To Herd or Not to Herd: Do Intangible Assets Affect the Behavior of Financial Analyst Recommendations?

Abstract

The extent to which financial analysts provide “herd” rather than “bold” (or anti-herd) earnings forecasts has important implications for market efficiency. Identifying any contributing factor(s) for financial analyst herding behavior can lead to policies to help reduce such harmful conduct. Our 2 proxies for intangible asset intensity are found to have a differential impact on analyst herding behavior. More specifically, increases in firm-specific reported balance sheet intangible assets (level of patents granted) are associated with heightened (reduced) herding behavior. Our findings highlight the need for regulatory reforms such as more transparent disclosure, and standardized accounting treatment to better capture a firm’s investment in intangible assets in their financial statements.

Keywords:

• Financial analysts
• Herding behavior
• Intangible assets
• Patents
• Information complexity

JEL Classification::

• G10
• G12
• G14
• G20

Notes

1 This study defines herding as the “propensity of analyst to follow the consensus” (see Welch 2000). In contrast, “bold” (or anti-herd) forecast is the opposite of forecasting close to the consensus estimate.

2 Intangible assets are defined by the Financial Accounting Standards Board (FASB) as “assets, other than a financial asset, that lack physical substance,” while the International Accounting Standards Board definition is “an identifiable nonmonetary asset without physical substance.” Examples of intangible assets include (1) brand name; (2) trademarks; (3) Internet domains and social media presences; (4) Patents; (5) Proprietary computer software; (6) Trade secrets, recipes, processes; (7) Customer lists and databases; and (8) Licenses, leases and royalties. Unfortunately, intangible assets do not have a physical form precluding them from easily being converted into cash. Consequently, working out what intangible assets are truly worth can be challenging.

3 Please refer to our literature review in the second section for more detailed discussion.

4 For example, in 2018, intangible assets represented 84% of all enterprise value of firms on the S&P 500 (USD21.03 trillion out of USD25.03 trillion), a massive increase from just 17% in 1975 (USD122 billion out of USD715 billion). Likewise, in most Organization for Economic Cooperation and Development countries, intangible assets constitute about 5% to 10% of gross domestic product, in which the average growth rate of intangible assets from 1995 to 2014 have outpaced those of tangible assets for a select group of developed countries (see Demmou, Stefanescu, and Arquié 2019).

5 André, Dionysiou, and Tsalavoutas (2018) found that mandated disclosures under the International Accounting Standard (IAS 36) Impairment of Assets help improve analyst forecast accuracy. This evidence is consistent with the argument that mandatory disclosures results in more transparent financial statements (see Pownall and Schipper 1999), thereby reducing economic uncertainty about firms (see Hope 2003; Anctil et al. 2004).

6 When a company files for a patent, the cost to acquire the patent (e.g., registration, documentation, and other legal fees associated with the application) will be recorded as the initial asset cost. In contrast, if a firm instead bought a patent from another party, the purchase price is the initial asset cost.

7 There are several ways in which analysts can collect information about patents. They can search in the U.S. Patent and Trademark Office’s patent and trademark system, reading the Official Gazette published by the U.S. Patent and Trademark Office, and/or subscribing to electronic news services that provide updates on firm patents.

8 Psychological research verifies that disagreement of opinion creates anxiety and a desire to seek consensus (see Asch 1952).

9 It is important to note that Kim and Pantzalis (2003) only adjusted their dispersion variable according to the firms’ industry, not the forecast error. Our study adjusts for both forecast error and dispersion level for industry-specific effects, which we believe provides for a more robust and comparable herding tendency measure across firms in different industries.

10 We note that there are data limitations when capturing information on specific intangible asset types. For example, the data availability and estimated size of business spending on intangibles as outlined by Corrado, Hulten, and Sichel (1995) is primarily obtained from survey information and/or cost items from a firm’s income statement.

11 We note that some researchers (see Barth and Kasznik 1999; Barth, Kasznik, and McNichols 2001) have used advertising expenses; and research and development expenses as a proxy for intangible assets. This article does not use these measures as proxies for intangible assets as they are expenses. For example, R&D is inherently risky, without assurance of future benefits, so it should not be considered an asset.

12 LagAccuracy_i,j,t measures the value of forecast accuracy of analyst $i$ for firm $j$ in the previous year t-1. Accuracy_i,j,t, is computed using a similar technique to Clement and Tse (2005), where: $\mathit{Accurac}{y}_{i,j,t}=(\mathit{AFEma}{x}_{j,t}-AF{E}_{i,j,t})/(\mathit{AFEma}{x}_{j,t}-\mathit{AFEmi}{n}_{j,t})$ For each firm j, we compute AFE for analyst $i$ for year $t$ according to Eq. 2. AFEmax_j,t (AFEmin_j,t) is the maximum (minimum) absolute forecast error for analysts who are covering firm $j$ in year $t.$ Since the absolute forecast errors are all relative to the stock price (i.e., price-deflated forecast errors), Accuracy_i,j,t is standardized from 0 to 1, indicating the least and most accurate forecasts, respectively.

13 We apply Clement and Tse’s (2005) below equation to transform our characteristic variables from 0 to 1: $$\mathit{Characterist}{ic}_{i,j,t}=\frac{\mathit{Raw}\_\mathit{Characteristi}{c}_{i,j,t}-\mathit{Raw}\_\mathit{Characteristi}{c\mathrm{}(\min )}_{\mathrm{}j,t}}{\mathit{Raw}\_\mathit{Characteristi}{c\mathrm{}(\max )}_{\mathrm{}j,t}-\mathit{Raw}\_\mathit{Characteristi}{c\mathrm{}(\min )}_{\mathrm{}j,t}}$$ where Raw_Characteristic_i,j,t is analyst $i$’s raw characteristic. Raw_Characteristic (max)_j,t [Raw_Characteristic (min)_j,t] is the maximum [minimum] value of a characteristic for analysts who are following firm $j$ in year $t.$ By way of example, let’s apply this standardization procedure to the variable BrokerSize. Let’s say analyst i’s broker size is 12 and the minimum (maximum) brokerage size for analyst who follow firm j in year t is 5 (30). The standardized broker size is computed as 0.28 (i.e., [(12-5)/(30-5)]). This means that analyst i sits at the 28% level of the range of brokerage sizes for all analyst who follow firm j in year t.

14 We note that a similar procedure is used by Clement and Tse (2005), who omitted price-deflated forecast revisions above 0.10 or below −0.10 and price-deflated analyst forecast error above 0.40 and below −0.40. As our study covers different periods, we eliminate the highest and lowest 1% of observations for these two items.

15 It is noted that the mean value of the dummy variable Herd in the full sample of observations (i.e., 104,332) is 0.264, implying that only 26.4% of all forecast revisions were identified as herd forecasts.

16 Robustness tests were carried out to ensure that our results are not driven by the “dot-com” bubble (or financial crisis periods), in which data from 1997 to 2000 (1997–1998; 2007–2008) were excluded. Overall, the robustness tests indicate that the obtained results are not sensitive to bubble or crisis periods. These results are available to the interested reader upon request.