated TLCF and stock returns (Amir and Sougiannis 1999; Chang et al. 2009; McGuire et al. 2016). Without offsetting options, the TLCF is worthless and can be interpreted as a signal for further losses in future periods (Amir and Sougiannis 1999). Investors can be assumed to be particularly interested in TLCF information when it is not obvious which of the two scenarios applies, i.e., when there is uncertainty about the usability of the TLCF. Disclosing information that goes beyond mandatory disclosures, for example, reasons for changes in the amount of TLCF or when and why the firm expects to use the TLCF, can be a way for firms to reduce this uncertainty.

Nevertheless, there are at least four reasons why firms might not disclose this information. First, gathering and editing the information is costly. Usually, firms derive tax information from single-entity tax returns, often from different jurisdictions. Aggregating this information at the group level can be a challenging task. According to researchers and practitioners, tax footnotes are very complex and costly to produce (PwC 2012; Smith Raedy et al. 2011). In particular, the recognition of deferred tax assets is a controversial issue under both IFRS and US-GAAP. According to Petree et al. (1995), the recognition of deferred tax assets is probably “the most complex and subjective area of Financial Accounting Standards Board Statement no. 109” (p. 71). Anecdotal evidence highlights the difficulties firms face in estimating the amount of TLCF for the consolidated statement. Deutsche Post DHL states in their annual report for 2011 that a “… refined method for determining unused TLCF was applied for the first time as at the current balance sheet date. The prior-period amounts were adjusted” (p. 184). The adjustment of the prior period’s unused TLCF amounts to 2.4 billion euros, which can be expressed as six percent of the firm’s total assets. This example illustrates that firms need sophisticated methods to determine the amount of TLCF at the group level, implying costs for the creation of the tax footnote.⁶

Second, the provision of detailed tax information can reveal insights into firm performance (Lenter et al. 2003) and help estimate firms’ tax returns (Kvaal and Nobes 2013). These insights can involve proprietary costs (Verrecchia 1983) and costs from stakeholder reactions, e.g. scrutiny of enforcement bodies (Bozanic et al. 2017; Flagmeier et al. 2023). In the TLCF context, sensitive information can for example be details on the TLCF amounts and expiration dates in different subsidiaries. Subsidiary level information is generally associated with proprietary costs as it could be a proxy for tax avoidance behavior if tax-haven activities are revealed (Dyreng et al. 2016). In line with this notion, we find only very few examples of firms disclosing region-specific TLCF information.⁷ Further, Bozanic et al. (2017) find that Internal Revenue Service (IRS) attention to tax positions is positively associated with TLCF, indicating that tax authorities are generally interested in this item.

Third, once certain information is disclosed voluntarily auditors and stakeholders expect such disclosures also in future years. The principle of consistency of accounting choices (IAS 8.13) renders the disclosure costly since any future deviation needs to be substantiated and might raise stakeholders’ concerns. Thus, the requirement to disclose such information continuously may cause preparation and proprietary costs.

Fourth, if investors deem voluntarily disclosed TLCF information as not valuable, particularly in predicting future earnings or cash flows (as indicated by Dreher et al. 2024), the effective demand for such information may diminish. Hence, firms may refrain from responding to uncertainty signals by increasing voluntary disclosure. In fact, there might be a tendency to reduce disclosures overall, as the perceived benefits of providing additional TLCF-related information seem diminished. In sum, capital market pressure can incentivize firms to increase their disclosure in the case of uncertainty, but they also have to consider the costs. We analyze whether firms expect a net benefit and therefore increase their disclosure for uncertain TLCF, and we pose the following hypothesis:

Hypothesis 1.:

The level of voluntary disclosure is positively associated with uncertainty about the usability of TLCF

3 Research Design

3.1 Sample

To investigate the financial statement disclosure of TLCF, we employ a sample of hand-collected data from annual reports’ notes on income taxes. Further firm-specific accounting information is obtained from Refinitiv’s Worldscope and the I/B/E/S database. Our sample comprises financial statements of listed DAX-30 and M‑DAX firms over fiscal years 2005 to 2016 that are prepared in accordance with IFRS. Our sample period ends in 2016 because one of our main control variables (AR_DISCL: a disclosure score of the annual report content quality to control for the overall disclosure quality of a firm; for details, see the Model section) is no longer available after 2016. DAX-30 and M‑DAX include the 80 largest and most liquid firms on the German stock market, based on free float market capitalization and exchange turnover.⁸ Moreover, the information environment (e.g., disclosure requirements, investor relations, visibility, analyst following) is rather homogeneous for firms in these market segments in comparison to other German stock-listed firms. We exclude 17 non-IFRS observations and lose 38 observations due to a lack of availability of annual reports. Furthermore, the sample is reduced by 183 observations with missing data, resulting in a final sample of 79 firms with 723 firm-year observations. Of these 723 observations, we can infer from the collected data that at least 721 firm-years have TLCF.⁹ Our sample selection is detailed in Table 1.

