How Analysts Process Information: Technical and Financial Disclosures in the Microprocessor Industry

Abstract

Following Bradshaw (‘Analyst information processing, financial regulation, and academic research’ [2009], and Analysts' forecasts: What do we know after decades of work? [2011]), this paper examines how analysts process information, particularly in an information environment characterised by multiple and potentially complementary information sources. The setting is the microprocessor industry, one in which technical information is particularly significant and complex to digest. Based on 3837 analyst earnings-forecast revisions, issued by 134 analysts, we examine quantitatively the speed, magnitude, and information content of the reactions of individual analysts and subgroups of analysts to both periodic and timely technical disclosures, and as a complement to periodic financial disclosure. We find that analysts are much slower to react to timely technical disclosures than they are to periodic financial disclosures. We find also that technical and financial disclosures complement each other. Furthermore, we find that there is a ‘hierarchy’ of analysts in this particular industry, as evidenced through the strength of reaction to timely technical disclosures. Finally, we find that lower speed in reacting to timely technical disclosures and a higher intensity in the use of timely technical disclosure (in conjunction with periodic financial disclosure) result in greater accuracy, and that more experienced analysts tend to be less accurate. We suggest that the findings may have implications for other industries such as Bio-Tech Pharma.

Acknowledgements

The work reported here was part of the programme of the ESRC Centre for Analysis of Risk and Regulation. We would like to thank Saverio Bozzolan, Salvador Carmona, Christian Leuz, Andrea Menini, Cathy Shakespeare, Ana Simpson, Wim Van der Stede, and the anonymous reviewers of this journal for their helpful comments. We would also like to thank participants at the EFMA Conference (held at Bocconi University, June 2009).

Funding

We would like to acknowledge the financial contribution received from inclusion in the 2005 and 2007 PRIN funding (Programmi di Ricerca Scientifica di Rilevante Interesse Nazionale). We would also like to acknowledge the support of the Economic and Social Research Council (ESRC).

Supplemental Data

Supplemental data for this article can be accessed on the Taylor & Francis website, http://dx.doi.org/10.1080/09638180.2014.969526

Notes

¹ (Baginski, Hassell, & Kimbrough, 2004; Ball & Brown, 1968; Banker & Mashruwala, 2007; Barron, Kile, & O'Keefe, 1999; Bernard & Thomas, 1989; Bozzolan, Trombetta, & Beretta, 2009; Chandra, Procassini, & Waymire, 1999; Clarkson, Kao, & Richardson, 1999; Espinosa, Gietzmann, & Raonic, 2009; Gu & Wang, 2005; Hussainey, Schleicher, & Walker, 2003; Ittner & Larcker, 1998; Rajgopal, Shevlin, and Venkatachalam, 2003; Schleicher & Walker, 1999; Tellis & Johnson, 2007; Vanstraelen, Zarzeski, & Robb, 2003; Xu, Magnan, & André, 2007) See also Sievers, Mokwa, and Keienburg (2013) on the incremental value relevance of non-financial metrics in the context of venture capital-backed firms.

²Of these, Hunton et al. (2010) is the only one to study analysts, although the focus in that paper is on buy-side rather than sell-side analysts.

³Also, since the early 1990s, the firm has played a lead role in ensuring that innovation on the part of suppliers and complementors matches the ambitions and time-lines of leading chip makers, including Intel itself (Miller & O'Leary, 2007).

⁴While the microprocessor industry may be distinctive in this respect, it may also have similarities to the biopharmaceuticals industry, as examined by Espinosa et al. (2009).

⁵Consistent with Francis et al. (2002, p. 315), we use the word complement to capture the notion of a positive association between two information signals or sources, and not to indicate any particular structure or mechanism that may produce it. When we refer to ‘non-complements’, we mean the use of a single signal or information source.

⁶We obtained I/B/E/S data across a sequence of two downloads (in December 2006 and June 2008) for the entire sample period. Differently from Ljungqvist, Malloy, and Marston (2009), we do not observe any change in the number of recommendations per analyst, in the value of the earnings forecasts and in the release forecast date. Data are available on request from the authors.

⁷According to Miller and O'Leary, technical analysts ‘play a pivotal role in the evaluation of products and processes in the industry … and function both as a “filter” and as a third-party evaluation and validation resource for analysts’ (2000, p. 2).

⁸Microprocessors are built from silicon wafers, which are thin disks. Each wafer may contain many chips of the same type. An individual chip is called a die. Chips are usually laid out in a grid pattern, and arranged to fit as many as possible on the wafer. A single wafer can hold more chips if they are smaller. Because chips are so small, many external factors (i.e. particle of dust or tiny impurities in the silicon) can cause defects in the die.

⁹As a robustness test, we also analysed the entire sample of technical disclosure events reported on the Intel website, without any assessment regarding the relevance of the disclosure event. These findings appear to support the classification carried out by the authors. There are 79 forecast revisions associated with low relevance technical disclosures. When we compute analyst forecast revisions preceded by these low relevance technical disclosures, the coefficient of the revision is positive (0.028) but not statistically significant (t = 1.379). When we combine high-relevance technical disclosure and low-relevance technical disclosures, the analysts forecasts revisions preceded by all technical disclosures is positive (0.009) but again not statistically significant (t = 0.810). However, when we focus on highly relevant technical disclosure only, the coefficient of the revisions is negative (−0.002) and significant at 10% (Table 4, Panel A). This suggests that disclosures qualified as low-relevance by the authors are also considered to be low-relevance by analysts, who do not revise their forecast following these disclosures.

¹⁰No multicollinearity problem affects the fundamental accounting variables in Equation (2).

¹¹We include the periodic technical variables one at a time because they are highly correlated (correlation coefficients above 70%), and multicollinearity problems would affect a multivariate regression.

¹²To control for cross-correlations in the residuals across time, we repeat our tests separated for each of the quarters ending at March, June, September, and December from 2000 to 2007.

¹³The t-test (31.77, p < .01) confirms that the length of the forecast revision period is smaller for the non-financial disclosure subsample.

¹⁴The t-test (0.011, insignificant) suggests that the magnitude of the forecast revisions following financial disclosure is the same as the magnitude of forecast revisions with no financial disclosure in the forecast-revision period.

¹⁵Standardized coefficients are available from the authors upon request.