2.1 Determinant of investment efficiency

The neoclassical theory of investment suggests that a firm should keep investing until the marginal cost of investment equals its marginal benefit, assuming that the firm is operating in a perfect capital market without friction (i.e., Abel, 1983; Hayashi, 1982). Empirically, this idea suggests that corporate investment should be driven solely by investment opportunities as measured by Tobin's Q (Tobin, 1969). Since Tobin's Q captures the market's information about a firm's investment opportunities (Hubbard, 1998), the sensitivity of investment (measured by capital expenditure) to Tobin's Q is often used to measure investment efficiency in the finance and accounting literature (i.e., Biddle et al., 2009; Chen et al., 2007, 2011). While Cooper and Ejarque (2003) point out that Tobin's Q could be a noisy measure of investment opportunities, a recent study conducted by Andrei et al. (2019) finds that the classic Q theory of investment works surprisingly well in recent years, suggesting that Tobin's Q may be a better empirical proxy for the firm's investment opportunities than previously thought.

Previous studies find that firms may depart from the optimal investment level due to some imperfections (Chen et al., 2017). One of the major imperfections is the information asymmetry between corporate decision-makers and outsiders. While having an information advantage over the market may prevent managers from investing efficiently (i.e., Myers, 1984; Richardson, 2006), a failure to employ useful information from the market can also lead to the same outcome. This is because managers need to correctly evaluate investment projects to invest efficiently. This requires managers to take into consideration various decision-relevant factors (i.e., customer demand, industrial competition). However, it is highly unlikely for managers to have complete information about every decision-relevant factor. Therefore, making use of information from markets becomes essential for efficient investment. Since markets aggregate information from various sources and reflect it in the stock price, the informative stock price could produce new information to managers that is beneficial to their investment decision (managerial learning hypothesis).

A growing number of studies provide empirical evidence consistent with the managerial learning hypothesis. For example, Chen et al. (2007) provide empirical evidence that managers are influenced by private information in stock prices when they make corporate investment decisions. They find that investment efficiency, measured by investment-to-price sensitivity, increases with price efficiency, measured by price non-synchronicity and the probability of informed trading. Bakke and Whited (2010) offer an econometric methodology to disentangle stock price movements that are relevant for investment so that they can separately test the effects of mispricing and information on investment. Their work further supports the view that managers incorporate information obtained from the price in making investment decisions. Foucault and Frésard (2012) show that cross-listed firms in the US are more sensitive to investment-to-price sensitivity than firms that never cross-list. They argue that cross-listing strengthens price informativeness, which improves investment efficiency by allowing managers to obtain more useful information from market prices. Loureiro and Taboada (2015) find that firms invest more efficiently following firms' adoption of International Financial Reporting Standards (IFRS). They argue that the adoption of IFRS improves insiders' ability to learn from market prices, thus resulting in real economic gains. In summary, this preceding research shows that informative stock prices can guide managers to make investment decisions.

In the next section, I will discuss in detail how the attention of institutional investors can affect corporate investment efficiency.

2.2 Attention and investment efficiency

Stock prices can influence corporate investment decisions because they transfer useful information from financial market participants to managers. However, out of the massive volume of information, only those receiving market participants' attention would be processed and incorporated into the stock price through trading (Ben-Rephael et al., 2017). Since attention is a limited cognitive resource (Kahneman, 1973), the degree to which the market price can influence corporate investment decisions consequently depends on the attention given by investors. If investors do not pay enough attention to a firm, the stock prices of that firm will contain less useful information or fail to incorporate available information in a timely manner, reducing the informativeness of stock prices. For example, Gennotte and Trueman (1996) find that stock prices are more informative when an earnings announcement is made during trading hours rather than after the market has closed because investors do not pay enough attention to announcements outside trading hours. Similarly, Dellavigna and Pollet (2009) find that investors are less likely to pay attention to earnings announcements if they are released on a Friday. This leads to post-earnings announcement drift resulting from the underreaction caused by limited attention. Hirshleifer et al. (2009) suggest that investors' reaction to an earnings announcement is weaker, and the post-earnings announcement drift is stronger when the earnings are announced on a day of many competing announcements; and Barber and Odean (2008) find that individual investors' trading decisions are influenced by salient, attention-grabbing events. Overall, the above literature suggests that attention is a crucial factor in determining price informativeness.

