Earnings Surprise, Portfolio Inertia and Stock Price Volatility

3.1 Hypothesis

Earnings surprise depend largely on the accuracy of the investors’ expectation on the company’s future earnings. The higher the degree of information ambiguity, the less is the accuracy of earnings expectation, and the greater is the earnings surprise^[6,24]. In a mature market, this accuracy mainly comes from two ways: The first is the relevance of the company’s future earnings and its past public financial information; the second is investors’ forecast ability on the company’s future earnings. In A-stock market, however, the factors that influence the earnings surprise are much more complicated, and the phenomenon of incomplete information is much more serious, such as the immature market, defects of system design, inadequate disclosure of information of listed companies, and the limited ability of investors to obtain and digest information, etc. Therefore, it is common to find that the listed company’s actual earnings are above or below expectations when the company releases the earnings announcement. In this context, we propose the following hypotheses:

H1: It is often observed the existence of earnings surprise in A-share market.

Generally, institutional investors comply with the asset portfolio theory to manage their portfolio investments. They decide whether to adjust portfolios through all the information they collect and process. There are a variety of factors influence this decision, such as earnings surprise, transaction costs, heterogeneous beliefs, emergencies, etc.

H2: After controlling for other factors, the greater the earnings surprise, the larger the magnitudes for institutional investors to adjust their portfolios.

Institutional investors have gradually dominated A-stock market, enabling them to impact share prices significantly. Assuming other factors constant, institutional investors actively adjust their portfolios will lead fluctuations on related stock prices. Specifically, largely reducing shares may produce negative abnormal returns, while largely increasing shares may produce positive abnormal returns. On the other hand, if institutional investors hold their portfolio constant or adjust slightly, that is, to present a portfolio adjustment inertia, the related stock price volatility should be relatively small.

H3: In terms of the earnings surprise of listed companies, the stronger the inertia of institutional investors to adjust their portfolios, the smaller is related stock price volatility, and vice versa.

H4: (Existence of the mediation effect): If the hypothesis 2 and 3 are well established, then institutional investors’ portfolios adjustments will impose a significant mediation effect on the relationship between earnings surprise and share price volatility.

3.2 Data

Because of the relatively small amount and scale of institutional investors in China before 2007, we choose all A-share listed companies over the period 2007–2012 as the initial sample and then exclude some observations according to the following standards:

1) Remove the listed companies without a full fiscal year.

2) Eliminate the financial sector shares, ST’special Treatment’shares, and companies with cross-listings.

3) Remove the samples with major assets reorganization during the sample period.

4) Eliminate observations in which the percentage share of institutional quarter-end stock holdings is less than 5% of the total shares outstanding during the sample period.

5) Exclude observations in which the number of institutional investors among the quarter-end top 10 unrestricted stakeholders is less than two during the sample period.

6) Exclude securities with less than one year of listing in order to reduce the possible IPO effect.

7) Eliminate abnormal and missing data from the sample.

8) Continuous variables are winsorized at the 1% level and 99% level in order to control the effect of extreme values.

In this paper, we consider every regular report from each listed company as an independent observation and get a total of 20143 valid observations. The report involved is between the first quarter of 2007 and the third quarter of 2012 and the data covered is for the period December 1, 2006, through October 31, 2012. We obtain the data from Wind database and CSMAR database.

This paper defines earnings announcement day as the event day. As the disclosure time requirements for the quarterly, semi-annual and annual report are different, and the earnings announcement date for each sample also varies, we set the different event window in Table 1 to reasonably examine the change of each observation in the window period, especially to reduce the impact of data overlap.