Table 1 Sample Overview (Years 2005–2016)

3.2 Disclosure Level

To investigate the voluntary disclosure level, our measure captures all information that is provided beyond mandatory IAS 12 requirements.¹⁰ To measure disclosure, prior studies sometimes use a disclosure index (e.g., Chen et al. 2017; Chow and Wong-Boren 1987; Makhija and Patton 2004; Raffournier 1995) that compares an expected or preferred list of items to the effectively disclosed figures and aggregates the results to a score. We do not want to restrict the collected data to a predefined set of items, and we want to avoid the subjectivity involved in setting up a list of disclosures. Thus, we employ a disclosure score based on the number of all voluntary TLCF disclosures weighted by their quality of presentation. While all items are considered in our disclosure score, we record the content of each item and classify them into different content groups, enabling us to analyze each content category separately in additional tests. This approach provides detailed insights into firms’ disclosure behavior without restricting the scope of the analysis.

We measure the quality of presentation because prior research indicates that financial statement users rely on readily available and salient information (e.g., Huang et al. 2018; Miao et al. 2016). Hirst and Hopkins (1998) find that alternative presentations of the same information affect analysts’ firm valuation judgments differently. Furthermore, Atwood and Reynolds (2008) document that the pricing of realized tax benefits from TLCF is affected by its presentation in income statements. Taken together, the way in which information is presented is important for the information processing of financial statement users. It can be assumed that clearer and more salient disclosure is easier to process and thus is preferred by analysts and investors.

Hence, to measure the quality of tax disclosure, we examine the way in which an item is presented.¹¹ We apply a scale developed for the German ‘Best Annual Report’ competition and used by a number of prior studies, e.g., Daske (2005) and Glaum et al. (2013). Disclosed items are analyzed and scored according to their level of detail and form of disclosure, which we summarize as disclosure type. A higher score is assigned if an item is a precise number vs. an interval or qualitative information. The degree of quantification is commonly used in recent literature to assess the readability or salience of disclosure (e.g., Huang et al. 2018; Lundholm et al. 2014) and the assignment of a higher score for a higher quantification degree reflects the enhanced usefulness, in line with prior evidence on investors’ limited attention and processing costs (e.g., Blankespoor et al. 2020; Miao et al. 2016). Table 2 panel A provides an overview of the scale, and panel B and Appendix C give examples of how points are assigned. We apply the scale to every voluntary TLCF item to account for disclosure type.¹² We then add the scores of the same year and firm to obtain one score for each firm-year observation. The resulting score is our dependent variable DISCL, as presented in Fig. 1.

Table 2 Disclosure Subcategories: Type of Disclosure

Fig. 1

Construction of Disclosure Score DISCL.

Notes: This figure presents an example for the construction of the disclosure score DISCL. It is an outcome of multiplying the two disclosure subcategories content and type of disclosure. The score is calculated for each voluntary TLCF item and the sum of all scores per firm-year is the dependent variable DISCL. Details for the subcategory content are provided in Table 3 and for the subcategory type of disclosure in Table 2

3.3 Uncertainty About the Usability of TLCF

The usability of TLCF depends on the availability of sufficient taxable income in future years. Our definition of uncertainty about this usability is derived from IAS 12.34: deferred tax assets “shall be recognized for the carryforward of unused tax losses […] to the extent that it is probable that future taxable profit will be available against which the unused tax losses […] can be utilized”. Hence, the expectation of future taxable profits is one of the key criteria to assess the value of TLCF. Another important aspect is the existence of a “history of recent losses” (IAS 12.35), which potentially increases the uncertainty about future taxable income. Firms are supposed to consider both factors in their recognition of deferred tax assets for TLCF, and we propose that investors rely on similar criteria to derive the value of TLCF. Based on these criteria, we derive two groups of indicators of TLCF usability uncertainty: future indicators related to the expectation of future income and historic indicators based on a recent history of losses.

3.3.1 Future Indicators

The future indicators of TLCF usability uncertainty measure different aspects of the ratio of TLCF to expected earnings. The best available proxy of future earnings is analysts’ earnings forecasts.¹³ For our main uncertainty measure, we compare expected earnings with the amount of usable TLCF. We investigate three possible cases: I) forecasts and carryforwards are close to each other, II) forecasts are considerably higher than carryforwards, and III) forecasts are considerably lower than carryforwards.

We expect uncertainty to be particularly high in the first case (I), when the amounts are close to each other, because minor deviations of earnings from a forecast can lead to the unexpected offsetting or non-offsetting of the TLCF. In this situation, we expect that investors demand more information about the TLCF. In contrast, if forecasts far exceed carryforwards (II), there is little doubt that the TLCF can be used, and additional disclosure is less necessary. If, on the other hand, carryforwards largely exceed forecasts (III), the prediction is less clear. Two different scenarios can apply. If the forecast is smaller than the TLCF and negative, it is very likely that the TLCF cannot be offset in the near future, and there is little uncertainty about usability. In contrast, if the forecast is smaller than the TLCF and positive, part of the TLCF is usable, but for the remaining part of the TLCF, usability is hard to predict. In this second scenario, uncertainty increases. Since our subsample of case (III) does not include any observations with a negative forecast, we are limited to testing the second scenario of case (III) and anticipating an increased demand for information regarding TLCF.