If the stock price is not informative, managers may fail to obtain useful information from the stock price to gauge market preference when making an investment decision, resulting in inefficient investment. Based on the above argument, I expect a positive relationship between abnormal institutional attention and investment efficiency (managerial learning hypothesis).

Alternatively, institutional investors may be exposed to behavioural bias (i.e., Coval & Shumway, 2005; Frazzini, 2006; Locke & Mann, 2005). If this is the case, attention may amplify their behavioural bias, thus reducing price efficiency. Since stock prices do not accurately reflect firms' fundamental value, this weakens managers' ability to improve investment efficiency. According to this behavioural bias hypothesis, AIA may weaken or impose no impact on investment efficiency or price informativeness.

It is also possible that institutional investors simply act as information mediators instead of information producers on the earnings announcement day. In this case, instead of producing new information and incorporating it into the stock price, an institutional investor may simply transfer managerial information from firm managers to the market. In other words, institutional investors may trade on firms' public information instead of their private information on the earnings announcement day. If this is the case, the incremental information they incorporated into stock prices would not contain any information that is new to corporate managers. As such, managers cannot benefit from such information when making investment decisions. Based on this information mediator hypothesis, AIA may have no impact on corporate investment efficiency. Therefore, the net impact of AIA on corporate investment efficiency remains an empirical question.

3.1 Data

To examine the effect of the abnormal attention of institutional investors on investment efficiency, I employ a sample of US firms appearing in the Compustat database from 2011 to 2018. The sample period starts in 2011 because the data on AIA have only been available since then. Following Da et al. (2011), I begin with a sample of Russell 3000 stocks. I then remove firm-quarter observations with missing financial data and exclude firms operating in the financial industry (SIC codes 6000–6999) because they are subject to different financial accounting processes, and regulated utilities (SIC codes 4900–4999) because they are more regulated than firms operating in other industries. I also require the stocks in the sample to have measures of news searching and news reading activity on Bloomberg terminals and to appear in the Institutional Brokers Estimate System (I/B/E/S) database. The final sample contains 24,463 firm-quarter observations for 1068 firms over the period from 2011 to 2018.

3.2 Measuring abnormal institutional attention (AIA)

To measure institutional investors' attention, I follow Ben-Rephael et al. (2017) and measure abnormal attention based on news searching and news reading activity for specific stocks on Bloomberg terminals daily. Compared to the Google search volume index first proposed by Da et al. (2011), the news index from Bloomberg terminals can better capture institutional attention. After all, nearly 80% of Bloomberg terminal users are professionals from the financial industry, with the most common job titles being portfolio/fund/investment manager, analyst and trader (Ben-Rephael et al., 2017). This implies that most Bloomberg terminal users are likely to be institutional investors. Bloomberg ranks the news-searching behaviour of users for a specific stock by basing it on the number of times each article is read by its users and the number of times users search for news. Based on the news-searching and news-reading activity criterion, Bloomberg then assigns a score of 1, 2, 3 or 4 (from low to high uses) to each stock on a daily basis. To capture abnormal institutional attention, I follow Ben-Rephael et al. (2017) and generate a dummy variable that is equal to one if Bloomberg's score is 3 or 4 and zero otherwise. This treatment allows us to capture the right tail of the measure's distribution.

While the focus of this paper is to explore the impact of institutional attention on corporate investment decisions, it is important to note that news attracting institutional attention does not necessarily have to be investment-specific to enhance corporate investment efficiency. While news specifically addressing investment may direct attention to a firm's investment policy, attracting institutional investors' attention – even with non-investment-specific news – can spur increased monitoring. This enhanced monitoring effort plays a crucial role in disciplining managers and thereby improving a firm's investment efficiency. For instance, news about an inefficient compensation structure, even unrelated to investment policies, can capture investor attention, prompting intensified monitoring and leading to more efficient investment practices.

3.3 Measuring investment efficiency

Drawing from the literature, I measure investment efficiency by the sensitivity of investment expenditure to investment opportunities (e.g., Chen et al., 2007, 2011). Specifically, I measure investment opportunities using Tobin's Q, calculated as the sum of the market value of outstanding stocks and total liabilities divided by total assets on a quarterly basis. Investment expenditure is measured as a firm's quarterly capital expenditures scaled by the beginning-of-quarter total assets. Because the quarterly capital expenditure variables are reported on a year-to-date basis, I calculate the second, third and fourth quarterly values by subtracting the lagged value from the current value (see Kahle & Stulz, 2013).