3.3 Variables

1) Earnings surprise

In this paper, we follow the method of Yu and Wang^[6] to measure accounting earnings (E) using the rate of total return on assets. Under the normal circumstance, if company’s information quality is high, there is slightly change in rate of return on assets every year. The formula of earnings surprise (UE) for stock i at time τth is

2) Portfolio adjustment by institutional investors

The institutional investors consist of securities investment fund, QFII, social security fund, insurance fund, company annuity, trust, financial companies, brokerage, securities financial products, etc. Because it is difficult to obtain each institutional investor’s quarter-end holdings for each equity at each regular report, we use the aggregate holdings of each security owned by all institutional investors as an alternative portfolio, and the change in the proportion of institutional shareholdings within two consecutive regular reports is considered as the adjustment of its asset portfolio. If the change is significant, it is said to be portfolio adjustment activeness; otherwise, it is called portfolio adjustment inertia. Our observations are based on the Wind database of “the proportion of the unrestricted shares outstanding owned by all institutional investors. The formula for stock i at th period in institutional investors’ portfolio adjustment is

3) Stock price volatility

In order to remove the effect of systemic factors on the individual stock prices, we use market adjustment model to calculate the abnormal rate of return so as to more objectively examine the influence of institutional portfolio adjustment on individual stock prices:

where t stands for trading day, p is stock price, p_m represents the market index, AR_it, R_it and R_mt are the abnormal rate of return, real rate of return, and normal rate of return for stock i at time t, respectively. The normal rate of return for main board stocks in Shanghai and Shenzhen stock market is calculated as the Shanghai composite index, while normal rate of return for small and medium-sized market, and growth enterprise market (GEM) is calculated as the small and medium-sized board index.

The average abnormal rate of return for stock i at τth period in the window period (t₁ ~ t₂)

The common way to measure stock price volatility is standard deviation of stock daily rate of return during a long period (usually a quarter). The stock price volatility for stock i at τth period in the window period (t₁ ~ t₂) is expressed as the standard deviation (σ) of abnormal rate of return:

4) Control variables

With reference to the related literature, this paper selects the following control variables, which are shown in Table 2.

4.1 Statistics Analysis

1) Institutional shareholdings

Table 3 shows the institutional shareholdings during the sample period. It can be seen that the number of institutional shareholders in each listed company has fluctuation, but the proportion of institutional stockholdings has increased steadily. Specifically, average number of institutional shareholders in each listed company increased from 19.63 in 2007 to 31.44 in 2012, and average proportion also increased from 22.32% in 2007 to 37.65% in 2012. This shows that Chinese stock market has experienced the transformation from individual investors dominance to institutional investors dominance, and the behavior of institutional investors have a more pronounced impact on individual stocks.

2) Descriptive statistics of variables and correlation tests

Table 4 shows the descriptive statistics of the variables. The mean, median value and standard deviation for all the sample of UE are -0.03%, -0.1% and 2.44%, respectively. This suggests that earning surprise is widely existed among Chinese listed corporations, and the wider distribution. This, to some extent, supports Hypothesis 1. The standard deviation of 10.62 for ΔIO indicates that the difference of institutional investors to adjust their portfolios is very significant. The minimum level of 0.49 and maximum level of 25.92 for the total sample of VOL suggest that it is notable for the difference on the individual stock price volatility in the sample period. For control variables, the mean of E is 3 and its standard deviation is 3.93, stating that the average profitability for the sample stocks is quite low, but the variation is significant. The mean value of IO is 37.42, showing that the average proportion of sample stocks owned by institutional investors is relatively high. This is consistent with the results shown in Table 3. The standard deviation of IO is up to 19.93, revealing the large difference on the proportion of institutional holdings for the sample stocks. The sample stocks tend to be overvalued, which can be seen from average value of 0.33 for BM. The standard deviation of TRAC is 10.84, showing that there is large variation on sample stocks’ trading costs. The mean level of Dpro is 0.05, which suggests that the percentage of listed companies engaging in large stock dividends is relatively low. However, the distribution of profits is only presented in the annual or semi-annual reports, resulting in the decreased value of Dpro, which is selected from quarterly observations. The mean of 1.69 and standard deviation of 1.96 for TURV show that abnormal turnover ratio for sample stocks are relatively low, but the difference is still significant. Apparently, this does not indicate the low turnover ratio for A-share. Due to the logarithmic mode, the standard deviation of SIZE is only 0.49, leading to the smaller differences in asset size. The mean of 0.21 for FORC show that there are relatively fewer listed companies releasing earnings pre-announcements. This can be attributed to China’s relevant policies and laws. More specifically, it is the listed companies with significant changes in earnings that require issuing earnings pre-announcements. Seen from the change in LAG, we find that the number of listed companies to issue an earnings announcement on a regular basis will increase with the passage of time.