As a basis for our future indicators, we use the last mean earnings before tax (EBT) analyst forecast for year t + 1, issued before the end of fiscal year t, provided by I/B/E/S.¹⁴ At this point in time, financial statement notes for period t are not available and cannot affect the forecast, reducing concerns about reverse causality.¹⁵ Moreover, this forecast is available for all observations in our sample. The amount of TLCF is not available in standard databases and must be hand-collected from tax footnotes. However, IAS 12 does not require the disclosure of this amount, and only 16% of our observations disclose it voluntarily. Thus, we calculate TLCF by grossing deferred tax assets recognized for TLCF. If the tax rate is disclosed, we divide deferred tax assets for TLCF by the tax rate that the firm uses to calculate deferred taxes; otherwise, we divide it by the (average) statutory tax rate as stated by the firm.¹⁶

Based on the forecast and the TLCF, we calculate the difference between the two amounts. For case (I), we partition the differential amount into deciles and identify the two deciles that are closest to zero (the total range of the two deciles combined is −0.628 to 0.030 Bn €). We create an indicator variable, AEF≈TLCF, that has a value of 1 if the difference lies in these two deciles. We expect AEF≈TLCF to have a positive association with disclosure because uncertainty increases when forecasts are close to TLCF. For case (II), we create a second indicator variable, AEF>TLCF, that has a value of 1 when earnings forecasts are higher than TLCF and the difference is not included in the two deciles around zero. We expect a negative relation because less disclosure is needed when the forecast is considerably higher than the TLCF. For case (III), our third indicator variable, AEF<TLCF, has a value of 1 when the amount of TLCF is higher than the forecast and the difference does not lie within two deciles around zero. We expect a positive association for AEF<TLCF.

Another factor to consider in assessing uncertainty is the dispersion of forecasts. Large variation in forecasted earnings indicates disagreement between analysts and complicates the assessment of TLCF usability for an investor. Therefore, our fourth future indicator, STDEV, measures the standard deviation of the last mean EBT forecast for the following fiscal year, scaled by the absolute value of the last mean EBT forecast (AEF). We expect a positive association because higher dispersion of analyst forecasts induces uncertainty for investors and thus increases the demand for disclosure.

3.3.2 Historic Indicators

The role of historic indicators is twofold. First, they indicate whether a firm has any information to disclose about TLCF. Without prior negative earnings and accordingly without TLCF, no information about this item can be disclosed. Although nearly all of our sample firms have TLCF (see Footnote 9 for details), it seems plausible that a firm has more to tell about TLCF and that this information is more relevant if the loss emerged in the recent past than if TLCF resulted from negative earnings ten years ago. Accordingly, we expect a higher level of disclosure if a firm had losses in recent years, indicating more disclosure if more information is available and if it is more relevant. Nevertheless, a firm still must decide whether to publish the information in light of the incentives and costs discussed above. Second, the historic indicators are alternative proxies for uncertainty. Hayn (1995) finds that losses are less informative about future earnings than profits are. Hence, it is harder to assess the usability of TLCF if a firm has a recent loss since future earnings are harder to predict. Based on the literature, we expect higher uncertainty and hence higher disclosure if a firm has a history of recent losses.

We use three different historic indicators. The first indicator variable, LH_5Y, measures whether a firm had at least one negative EBT in the past five years. The second variable, LH_CUM, is based on the accounting standards’ definition of a loss history and is more restrictive than the first proxy: the variable has a value of 1 if a firm has a cumulative negative EBT in the current and the two previous years.¹⁷ Hence, only very large losses are considered here. The third measure, N_LOSS, counts the years with negative EBT in the past five years and ranges from zero to five. For all historic indicators, we expect a positive association with DISCL because the existence and a higher frequency of losses should increase uncertainty and hence disclosure.

3.4 Model

To analyze the association between disclosure and uncertainty, we estimate the following regression model:

$$\mathit{DISCL}_{it}=\beta _{0}+\beta _{1}\mathit{UNCERTAINTY}\,\text{INDICATOR}_{it}+\sum \beta\ \mathit{control}_{it}+\varepsilon _{it}$$

(1)

where firms are identified by i and years by t. Detailed variable definitions are presented in Appendix B. UNCERTAINTY INDICATOR represents the individual FUTURE INDICATORS (AEF≈TLCF, AEF>TLCF, AEF<TLCF, and STDEV) or one of the HISTORIC INDICATORS (LH_5Y, LH_CUM, or N_LOSS). We estimate the model separately for each of our indicators.¹⁸