3.4 Regression model and variables

Following prior studies on investment (Chen et al., 2007; Zuo, 2016), I also include the following set of control variables: $C{F}_{i,t}$ is the quarterly cash flow, measured as the sum of the income before extraordinary items, depreciation and amortisation, scaled by lagged total assets. I include $C{F}_{i,t}$ to control the well-documented effect of cash flow on investment (Abel & Eberly, 2011; Lamont, 1997). $\mathit{\text{Asset}}{1}_{i,t}$ is the inverse of quarterly total assets (i.e., 1/Asset). I include this variable to isolate the correlation between $\mathit{\text{IN}}{V}_{i,t+1}$ and $Q_{i,t}$ induced by the common scaling factor (i.e., ${\mathit{\text{Asset}}}_{i,t}$) following Chen et al. (2007).

I include control variables that are found to be correlated with corporate investment (Chen et al., 2017, 2011) or AIA (Ben-Rephael et al., 2017). For example, I include $\mathit{\text{Leverag}}{e}_{i,t}$, the ratio of the book value of debt to the book value of assets, to control for the impact of corporate leverage on investment efficiency. As documented by prior studies (i.e., Chen et al., 2011), firms with higher leverage may have more interest expenses and are less likely to obtain additional debt financing, both of which would limit their ability to invest. I also control for ${\mathit{\text{MrtCap}}}_{i,t}$, the natural logarithm of firms' market capitalisation, and ${\mathit{Age}}_{i,t}$, the age of the firm, since firm size and firm age are both related to the available resources that a firm can use for investment. Apart from these, I control for ${\mathit{\text{LnNumEst}}}_{i,t}$, the natural logarithm of the number of analysts covering the stock using the most recent information, as firms with greater analyst coverage are more likely to receive institutional attention (Ben-Rephael et al., 2017). Institutional ownership (${\mathit{\text{Inst}}}_{i,t}$) and earnings surprise (${\mathit{SUE}}_{i,t}$) are also included in the control set as they are important factors affecting the market reaction of an earnings announcement. Firm fixed effects and year-quarter fixed effects are added to control for fixed firm characteristics or time-varying characteristics. Standard errors are clustered by firm to mitigate the statistical concern arising from autocorrelated residuals (Petersen, 2009). All continuous variables are winsorised at the top and bottom 1% levels to reduce the impact of outliers.

3.5 Summary statistics

Table 1 presents the summary statistics of the main variables used in this study. The mean (median) value of AIA is 0.608 (1), similar to that reported by Ben-Rephael et al. (2017) (mean AIA = 0.620; median AIA = 1 around the earnings announcement day), suggesting that most firms receive abnormal institutional attention on the earnings announcement day. The mean and median values of other variables are consistent with prior studies (Ben-Rephael et al., 2017; Kahle & Stulz, 2013). For example, the mean (median) value of Q is 1.978 (1.659), similar to that reported in Chen et al. (2007) (mean = 1.81; median = 1.31).

Table 2 presents the pairwise Pearson correlations for the main variables used in the regression analysis. As shown in Panel B, AIA and Inv are significantly correlated with all of the control variables, suggesting that the control set is valid.

4.1 AIA and investment efficiency

Table 3 estimates the impact of AIA on investment efficiency using different models. Specifically, columns (1) and (2) estimate the influence of AIA on investment efficiency in the firm fixed effects model and industry-YearQuarter fixed effects model, respectively, following Bird et al. (2021). In column (3), additional control variables are added to the firm fixed effects model. I find that the coefficient on the interaction term (AIA*Q) is positive and significant across all specifications. This result is consistent with the managerial learning hypothesis that AIA encourages institutional investors to incorporate their information into the stock price, allowing managers to invest efficiently through learning from the price.