In addition, the Pearson correlation coefficient test shows that three core variables (UE, ΔIO and VOL) are significantly positively correlated and the estimated coefficient is less than 0.5. (Due to space limits, we omit the table for the correlation coefficients). This suggests that there do not exist apparent multicollinearity among core variables.

3) Statistics of core variables classified by earnings surprise

In order to observe earnings surprise, and changes in institutional investor’s portfolio adjustments as well as stock price volatility, we classify the sample according to two criteria. One is to put the positive UE sample into “good group” and negative UE sample into “bad group”. The other is to rank UE in an ascending order and then equalize them into ten parts. Among them, the closer to the two ends of the percentile, the larger is the changes in earnings surprise.

We can see from Table 5, no matter what classification used, it is common for the presence of earnings surprise, and again it supports Hypothesis 1. The M-W test results show that, due to the average value, there is no significant difference on the proportion adjustment of institutional stockholdings between good and bad information groups, but the high volatility and significant difference on the proportion adjustment are found during the window period. This suggests that, compared to the “good information group”, institutional investors show little sign of reducing “bad information group”. That is, for stocks with declining earnings surprise, institutional investors show the inertia of portfolio adjustment.

To further examine the classification of SUE from the lowest to the highest, we can see that when UE increases from -4.82 to 0.07, ΔIO rises from 0.47 to the highest level of 1.89; and when UE rises from 0.07 to even further, ΔIO falls and eventually reduces by 0.23. From the trend of change, we find that ΔIO exhibits the feature of “inverse U-shape” curve (that is, to rise first, and then to fall). This suggests that as earnings surprise increase, institutional investors gradually purchase more stocks, but when it reaches the highest level, the increase in purchasing stocks is getting smaller and smaller with the further growth of earnings surprise, which manifests the inertia of portfolio adjustment. Again, this reflects the different trading features of individual investors from institutional investors, since the latter is more rational and able to inverse trade timely. Meanwhile, with the rise of UE, VOL increases from the lowest group of 2.38 to the highest group of 2.63 (albeit with fluctuation). It means that stock price volatility is becoming greater with the rise of earnings surprise. Combined the variability of UE, ΔIO and VOL, we can tell that institutional investors prefer steady-going business; before stock prices rise to the high levels, they keep purchasing more shares to steadily push up the trading volatility; after the high levels, the increasing volatility apparently cannot be attributed to institutional investors, but mainly to the non-rational trading of individual investors. Due to the portfolio adjustments inertia, institutional investors usually do not respond fully to large earnings surprise. In contrast, because of the disposition effect, individual investors prefer to continue holding stocks when earnings surprise is negative and stock prices fall, while they prefer to sell shares to push up stock price volatility when earnings surprise is positive and stock prices rise.

4.2 Regression Results and Analysis

In terms of the mediation effect proposed in [25] and their test procedures, this paper constructs the following regression models to examine the relationship among earnings surprise, institutional investors’ portfolio adjustment and share price volatility:

where, Z¹, Z² and Z³ are corresponding control variables respectively, μ¹, μ² and μ³ are corresponding error measures, respectively.

1) Regression results and analysis of Model (5)

Model (5) is used to investigate whether institutional investors adjust their portfolio positions in line with earnings surprise of listed companies. Table 3 shows that there is an annual increase in the proportion of institutional shareholdings, so this paper divides the total sample into six sub-samples based on the year. Following the method of [26], we select previous values of BM, TRAC, Dpro, E and IO as control variables in Model (5). With expectation, if institutional investors adjust their portfolios according to the changes in earnings surprise, then α₁ should be positive. That is, the larger the earnings surprise of listed companies, the greater is the scales of portfolio adjustment for the institutional investors. Regression results are shown in Table 6.