A common concern in disclosure research is that specific financial statement disclosures simply reflect a firm’s overall disclosure behavior. We can address this concern in our one-country setting by employing a unique dataset. Our research design includes a proxy for the annual report disclosure quality (AR_DISCL). This is an important control variable, as it is a measure for the general (TLCF-unrelated) disclosure behavior of a firm. The score is based on the German yearly annual report competition of ‘Manager Magazin’ (the same as the scale for our dependent variable DISCL). For this competition, every year since 1995, the annual reports of large German listed companies are analyzed with respect to content, design, and language by an expert research group of the University of Münster. Therefore, points are assigned with a checklist including more than 300 criteria grounded in theory which has been compiled based on several surveys of fund managers, financial analysts, auditors, and private investors (Glaum et al. 2013). The criteria refer to different parts of the annual report, including reports that are mainly reflecting past events (e.g., the notes) and parts with forward-looking information, such as the management report. The experts examine whether the information is reported, in how much detail it is reported, and what kind of disclosure it is (comparative, qualitative etc.) and aggregate the assigned points to a summary score. We use the results of the ‘annual report content’ category for which the scores range from zero to 100 (100 denotes the highest level of content quality).¹⁹ The score is divided by 100, resulting in a score between zero and one for our variable AR_DISCL, with a higher score indicating higher quality of content. The advantage of using this well-established score (Daske 2005; Glaum et al. 2013) instead of, for example, the length of the annual report is that it incorporates the same disclosure characteristics for the full annual report that we study for TLCF, i.e., content and disclosure type. Further, it is a comprehensive and time-variant measure of a firm’s disclosure quality, capturing potential firm-specific strategies and incentives.

Our other firm specific control variables are derived from the disclosure literature and the specific characteristics of TLCF. The first is ∆DTA_TLCF, denoting the change in deferred tax assets for TLCF. This item can indicate management’s earnings expectations because international accounting standards allow the recognition of deferred tax assets only if sufficient future taxable income is likely to be available. However, whether a change in deferred tax assets increases or reduces the voluntary disclosure level is unclear. An increase in deferred tax assets for TLCF can, on the one hand, indicate new TLCF and hence the opportunity to disclose more information. On the other hand, an increase can indicate improved earnings expectations that result in the recognition of deferred tax assets on existing TLCF. In both scenarios, it is not obvious whether or how new deferred tax assets affect uncertainty about TLCF usability and hence the voluntary disclosure level. We calculate ∆DTA_TLCF as the change in deferred tax assets for TLCF from the previous to the current year, scaled by total assets and (for ease of exposition) multiplied by 100. The second control variable indicates whether the firm has negative earnings in the current year. A current loss can increase existing TLCF and draw more attention to the topic, possibly increasing disclosure. The indicator variable LOSS has a value of 1 if the firm has a negative EBT in the current year and zero otherwise. Analyst following (AN_FOL) (e.g., Lang and Lundholm 1996) proxies for the information environment of a firm and supports governance mechanisms through information dissemination. Moreover, it may also capture whether TLCF information is publicly available beyond annual report disclosures. Furthermore, previous research indicates a significant association between the level of disclosure and firm size (SIZE) (Cooke 1989; Craig and Diga 1998), leverage (LEV) (Ismail and Chandler 2005; Meek et al. 1995), profitability (∆EBT) (Broberg et al. 2010; Singhvi and Desai 1971), audit firm (AUD) (Singhvi and Desai 1971), and CEO turnover (CEO_TO) (Kwak et al. 2011). In addition, we control for experience with accounting regulations by measuring the number of years that have passed since the firm adopted IFRS/IAS (IFRS_AD).²⁰ To control for other unobserved effects, we include year and industry fixed effects (one-digit Standard Industrial Classification (SIC) code).²¹

4 Firms’ Disclosure Choice

4.1 Descriptive Results

To assess disclosure, we first investigate the content of the TLCF information in annual reports’ notes. We identify 15 different voluntary items. Table 3 gives an overview of the items and the frequency of these disclosures. Each item appears in several firm-years, and the column “Absolute” presents the number of observations that disclose the item. The column “Percent of Total Observations” puts this number in relation to the total number of 723 observations.

Table 3 Disclosure Subcategories: Content

The most frequent disclosure is the effect of TLCF on tax reconciliation in 52% of all annual reports. In addition, 43% of the observations disclose the total amount of TLCF, which is comparable to 51.7% in Dreher et al. (2024, p. 11). Four other voluntary disclosures are reported in at least 30% of statements: the expiration date of total TLCF, deferred tax income/expenses recognized in the current year’s income statement due to unused TLCF, valuation allowance for deferred tax assets on TLCF, and distinction between corporate tax loss and trade tax loss. To group the different items, we divide the voluntary disclosures into four subcategories. The first contains basic information about TLCF, i.e., amount and expiration date. The second category comprises items that explain why or to what extent TLCF (or the recognized deferred taxes) have changed, have been used or have affected the current fiscal year’s income. The third group gives information about valuation allowances and deferred taxes that have not been recognized. Any other kind of disclosure is included in the fourth category. We further analyze the four categories by estimating our main model separately for each of the disclosure categories. Details are outlined in the subcategory regression section.

Next, we present several figures that describe our aggregate disclosure variable DISCL. Figure 2 shows the development of the 25th percentile, the mean, and the 75th percentile of DISCL over our sample period 2005 to 2016. The graph shows no clear trend but indicates that the disclosure behavior varies over time. Further, the large range between the 25th and 75th percentile is an indicator for a substantial cross-sectional dispersion of the DISCL variable in the respective year. To explore the within-firm dispersion of DISCL, a histogram in Fig. 3 presents the frequency of each sample firm’s standard deviation of the disclosure score. The bin width is one unit of the within-firm standard deviation of DISCL. The graph illustrates that the vast majority of firms have a standard deviation between zero and three; the average within-firm dispersion is 2.22. This finding indicates a modest change of voluntary disclosure within firms over time and corroborates our earlier argument in the hypothesis development section on the quasi-mandatory continuity of disclosure choices.