The economic magnitude of the improvement is also significant. The positive and significant coefficient of AIA*Q (coefficient = 0.0245; column (3)) implies that, on average, the investment of firms receiving abnormal institutional attention on the earnings announcement day is approximately 1.8 times more sensitive to their stock price than others. To explain this substantial economic magnitude, consider a one-standard-deviation increase in Q (1.885). This shock is associated with a greater investment for both firms that received abnormal institutional attention and those that did not. However, this impact is stronger for those who received abnormal institutional attention because their investment increased by 10.37% ($({\beta }_1+{\beta }_2$)*SD(Q) = (0.0305 + 0.0245)*1.885) on average. In contrast, the investment of firms that do not receive abnormal institutional attention increases by 5.75% (${\beta }_1$*SD(Q) = 0.0305*1.885) on average. As such, the investment of firms receiving abnormal institutional attention is 1.8 times (10.37%/5.75%) more sensitive to stock prices. This result is consistent with the literature. For example, Edmans et al. (2017) find that increasing the amount of outsider information in stock prices can lead to a 38% increase in investment-Q sensitivity; Foucault and Frésard (2012) propose that cross-listing experience doubles corporate investment-Q sensitivity, suggesting that information in financial markets can be essential to corporate investment decisions.

4.2 Investigating the channel

Thus far, the results suggest that AIA is associated with increased investment-Q sensitivity. In this section, I explore the channel through which AIA affects investment-Q sensitivity. According to the managerial learning hypothesis, managers can obtain useful information from stock prices and engage in more efficient investment if the price is informative (Bond et al., 2012; Chen et al., 2007). As such, I argue that AIA improves firms' investment-Q sensitivity by encouraging institutional investors to incorporate more information into the stock price, allowing managers to obtain useful information from the stock price to invest efficiently.

5.1 Instrumental variable regressions

To alleviate potential endogeneity from reverse causality or omitted variables, I perform instrumental variable analysis. Specifically, I use NoAnnouncement (number of earnings announcements) and EAIA (expected AIA) as instruments for AIA because they both related to institutional attention and are arguably not related to investment efficiency through any channels other than the attention channel. The choice of the first instrument (NoAnnouncement) is inspired by the idea that it captures the likelihood of institutional investors being distracted by the earnings announcement of other firms. This indicator takes the value of one if the number of competing announcements of a firm is above the firm-quarter median. Indeed, investors' reaction to an earnings announcement is weaker when earnings are announced on a day of many competing announcements due to limited attention (Hirshleifer et al., 2011). Thus, a higher value of NoAnnouncement implies a lower amount of attention being allocated to a specific firm (i.e., the correlation between AIA and NoAnnouncement should be negative). The second instrument (EAIA) refers to the median AIA across the previous four earnings announcements (Chiu et al., 2021). As documented by Ben-Rephael et al. (2021), attention to a firm's past earnings announcements can predict the attention to a firm's current earnings announcements. Therefore, a positive association between AIA and EAIA should be observed.

Column (1) of Table 6 presents the first-stage regression results. Consistent with my prediction, the correlation between NoAnnouncement and AIA is negative whereas the correlation between EAIA and AIA is positive. The significant Kleibergen-Paap rk LM statistic and Kleibergen-Paap Wald F-statistic reject the null hypothesis that the instrumental variables are weak or under-identified. The insignificant Hansen J statistic suggests that the instrumental variables are not subjected to overidentification. In the second-stage regression, I replace AIA with its fitted values predicted from the first stage and repeat the baseline regression. As shown in column (2) of Table 6, the coefficient of AIA*Q is still positive and significant, indicating that the finding is not driven by unobservable omitted variables or the reverse causality issue.

5.2 Exogenous return shocks from unrelated industries

To address concerns related to endogeneity and omitted variables in the findings, I further employ the exogenous return shocks from unrelated industries to estimate the causal relationship between AIA and investment efficiency. As highlighted in the study by Kempf et al. (2017), institutional investors tend to focus more on industries with extreme returns, be they exceptionally low or high.6

To test, I first estimated the industry return for each year-quarter based on the SIC3 industry classification. I defined an industry shock (IS) as a dummy variable, taking the value of one if the return of an industry in a given quarter was located in the top or bottom quartile across all SIC3 industries. Firms were then categorised into treatment and control groups based on the IS of unrelated industries. Specifically, treatment firms (Distraction = 1) were those from industries with non-extreme returns (IS = 0) because institutional investors of such firms were highly likely to be distracted by the extreme returns in other unrelated industries. Conversely, control firms were those from industries with extreme returns (IS = 1). For a more precise analysis, I further exclude those firms that themselves experienced extremely high or low returns (i.e., in the top or bottom quartile across all firms in a specific industry) as the impact of distraction would be less pronounced among such firms.

Table 7 reports the result of the test. The negative and significant coefficient of Distraction*Q indicates that institutional distraction negatively affects a firm's investment efficiency. Importantly, by employing an exogenous shock in the form of industry returns, these findings help alleviate concerns that the main results are driven by omitted endogenous factors.

