Model Design

Lin and Liu (2012) measure the multi-period effect in reverse regression as Basu (1997), and employ multi-period annual returns as proxies of good (bad) news to measure the asymmetric timeliness of earnings. Comparatively, in the event-based approach, I measure the impact of specific news by using short-window returns as proxies of good (bad) news because the annual returns might contain information of news in whole year and obscures the information of specific news. To consider the multi-period effect, I refer to regression in Warfield and Wild (1992), who use regression to measure the explanatory power of earnings over multi-period in explaining specific returns. The following equation is to examine impact of speed of news recognition itself on news recognition and asymmetric timeliness in multi-period under conservatism (i.e. H1, H2, H3.a and H3.b).

21

RETi, t = α0Di,t + β0Ei,t + β1∑ Ei,t+n +γ0Et ×Di,t +γ1∑ Ei,t+n × Di,t

+δ0Fi,t + ζ0Fi,t×Di,t +θ0 Ei,t ×Fi,t + θ1∑ Ei,t+n ×Fi,t

+ λ0Ei,t ×Di, t× Fi,t + λ1∑ Ei,t+n ×Di,t×Fi,t

RETi,t are short-window (three-day) stock returns of firm i at year t. Ei,t are the earnings of firm i at year t, deflated by the starting stock prices of the year. ∑ Ei,t+n

are the aggregating earnings of firm i in mutli-period. I use different lengths of periods as proxies of future periods. If max n =1,2,3 or 4 , the aggregate earnings are EPSt+1, (EPSt+1 + EPSt+2), (EPSt+1 + EPSt+2+ EPSt+3) or (EPSt+1 + EPSt+2+ EPSt+3+ EPSt+4). Di,t denotes dummy variables of RETi,t, if RETi,t < 0 (bad news), Di,t equals 1;

instead, Di,t equals 0 (good news). Fi,t denotes dummy variables of Future Group, if news is classified in Future Group, Fi,t equals 1; instead, Fi,t equals 0.

To measure the news recognition in multi-period, I employ Ei,t and ∑ Ei,t+n

as proxies of earnings in current (short-term) period and in future (long-term) periods.

Aggregate earnings are employed as variable rather than multi-period earnings for two reasons. (i) The regression is to measure the recognition of news in short-term and long-term periods, rather than recognition of news in each period. (ii) As time periods increase, the noise of news in earnings-returns relation increases, but the information of news is less recognized. Hence, aggregate earnings are employed as

22

proxies to reduce the noises in the earnings-returns relation.

The information of news in Future Group is recognized in longer-term periods, and it’s difficult to measure the information in future. Hence, news in Future Group contains more uncertainties than in Current Group. Due to the uncertainties of news in Future Group, it’s natural that firms take more conservative way (i.e. more asymmetric timeliness of earnings) to recognize news in Future Group. Namely, the bad news in Future Group is recognized earlier, and the good news in Future Group is recognized more slowly than in Current Group to avoid overstatement of net income.

In the following, I measure the different earnings responses (DERs) to good (bad) news between Future Group and Current Group, and explain the changes in DERs in multi-period.

The coefficient (θ + λ) denotes the DERs of bad news between Future Group (β + θ + γ + λ) and Current Group (β + γ) in different periods. Due to uncertainties of news in Future Group, firms would pre-recognize bad news earlier than in Current Group to avoid overstatement of net income. Namely, the information of bad news in Future Group is recognized more than in Current Group in current period. Hence, the coefficient in current period (θ0 + λ0) is considered positive.

In long-term periods, the earnings responses to bad news in Future Group decay

23

more than in Current Group as time periods increase, because the information of bad news in Future Group is recognized more in short-term periods than in Current Group.

Hence, the coefficient (θ1 + λ1) would decreases and turns negative in future periods.

To test H1, I expect that DERs of bad news between Future Group and Current Group are positive in current period, and decrease in future periods.

The different earnings responses (DERs) of good news between Future Group (β+ θ) and Current Group (β) are denoted by value (θ). Firms need more time to verify the good news in Future Group, and recognize good news in Future Group in longer-term periods than in Current Group. Hence, good news in Future Group is recognized less than in Current Group in short-term periods, and the coefficient in current period (θ0) is considered negative in short-term periods.

As time periods increase, information of good news in Future Group is gradually recognized, and thus earnings responses to good news in Future Group increase in long-term periods. Comparatively, the information of good news in Current Group is recognized in shorter-term periods, and thus earnings responses to good news in Current Group gradually decay in long-term periods. Due to the increase and decrease in recognition of good news in Future Group and Current Group, the earnings responses to good news in Future Group would be more than in Current Group in

24

long-term periods. Hence, the DERs in future periods (θ1) to good news between Future Group and Current Group are considered positive. To test H2, I expect that DERs of good news between Future Group and Current Group are negative in current period, and turn positive in future periods.

It’s natural that firms take more conservative accountings (i.e. more asymmetric timeliness) to recognize news in Future Group, because the news in Future Group contains more uncertainties than in Current Group. Namely, the bad news in Future Group is recognized earlier, and the good news in Future Group is recognized more slowly than in Current Group to avoid overstatement of net income. The reverse moving direction (increase or decrease) in recognition of bad news and good news magnifies the asymmetric timeliness in Future Group in short-term periods. The asymmetric timeliness of earnings in Future Group (γ+ λ) would be more than in Current Group (γ) in current period. I define the difference in asymmetric timeliness of earnings in Future Group and Current Group as magnification effect. To test H3.a, I expect that magnification effect (λ0) in Future Group is positive in current period.

Considering the multi-period effect, the magnification effect in Future Group decays in future periods. The information of bad news in Future Group is recognized more in short-term periods. Hence, the earnings responses to bad news decay more in

25

Future Group than in Current Group in long-term periods. In another aspect, information of good news in Future Group is gradually recognized in long-term periods, and thus the earnings responses to good news in Future Group are more than in Current Group. Due to the increase and decrease in recognition of good news and bad news in Future Group, the asymmetric timeliness in Future Group decreases more than in Current Group. To test H3.b, the magnification effect (λ1) in Future Group is expected to decrease compared with λ0.