Fig. 2

Distribution of the Disclosure Score DISCL over Time.

Notes: This figure presents the 75th percentile (upper end), mean, and 25th percentile (lower end) of the disclosure score DISCL over our sample period

Fig. 3

Frequency of Within-Firm Standard Deviation of the Disclosure Score DISCL.

Notes: This figure presents a histogram for the within-firm standard deviation of the disclosure score DISCL. The bin width is one unit of the within-firm standard deviation of DISCL

Figure 4 shows boxplots of the disclosure score DISCL for the three uncertainty indicators that compare the earnings forecasts and TLCF: AEF≈TLC, AEF<TLC, and AEF>TLC. The figure indicates the highest disclosure score if earnings forecasts and TLCF are close (AEF≈TLC) with respect to the upper adjacent value and the outside values. We expect this variable to capture a situation with high uncertainty. The lowest disclosure scores are visible for AEF>TLCF, which we assume to signal a low level of uncertainty.

Fig. 4

Disclosure Boxplot.

Notes: This figure presents boxplots for the disclosure score DISCL over the three FUTURE INDICATOR variables AEF>TLCF, AEF≈TLCF, and AEF<TLCF

Table 4 presents summary statistics, and Table 5 presents a Spearman correlation matrix. The average voluntary disclosure score per year and firm (DISCL) amounts to 8.88 with a median of 7.5 and a range from 0 to 42. Approximately 18% of our observations have a difference between forecasts and carryforwards close to zero (AEF≈TLC), 10% have a carryforward surplus (AEF<TLC) and 72% have a substantively higher forecast than TLCF (AEF>TLC). Approximately 8% of observations have a dominating loss in the current three-year period (LH_CUM), and 31% have at least one loss in the five previous years (LH_5Y). With regard to the control variables, 10% have a loss in the current year, 90% are audited by a Big 4 auditor, and the average annual report disclosure quality score is 0.59. Summary statistics are also presented for the disclosure content subcategories derived from Table 3 and the different types of disclosure as presented in Table 2. The highest average score for content is observable for BASIC information with 2.78 and a maximum of 16. The subcategories for disclosure type show that QUANT has the highest average score with 3.54, followed by DETAIL with 3.18.

Table 4 Summary Statistics

Table 5 Spearman Correlation Matrix

The correlation matrix indicates a positive and significant association of DISCL with the uncertainty variables and a negative association with AEF>TLCF, which represents lower uncertainty. These correlations provide preliminary support for the hypothesized increase in TLCF disclosure under TLCF usability uncertainty.²²

4.2 When Is Additional TLCF Information Disclosed? Disclosure Score Regression Results

We estimate a pooled cross-sectional regression with standard errors clustered by firm (reported in parentheses).²³ Panel A of Table 6 gives an overview of the results for the FUTURE INDICATORS in columns I–IV; panel B shows columns V–VII with the different HISTORIC INDICATORS. Year and industry fixed effects are included in all models but are not reported. Consistent with our expectations, AEF≈TLCF and the dispersion of earnings forecasts measured by STDEV have significant coefficients with a positive sign. The results indicate that firms disclose on average more information when TLCF usability is uncertain because TLCF is close to earnings forecasts or because it is harder to estimate future earnings. We expect and find a negative and significant coefficient for AEF>TLCF, indicating that if TLCF is expected to be offset in the following fiscal year, uncertainty is low and less disclosure is provided. We do not find significant results for AEF<TLCF. In columns V–VII (panel B) with the HISTORIC INDICATORS, we find positive and significant coefficients for all loss history variables. At this point, we cannot disentangle whether the relation between disclosure and recent losses is driven primarily by the availability of TLCF information or by uncertainty about future earnings. To control for the availability of information, we estimate a selection model in the robustness section.

Table 6 OLS Regression Estimates (Dependent Variable: DISCL)

Regarding the magnitude of the documented relations, firms’ disclosure score is on average 2.8 points higher if their earnings forecasts and TLCF are close to each other (AEF≈TLCF = 1). This difference can, for example, result from an additional quantitative (2 points) and a qualitative item (0.5). Mind that the sample mean for the disclosure score DISCL is 8.88 (median 7.5). Observations with a cumulative loss history (LH_CUM) have, on average, a disclosure score nearly four points higher than the score of firm-years without a cumulative loss history, which can, for example, indicate additional quantitative information (2 points) presented in a table (1.5 points) and an additional qualitative item (0.5 points). For our control variables, we find significant coefficients for SIZE (positive) and AN_FOL (negative), in line with higher disclosure scores for larger firms with fewer analysts following them. With respect to the control variable AR_DISCL, we find a strong positive relation with DISCL, as expected. This finding suggests related disclosure behavior in the tax footnote and the overall annual report, yet incremental TLCF disclosure is significantly associated with uncertainty.

