3D Printing Technology and the Market Value of the Firm

Abstract

3D printing, sometimes known as additive manufacturing or digital direct manufacturing, is an innovative manufacturing technology that has gained notoriety recently. However, despite the promising potential and popularity of the technology, there is not yet evidence of the shareholder value implications of firms’ commitments to 3D printing. In this paper, we assess the effects on shareholder wealth of 3D printing-related announcements for publicly traded firms from 2011 to 2017. We find that the stock market places significantly positive value on announcements associated with rapid prototyping or ad hoc customization applications, while the reaction to announcements related to the use of 3D printing technology for mass production was far less enthusiastic. For firms faced with the decision of whether to implement 3D printing in their manufacturing processes, our findings suggest that the market greatly values 3D printing in several important contexts, but the technology is not a universal panacea.

Appendices

Appendix A. Sample 3D Printing Investment Announcements and Coding

Table 7 Sample 3D printing investment announcements and coding

Appendix B. Event study statistical analysis

In addition to computing the mean abnormal return across all our events, we also perform a series of parametric statistical tests to assess the strength of our results. First, we use a simple t test to assess whether the mean abnormal return significantly differs from the expected value of zero. The t test is a well-accepted technique for assessing statistical significance in event studies (Kolari & Pynnonen, 2011), and Armitage (1995) found through simulation experiments that this test often outperforms more complicated statistical tests in event studies. We also perform the Patell test, which considers the variance of residual stock returns in assessing each abnormal return (Patell, 1976). Finally, we employ the standardized cross-sectional test, which builds upon the Patell test by considering a cross-sectional adjustment to observed variances to compensate for event-induced variance (Boehmer et al., 1991).

We further calculate two nonparametric statistics to ensure the robustness of our results. The first of these statistics is the generalized sign test proposed by Cowan (1992). This test is similar to a binomial sign test, but as opposed to assessing whether a binomial proportion differs significantly from 50%, this test accounts for the potentially non-zero skewness of daily stock returns by considering the returns observed in the estimation period (Brown & Warner, 1985). We also perform the rank test proposed by Corrado and Zivney (1992). This test is a modification of the Wilcoxon sign-rank test, which imputes a stock-specific rather than cross-sectional ranking in assessing statistical significance.

For robustness, we also compare the results with the market model to three other models: the market-adjusted model, the mean-adjusted model, and the Fama-French-Carhart model. The market-adjusted return model does not use an estimation period, but it instead calculates the abnormal return as the difference between the market return on a given day and the firm’s return on that day (Binder, 1998). The mean-adjusted return model does not consider the market return and instead calculates the abnormal return as the difference between the firm’s mean return during the estimation period and its return on the event date (Binder, 1998). Fama and French (1993) suggest a three-factor multiple regression model that considers market capitalization values and book-to-market ratios in addition to the market return. In addition, a final model suggested by Carhart (1997) adds a fourth factor, momentum, to the prior model.

McWilliams and Siegel (1997) note several tenets in designing event studies to ensure appropriate rigor and quality of results. First, as many statistical tests used to analyze abnormal returns assume a normal distribution, the sample size must be sufficiently large such that the sampling distribution is approximately normal. Generally, the Central Limit Theorem holds that the sampling distribution is approximately normal when N ≥ 30. Given our final sample size of 84, this assumption should hold. Second, as parametric statistics may be sensitive to outliers, McWilliams and Siegel (1997) exhort researchers to include non-parametric statistics as well, particularly noting the sign test. We include a battery of tests in our event study to ensure robustness, including the non-parametric generalized sign test and the non-parametric rank test. Third, McWilliams and Siegel (1997) argue that abnormal returns should be analyzed over shorter event windows, as longer windows reduce the power of statistical tests and may include the influences of extraneous market effects and/or confounding events. We report abnormal returns over a window ranging up to five days for completeness, but we focus our statistical analysis largely on day 0, or the smallest event window possible. Finally, regression analysis must be performed to control for various factors that can affect abnormal returns, as doing so best explains the reasoning behind observations. Thus, we perform regression analyses in section 5.2 to properly attribute the observed effect to the appropriate sources.