First, we analyze the regression results of the total sample. The second column of Table 6 shows that regression coefficient of UE is 0.07 with significance at 5% level, indicating that the institutional investors purchase more shares as earnings surprise of the listed companies becomes larger. That is, earnings surprise encourages the institutional investors to adjust their portfolios. This is consistent with expectations and supports Hypothesis 2. The regression coefficient of BM is 2.88, meaning that the higher the book-to-market ratio of the company, the more likely institutional investors tend to increase their shareholdings, that is, institutional investors choose companies with undervalued assets. For TRAC, its regression coefficient is 0.14. It seems that institutional investors inclined to buy more stocks with higher transaction costs, which is contrary to common sense. In fact, TRAC represents the average share price during the window period. On one hand it reflects the transaction cost; on the other hand it also shows the value of the stock. From this perspective, the increase in share price is in line with increase in institutional investors shareholdings. The coefficient of -4.00 for Dpro shows that institutional investors are more likely to reduce shares with large stock dividends. This finding supports the optimal sequence financing theory: Institutional investors believe that large stock dividend conveys information of over-valued share price. The regression coefficient for UE is -0.22, which reveals that the greater the earnings of the previous period, the more likely institutional investors tend to sell their shares. Since current earning is the best estimate for earnings of the next period, institutional investors are supposed to increase their holdings. Obviously, this is not consistent with our expectations. We believe that the proportion of irrational individual investors in Chinese stock market is relatively high. They tend to be easily influenced by Herding Effect, and over-react to good news with high earnings stocks, therefore, pushing up share prices as a result. While the rational institutional investors take advantage of the irrationality of individual investors to reverse the operation and sell their high earnings stocks. In Table 5, the inverse “U” shape curve characteristics for ΔIO and UE also prove this point. The coefficient of -0.14 for IO suggests that if institutional investors hold more shares in the previous period, they are more inclined to decrease their shares in the current period, which is shown in Table 5.

Second, we analyze regression results for the sub-sample. The coefficient for SUE is not stable, and the results of the year 2008, 2010, 2011 and 2012 are negative but not significant, showing that institutional investors do not significantly adjust their portfolio according to earnings surprise. To some extent the result verifies the existence of institutional investors’ portfolio adjustment inertia. Further analysis shows that the year which portfolio adjustments exhibit inertia, is when Chinese A-share market experienced a decline. Therefore, we believe that the essence of portfolio adjustments inertia is institutional investors’ disposition effect: In years when the stock market is falling, institutional investors still hold plenty of stocks with accounting losses and do not reduce their shares rapidly. In the year of 2007 and 2009, the coefficients for UE are significantly positive and it is these two years when there is a sharp increase in the A-share market. This shows that when the market as a whole is better off, institutional investors actively adjust the asset portfolio according to changes in earnings surprise. The regression coefficients for the control variables are almost consistent with results for the total sample.

Table 6 and its analysis show that if institutional investors express some activeness in portfolio adjustments during a rising stock market, then Hypothesis 2 is well established; instead, if institutional investors do so during a falling market, then Hypothesis 2 will not be established. Institutional investors, to some extent, exhibit portfolio adjustment inertia as they do not significantly adjust their portfolio according to variation of earnings surprise. Basili and Fontini^[22], Asano^[23] and Illeditsch^[8] regarded the ambiguity as necessary premise to form the asset portfolio inertia. Before the regular earnings announcements of listed companies, future accounting earnings are considered as the ambiguity information. Therefore, this paper not only provides strong empirical support for the related theory of asset portfolio inertia, but also, for the first time, concludes that the formation of portfolio adjustment inertia is subject to market conditions.

2) Regression results and analysis of Model (6)

In Model (6), we investigate the premise for the existence of mediation effect: The significance of β₁ is the main criterion of telling whether mediation effect exists. Referred to related literature, the choice of the control variables is slightly different from Model (5). As institutional investors show portfolio adjustment inertia in a declining market, we will further divide the total sample into inertia group and non-inertia group. Results are shown in Table 7.

Table 7 shows that the coefficients β₁ for UE are positive at 1% significant level in both the total sample and sub-sample, which satisfies the premise of mediation effect tests^[25]. After joining the control variable, β₁ is still significant but with a smaller value, and the size of adjusted R2 becomes larger. This shows that the inclusion of control variables reduces the influence of earnings surprise on share price volatility, but improves the fitting degree of the model. Except for FORC and LAG from the non-inertia group, the coefficients for other control variables are significant and consistent with expectations.