To further address the concern that tax disclosures mirror the overall disclosure strategy of a firm, we employ a placebo test and substitute our TLCF disclosure score DISCL with our control AR_DISCL as the dependent variable and rerun the baseline model. If firms increase their overall disclosure in the case of uncertainty, we should find similar results for the uncertainty variables as in our baseline model. The results for the different model specifications are reported in the online Appendix (Table OA15). The findings support the positive association between DISCL and AR_DISCL. However, the remaining results differ considerably from our baseline findings. The HISTORIC INDICATORS have negative coefficients (contrary to our baseline model), weakly significant only for N_LOSS. Among the FUTURE INDICATORS, STDEV is the only variable that has the same (positive) sign as in our main regression and a significant coefficient.²⁴ This finding seems plausible because the standard deviation of earnings forecasts STDEV is our least specific and rather generic uncertainty measure, while the other variables measure uncertainty very closely tied to TLCF. Our findings thus align with the results of Chen et al. (2002), indicating that the overall disclosure quality increases in the case of uncertain future earnings but also suggesting that this does not hold in the case of TLCF-specific uncertainty. This result corroborates our baseline finding that firms provide specific voluntary tax disclosures where they anticipate investors’ need for additional information.

4.3 What Kind of Information Is Disclosed?

4.3.1 Content

Next, we aim to shed light on the nature of disclosed information, i.e., content and type. In Table 3, we identify four categories of disclosed items: 1) Basic TLCF Information, 2) information about Changes and Effect on Income, 3) Valuation Allowance Information, and 4) Other information. To analyze for which type of information our main results are most pronounced, we repeat our basic regressions for subcategories 1) to 4).²⁵ Table 7, columns I to IV, shows the aggregated estimation results. Each coefficient is from a separate estimation of the basic Model (1) with the respective disclosure subscore as the dependent variable. Full estimation results for all models are available in the online Appendix (Tables OA1–OA4).

Table 7 Aggregated Results for Disclosure Subcategories (Dependent Variable: Disclosure Subcategory)

The dependent variable in column I is the disclosure score for Basic TLCF Information. We do not find a significant relation of the score with most future or historic uncertainty indicators. Only STDEV has a significant coefficient but with a negative sign, reverse to our main results. This finding suggests that the disclosure of essential TLCF information is basically unrelated to uncertainty about TLCF usability. Column II presents the results for Changes and Effect on Income. We see a strong significant positive association for the HISTORIC INDICATORS, while among the coefficients of the FUTURE INDICATORS only AEF>TLCF is weakly significant. Hence, in the case of recent losses, firms seem to report more information about how and why TLCF and the respective deferred taxes have changed and affected the current year’s income. The findings for the disclosure category Valuation Allowance Information are shown in column III. The results are opposite to those in column II: two of the FUTURE INDICATORS have significant coefficients with the expected sign, while the HISTORIC INDICATORS have insignificant coefficients. This finding indicates that the results for the forward-looking uncertainty proxies are associated mainly with valuation allowance information. When uncertainty about TLCF usability increases, firms must adjust the amount of recognized deferred tax assets and voluntarily report the respective valuation allowance activities in the tax footnote. Column IV shows results for the category Others. Coefficients for all FUTURE and HISTORIC INDICATORS are significant with the expected sign. The category includes diverse and unsystematic disclosed items, for example, the effect of TLCF on the tax reconciliation and a residual category Other disclosures with items that do not fit in any of the other categories (see Table 3 for details). The results indicate that firms further complement our identified voluntary items with additional individual information under uncertainty.

In sum, evidence from the disclosure content subcategories suggests four conclusions. First, in the case of forward-looking uncertainty, firms primarily provide valuation allowance details. Second, in the case of recent losses, firms mainly provide information about TLCF changes and the effect on the current year’s income. Third, providing basic TLCF information seems to be unrelated to both types of uncertainty. Fourth, firms report additional unsystematic information under uncertainty. These findings appear intuitive and corroborate the choice of our disclosure measures. At the same time, they provide granular and novel insights into firms’ tax disclosure behavior. Firms cater to the anticipated information needs of financial statement users by providing more explanations and guidance when TLCF usability is uncertain. Regarding the content of the voluntary disclosure, firms stick closely to the criteria set by IAS 12 for the recognition of TLCF-DTA. This finding is related to evidence that firms mainly use objective accounting standard criteria when setting a mandatory TLCF item, the US-GAAP valuation allowance (Dhaliwal et al. 2013; Goldman et al. 2022). We extend this research by identifying different uncertainty signals as explaining the content of voluntary TLCF disclosure.²⁶

4.3.2 Type of Disclosure

We document the choice of disclosure type in columns V to IX of Table 7 (online Appendix Tables OA5–OA9). The categories are based on the construction of the disclosure score (Table 2) and reflect the type and presentation of information. We find that Quantitative Information, Qualitative and Interval Information hardly exhibits any statistically significant association, indicating that firms seem not to use these disclosure types to reduce uncertainty. In contrast, providing Additional Detail appears to be a frequently used disclosure choice when TLCF uncertainty is high, as this category has significant coefficients with the expected signs for all FUTURE and HISTORIC INDICATORS. Highlighting information in Tables also shows significant associations with four uncertainty indicators. Taken together, under uncertainty, firms explain additional details, and enhance the processing of stakeholders by making this information visible in tables. These findings add to prior evidence on firms’ strategic use of certain presentation formats for tax disclosure (e.g., Chychyla et al. 2022), documenting how firms apply these disclosure choices to mitigate TLCF-related uncertainty.