5.3 Addressing sample selection bias

I perform two tests to address the sample selection bias issue. First, to address sample selection bias due to unobservability, I perform Heckman sample selection analysis. In the first stage, I use a probit model to predict the presence of abnormal institutional attention. Specifically, I regress AIA on NoAnnouncement and all the control variables from the baseline regression to predict the inverse Mill's ratio (MILLS). In the second-stage regression, I repeat the baseline regression with MILLS as an additional control variable to control for the self-selection effect. As reported in column (1) of Table 8, the coefficient of AIA*Q remains positive and significant after including MILLS in the regression. The coefficient of MILLS is significantly different from zero, implying the importance of controlling for potential selection bias.

To address sample selection bias that is occasioned by the correlation between AIA and observable firm characteristics, I also perform the baseline regression in the propensity score matching (PSM) setting. In the first stage, I use a logit model and regress AIA on CF, Leverage, Asset, MktCap, Age and LnNumEst to calculate the propensity score. I then match each treated unit (AIA = 1) with its neighbour (control group: AIA = 0) based on the propensity score. The matching results in a sample of 9893 firm-year observations. Appendix 2 reports the correlation between AIA and all other control variables before and after matching. The insignificant coefficients of the control variables in column (2) suggest that the control group and the treatment group share similar characteristics after matching.

After confirming that the matched sample is valid, I then rerun the baseline regression model on the matched sample. Column (2) of Table 8 reports the results. As reported in column (3), the coefficient of AIA*Q remains positive and significant, suggesting that the baseline finding is unlikely to be driven by the correlation between AIA and observable firm characteristics.

5.4 Other alternative explanations

The evidence is consistent with the idea that abnormal attention encourages institutional investors to incorporate their information into the stock price, which provides useful feedback for managers to invest efficiently. In this section, I perform several tests to rule out the alternative explanation that (1) the information contained in the stock price is from other sources (i.e., analysts, managers) and (2) the results are driven by institutional ownership.

5.5 Alternative measures

To ensure that my findings are not subject to measurement error, I rerun the baseline regression using alternative measures of abnormal institutional attention. Specifically, I follow Loughran and McDonald (2014) and measure abnormal institutional attention (AEDGAR) as the natural logarithm of the ratio of EDGAR on day t to EDGAR over the previous month. EDGAR refers to the daily number of unique requests for firm filings in the SEC EDGAR server. I obtain the data of the EDGAR filing request of 2011–2015 from the website provided by Loughran and McDonald (2017),7 and manually collect the request of 2016–2017 from the US Securities and Exchange Commission website.8 To better interpret the extent to which abnormal attention impacts investment-Q sensitivity, I first generate two dummy abnormal attention measures based on AEDGAR: (1) AEDGAR_P, a dummy variable that is equal to one if AEDGAR is a positive figure and zero otherwise – a positive AEDGAR suggests that firms received more filing download requests on day t compared to the previous month; (2) AEDGAR_MED, a dummy variable that is equal to one if a firm's AEDGAR is above the sample median and zero otherwise. After constructing the two dummy variables as alternative measures of abnormal institutional attention, I interact each of them with Q and rerun the baseline regression.

Additionally, as investment decisions may not be limited to specific days, such as earnings announcement days, I have repeated the baseline regression utilising a quarterly measure of institutional attention, denoted as AIAQTR. AIAQTR is a dummy variable taking the value of one when the quarterly median of the Bloomberg attention score for a firm falls within the range of 3 or 4, signifying abnormally high attention. In contrast, it is assigned a value of zero otherwise. Unlike the previous measure (i.e., AIA), which focused exclusively on the earnings announcement day, AIAQTR relies on the Bloomberg score for a given firm across the entire quarter. This broader scope accounts for institutional attention received by the firm throughout the quarter, which can mitigate the concern that investment decisions can be made at any time, not necessarily around earnings announcements.

Table 10 reports the results. I find that the results are consistent with the findings of the baseline regression, indicating that the findings are robust to these alternative proxies for abnormal institutional attention.

The residual of the above regression (InvInef) captures the deviation of current firm investment from its optimal level and thus measures the investment inefficiency of a firm. The negative coefficient associated with AIA in Appendix 5 indicates that higher levels of AIA can alleviate this investment inefficiency, consistently highlighting the positive influence of institutional attention on investment efficiency.