3) Regression results and analysis of Model (7)

The results of Model (5) support Hypothesis 2, and those of Model (6) prove that the premise for the existence of mediation effect is established. Model (7) is used to test Hypotheses 3 and 4, and the significance of coefficient γ₁ for ΔIO is the core variable to tell the existence of mediation effect. The results are shown in Table 8.

From Table 8, we can see that adjusted R² is between 0.11 and 0.50, indicating the strong explanatory power of variables. Comparing the inclusion of control variables to the exclusion of them, we find that the signs of correlation coefficients are the same, but the model fitting degree is improved significantly, suggesting that the selection of control variable is reasonable.

First, we analyze the regression results for total samples. The coefficient γ₁ for ΔIO is 0.0035 with significance at 1% level, showing that ΔIO has a remarkable effect on VOL, after controlling for other factors. Apparently, the influence of earnings surprise on the stock price volatility can be exerted by institutional investors’ portfolio adjustment. This is basically accordance with the conclusion of Yue and Zhou^[12] that fluctuation in proportion of institutional shareholdings is the most important source for volatility of the Shanghai Composite Index. Further analysis of Table 4 shows that the coefficient for ΔIO is the minimum when that for UE is the maximum, while the corresponding coefficient for VOL is at its maximum level. This illustrates that VOL is not completely dominated by institutional investors. In the sample period, individual investors still remain important to have a significant impact on share price volatility. Therefore, partial mediation effect is imposed by the trading of individual investors, which can be seen from the coefficients of UE. The coefficient of γ₂ is 0.5724, which is positively significant and greater than γ₁.

Second, we compare the regression results of inertia group to those of non-inertia group and conclude that hypothesis 3 is well established. The regression coefficient of ΔIO is insignificant in inertia group, but is positively significant in non-inertia group. This difference shows that: For inertia group, institutional investors inactively adjust their portfolio according to earnings surprise of listed companies, therefore, no remarkable volatility is found in share prices. For the non-inertia group, institutional investors actively adjust portfolio to significantly enhance share prices volatility. Hence, with respect to earnings surprise of listed companies, the stronger inertia the portfolio adjustment by institutional investors, the smaller fluctuations of related stock prices, and vice versa. Thus, Hypothesis 3 is established. Combined with the results of Tables 6, 7 and 8, we can see that, the characteristics of institutional investors’ portfolio adjustments (inertia or activeness) exert significant effects on related stock price volatility, based on earnings surprise of listed companies. Therefore, the existence of the mediation effect is strongly proved and hypothesis 4 is established.

Finally, we examine the regression results for the control variables. The coefficients for TRAC, TURV and Dpro are positively significant, suggesting that the higher the transaction costs, the stronger the investor sentiments, the more likely the listed company engages in large stock dividends, the greater is the share price volatility. The coefficients for BM, SIZE, FORC fand LAG are negatively significant, indicating that the higher the book-to-market ratio, the greater the size of company assets, the more easier is to stabilize the stock prices. The listed companies that issue pre-announcement earnings present relatively low share price volatility, it shows that pre-announcement earnings releases part of the profit information. The more delayed the date for actual earnings announcement, the less the impacts of announcement on stock price volatility. Since investors have more time to collect and process information before earnings announcements, they can predict earnings more reasonably and respond in advance. The similar results of whether to control for the industry factors indicate that the difference between industries has no influence on stock price volatility. Moreover, share price volatility has eliminated the systematic effect of the market, therefore, the results are barely affected by whether to control for the annual factors, such as the macroeconomics.

4.3 Robustness Test

First, this paper defines earnings surprise (UE) as dummy variables (Due) in two ways: One is to categorize by the signs, if UE is positive, then we assume a value of 1 for Due1; if UE is negative, then we assume a value of 0 for Due1. The other is to divide UE into ten equal parts in an ascending order and Due2 takes the values from 1 to 10 respectively. Second, we replace UE in Models (5), (6) and Model (7) with Due1 and Due2 respectively to run the tests. The same test results are obtained as the previous tests, thus, our conclusion supports the robustness of above empirical results (Limited to space, robustness test results are not reported here; if necessary, please contact the author).