4.4 Cross-Sectional Tax Avoidance Test

To provide further support for the relation between uncertainty and TLCF disclosure, we derive another indicator of uncertainty about TLCF usability from prior literature and use it in a cross-sectional test: firms’ tax avoidance behavior (McGuire et al. 2016). To assess the future usability of TLCF, firms should consider expected future taxable income and also take into account their own ability to generate such income via tax planning activities (IAS 12.29 (b)). Hence, if a firm is active in tax planning, this could reduce uncertainty about TLCF usability as it indicates that, if necessary, the firm can create taxable income to offset TLCF. In line with this notion, McGuire et al. (2016) find that investors positively value TLCFs when firms have high levels and high variability of prior tax avoidance. We split our sample into firms with high and low tax planning activities. Assuming that tax planning reduces uncertainty about TLCF usability, we expect higher uncertainty and disclosure for firms with low tax planning activities.

Building on McGuire et al. (2016), we use two measures of firms’ prior tax planning activities: the level and variability of the five-year cash ETR. The five-year cash ETR (CETR5) is calculated as the sum of taxes paid divided by pre-tax income less discontinued operations and extraordinary items over the five-year period t‑5 to t‑1 (Kerr 2019; McGuire et al. 2016). The variability (CVCETR5) is the coefficient of variation, calculated as the standard deviation of annual cash ETRs divided by the absolute value of the mean of annual cash ETRs for the five-year period t‑5 to t‑1 (McGuire et al. 2013; McGuire et al. 2016). We assign firms into low and high tax planning samples based on yearly median splits for CETR5 and CVCETR5. Results are presented in Table 8.

Table 8 Cross-sectional Tests for Tax Planning Subsamples (Dependent Variable: DISCL)

Again, results are aggregated and show only the uncertainty coefficients. Detailed results for all 24 estimations are available in the online Appendix (Tables OA10–OA13). Columns I and II show the sample split into high (I) and low (II) tax planning firm-years based on CETR5. While only two out of six uncertainty variables show significant coefficients in column I, all coefficients are significant in column II. Further, the coefficients’ size is considerably higher for almost all variables in column II. Results for high (III) and low (IV) tax planning based on the variability of the five-year cash ETR (CVCETR5) present an even more pronounced pattern: they show no significant coefficients for the high tax planning sample, but all uncertainty variables have significant coefficients with much higher magnitudes for the low tax planning sample firms. Taken together, results indicate that the association between uncertainty and voluntary TLCF disclosure is more pronounced if firms are less active in tax planning. This finding is in line with McGuire et al. (2016) and our expectation that active tax planners are able to create taxable income to offset TLCF if necessary, reducing uncertainty and the need for more disclosure.

5 Sensitivity Analyses

5.1 Exogenous Uncertainty Shock

To validate our baseline results, we examine the European sovereign debt crisis as an exogenous shock to uncertainty. The financial sector was hit severely by charges on Greek debt, and concerns emerged about whether the euro zone would prevail (Rooney 2011; Viñals 2011). This situation created high uncertainty about future earnings, particularly for the financial industry. 2011 was characterized by large write-offs on debt that hit profits for several financial institutions. To avoid the direct effect of current negative earnings on TLCF disclosure, we examine disclosure in 2012.²⁷ The advantage of the year 2012 is that while there was a high uncertainty about future earnings (Laurent and Slater 2012), current operating earnings in German Financial Institutions were positive.²⁸ To examine whether firms in the industries finance, insurance, and real estate increase TLCF disclosure in 2012, we include an interaction of a year-2012 indicator and a finance-industry indicator in our baseline model (1). The results for the interaction variable show positive and significant coefficients (tabulated in the online Appendix Table OA16). The remaining results for our other uncertainty indicators and control variables are qualitatively unchanged. This finding indicates that firms which face an exogenous rise in uncertainty about future earnings (and hence TLCF usability) increase TLCF disclosure, corroborating our baseline results and mitigating endogeneity concerns.

5.2 Disclosure Score

Next, we analyze the robustness of our disclosure score. First, we use the number of reported items as a quantitative measure instead of the disclosure index DISCL, i.e., we do not apply the scale to the disclosed items. The results are presented in the online Appendix Table OA17 and do not change qualitatively. This finding mitigates potential concerns about the effect of the application of our disclosure scale on the results. On the one hand, it could also indicate that the number of disclosures is the driving force of the observed main effect. On the other hand, results of our subcategory tests indicate that firms apply specific types of disclosure under uncertainty.

Second, we vary the weights assigned to the disclosed items in untabulated tests. While the scoring in our main results is based on the German ‘Best Annual Report’ competition which has been used in prior research (e.g., Daske 2005; Glaum et al. 2013), we test robustness of our results to a modified weighting. For example, we assign an incremental score of 1 point instead of 1.5 points for additional information (DETAIL) and the disclosure in a table or graph (TABLE). Applying the adjusted weights, results of our disclosure regressions and all respective subcategory estimations are qualitatively unchanged.

5.3 Forecasting Horizon

We examine the robustness of our future uncertainty indicators with respect to the forecasting horizon. Anecdotal evidence indicates that firms often use forecasting horizons of more than one year to assess TLCF usability.²⁹ We use the forecast for only one year in our main tests for the following two reasons. First, we expect to obtain the most precise forecast for t + 1, while forecasts for later years are much noisier. Second, the availability of analyst forecasts beyond t + 2 is limited and would further reduce our already modest sample size. However, in this robustness test, we repeat our main tests by using the sum of the forecasts for t + 1 and t + 2 to create the uncertainty proxies. Our sample is reduced by nine observations. We can no longer estimate the model with AEF<TLCF because under this classification, there is no case in which TLCF is considerably higher than the cumulated forecast. Untabulated tests show that the results for the remaining two uncertainty variables, AEF≈TLCF and AEF>TLCF, have the same sign and are both significant at the one-percent level, suggesting that our findings are robust to this modification.

5.4 Other Tests

In the online supplement, we further examine the robustness of our results. In particular, we address potential sample selection concerns estimating (i) a Heckman (1979) selection model.³⁰ The first step specifies a probit regression and models the availability of TLCF information. The second equation is our main model including the inverse Mills ratio from the first stage and models the relation between the level of disclosure and uncertainty. In sum, except for STDEV, the results of the selection model corroborate our baseline findings (Table OA14).

To control for the effect of outliers, we (ii) winsorize all continuous variables above the 99th percentile and below the 1st percentile or (iii) exclude bank, insurance, and financial firms from our sample (SIC Code 60-67). To examine the robustness to the effect of the financial crisis (iv), we drop the crisis years 2008 and 2009 from our sample. Finally, we (v) include different additional variables controlling for the materiality of TLCF, internationalization of the firm, and market-based uncertainty measures. Throughout these modifications our results remain robust. If we estimate our models without any control variables, AEF≈TLCF and LH_CUM have insignificant coefficients but the remaining results are unaffected.

6 Conclusion

This study provides a comprehensive analysis of the determinants of firms’ voluntary TLCF disclosures using data hand-collected from tax footnotes. We analyze when and how firms report additional TLCF information in a German setting characterized by a high demand for voluntary TLCF disclosure due to a lack of confidence in deferred tax assets on TLCF (Chludek 2011; Flagmeier 2022). We examine different future and historic signals for uncertainty about the usability of TLCF and find a strong positive association between disclosure and uncertainty. Disaggregating our disclosure score indicates that firms disclose information on changes and the effect on income if uncertainty is created by (historic) losses and disclose valuation allowance-related information if uncertainty is based on (future) forecasts. The type of disclosure is focused on additional details and information is reported in tables. Our findings are robust to our controlling for the availability of TLCF information in a selection model and to several other sensitivity tests. Furthermore, our research design assures that the disclosure inference we draw are tax specific and do not simply stem from the overall disclosure policy of a firm.

We contribute to the tax disclosure literature with a comprehensive textual analysis of a unique TLCF disclosure dataset and an innovative identification strategy to control for the overall reporting quality of a firm. TLCF information can be highly useful in estimating a firm’s future net income, and it is important to understand whether firms’ disclosure is helpful in this regard. Our findings indicate strategic disclosure behavior that mainly caters to investors’ need for additional information regarding the type of disclosure. With regard to content, firms’ disclosure depends on the signal of uncertainty about TLCF usability. Our results provide valuable insights for regulators, investors, and firms across various jurisdictions adopting IFRS, especially those where deferred tax assets are similarly perceived as less informative. Our study also outlines the complexity of TLCF information and calls for future research on how firms could further improve the informativeness of their tax disclosures.

Appendix

1.1 Appendix A

1.1.1 Differentiation in Mandatory IAS 12 and Voluntary Disclosures

We identify six mandatory IAS 12 disclosures concerning TLCF:

• the amount of deferred tax assets recognized in the balance sheet for unused TLCF (IAS 12.81 (g) (i))

• the amount of the benefit from a previously unrecognized TLCF of a prior period that is used to reduce deferred tax expense (IAS 12.80 (f))

• the amount of the benefit from a previously unrecognized TLCF of a prior period that is used to reduce current tax expense (IAS 12.80 (e))

• the amount of unused TLCF for which no deferred tax asset is recognized in the balance sheet (IAS 12.81 (e))

• the expiration date of unused TLCF for which no deferred tax asset is recognized in the balance sheet (IAS 12.81 (e))

• the amount and nature of evidence supporting the recognition of a deferred tax asset when the entity has suffered a loss in the current or preceding period (IAS 12.82 (b))

One item cannot clearly be classified as voluntary or mandatory: the amount of deferred tax income or expense recognized in the income statement due to unused TLCF (IAS 12.81 (g) (ii)). The classification is ambiguous due to the additional remark that this disclosure is only necessary if it is not apparent from changes in the amounts recognized in the balance sheet. Thus, the firm can choose whether it states the amount separately or leaves it to the reader to derive it from the balance sheet. Due to this discretion, we classify this item as voluntary. Repeating our analysis with this item classified as mandatory does not affect our main results. Furthermore, every disclosure beyond the six required items is considered voluntary.

1.2 Appendix B

Table B1 Variable Description

1.3 Appendix C

Table C1 Disclosure Examples from Tax Footnotes