Chapter 5: Findings and Discussion on the Value Relevance of Fair Value Hierarchy
5.5 Additional analyses and robustness checks
As pointed out by Song et al. (2011), the differences in the value relevance across firms could reflect differences in firm-specific characteristics, such as asset size, rather than differences in fair value levels. This might be the case if, for example, large firms tend to hold level 1 and level 2 fair values, while small firms rely more on level 3 fair values. Therefore, as a robustness check, this study tests the value relevance of fair value hierarchy for two sub-samples: small and large firms. Another reason for this segmentation is that information asymmetry between managers
178
and shareholders is likely to be higher for small firms (Atiase, 1985; Freeman, 1987). The small group was defined as firms with average total assets less than the whole sample median and large firms as those with average total assets greater than the sample median, which corresponds to 94 financial firms (349 observations) in the former group and 91 financial firms (350 observations) in the latter.
As shown in Table 5.8 Panel A for small firms, the estimated coefficients on non-fair values as well as level 1 and level 2 fair values are statistically significant, indicating their value relevance. Unlike the entire sample, the valuation coefficient on level 3 fair values is not statistically different from 0. That is, investors in small firms do not perceive fair value estimates based on unobservable inputs as relevant for valuation purposes. One possible reason for this result could be the higher information asymmetry associated with small firms. The valuation coefficients on both level 1 and level 2 fair values are higher than that on level 3 fair values, suggesting that investors in small firms depend more on fair value amounts with observable inputs for valuation purposes. The valuation coefficients on both level 1 and level 2 are significantly higher than non- fair value net assets. Finally, the valuations of non-fair values and level 3 fair values are not statistically different. Moving to Panel B on large firms, both non-fair value and fair value amounts are value relevant to investors. In terms of the differences in the value relevance across fair value levels, large firms show a pattern comparable to that of the entire sample: the valuation coefficient on level 1 fair values is significantly higher than that on level 3 fair values, whereas it is not statistically different from that on level 2 fair values. The valuation of level 2 fair values is higher than that of fair values at level 3, however the difference is not statistically significant. The three levels of fair values are more value relevant than non-fair value amounts.
For additional insight into the results on the value relevance of fair value hierarchy, this thesis also partitions the banks in the main sample into two groups: high versus low Tier 1 capital ratio banks. This partition is based on the findings of prior research indicating that managerial
179
discretion in banking industry is used to avoid violating regulatory capital requirements (Ramesh and Revsine, 2001; Shrieves and Dahl, 2003; Paananen et al., 2012). Interestingly, Nissim (2003) shows that banks with low Tier 1 ratios tend to overstate, to a large extent, their fair values of loans in order to affect the market perception of their risk and performance. Investors might consider the potential unreliability of fair values in valuing banks with low capital ratios. Relating to fair value hierarchy, Song et al. (2010) posit that the differences in capital ratio might be correlated with managersβ choice of fair value valuation levels. Furthermore, Goh et al. (2015)
argue that banks with lower capital adequacy might be forced to liquidate their positions, even though their assets might be sold at fire-sale prices. This is particularly the case in times of financial crisis, which is the study period in this thesis. Investors are more likely to discount the fair value estimates given the greater likelihood of being forced to sell their assets at unfavourable prices, especially those measured based on unobservable inputs (Goh et al., 2015).
This additional analysis is limited to financial firms engaging in traditional banking activities (138 banks in the whole sample) since the regulatory capital requirement are relevant only to banks. Banks are classified as low Tier 1 group (69 banks, and 262 observations) when having Tier 1 ratio below the sample median, and as high Tier 1 banks (69 banks with 263 observations) when their Tier 1 ratio is greater than the sample median.100
Table 5.9 Panel A shows that results for banks with low Tier 1 ratios. The valuation coefficients on non-fair values in addition to level 1 and level 2 fair values are statistically significant, suggesting their value relevance to investors. The valuation coefficient on level 3 fair values is not statistically significant. That is, investors consider level 3 fair values reported by banks with low Tier 1 ratios as not sufficiently reliable to be reflected in firm value. The valuations of level 1 and level 2 fair values are significantly higher than that of fair values at level 3. This might be explained by reliability concerns about fair value estimates for banks close to the minimum
180
regulatory capital ratio (Nissim, 2003; Paananen et al., 2012). Moreover, the valuations of the three levels of fair value are significantly different from those of non-fair value amounts. Turning to the results for banks with high Tier 1 ratios in Panel B of Table 5.9, the estimated coefficients on both non-fair value and fair value amounts are significantly different from 0, indicating their value relevance to investors. Interestingly, the valuations across the three levels of fair value are not statistically different. This suggests that investors in banks with high capital ratios perceive fair values based on unobservable inputs as value relevant as those measured using observable inputs. One explanation could be the lower incentive for managers to use accounting discretion for the purpose of capital management (Ahmed et al., 1999; Nissim, 2003; Vyas, 2011).101 Finally, with the exception of level 2 fair values, the valuation coefficients on fair value amounts are significantly greater than that on non-fair values.
101 However, some have argued that the impact of fair value adjustments alone on determining banksβ regulatory
181 Table 5. 8 Value relevance of fair values hierarchy for large versus small firms
Panel A Small Firms Panel B Large Firms
VARIABLES Coeff. Test F-stat p-value VARIABLES Coeff. Test F-stat p-value
NFVNA π1 0.181*** FVNA1= FVNA3 4.69 0.0311** NFVNA π1 0.105*** FVNA1= FVNA3 8.37 0.0041***
(0.0428) (0.0263)
FVNA1 π2 0.361*** FVNA1= FVNA2 0.16 0.6931 FVNA1 π2 0.296*** FVNA1= FVNA2 2.28 0.1323
(0.0661) (0.0424)
FVNA2 π3 0.315*** FVNA2= FVNA3 4.48 0.0349** FVNA2 π3 0.228*** FVNA2= FVNA3 0.99 0.3207
(0.0667) (0.0519)
FVNA3 π4 -0.199 FVNA1= NFVNA 4.03 0.0454** FVNA3 π4 0.164*** FVNA1= NFVNA 34.21 0.0001***
(0.237) (0.0477)
EPS π5 -0.484 FVNA2= NFVNA 4.29 0.0391*** EPS π5 2.640*** FVNA2= NFVNA 6.74 0.0098***
(0.822) (0.965)
Constant π0 6.329*** FVNA3= NFVNA 2.53 0.1129 Constant π0 4.029 FVNA3= NFVNA 4.94 0.0270**
(1.674) (2.664)
Year dummy π·π‘ Yes Year dummy π·π‘ Yes
Observations 349 Observations 350
No of firms 94 No of firms 91
R-squared 0.769 R-squared 0.656
Notes:Robust standard errors in parentheses. *, **, *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. The table reports the OLS estimation of the following equation πππ‘= π0+ π1ππΉπππ΄ππ‘ + π2πΉπππ΄1ππ‘ + π3πΉπππ΄2ππ‘+ π4πΉπππ΄3ππ‘+ π5πΈππππ‘ + Ξ΄π·π‘+ πππ‘, where πππ‘ is the market value per share of firm i three months following the end of fiscal year t. πΉπππ΄1ππ‘, πΉπππ΄2ππ‘ and πΉπππ΄3ππ‘are fair value of level 1, level 2 and level 3 net assets for firm i as reported at the end of fiscal year t. πΈππππ‘ is the reported net income of financial firm i for the fiscal year t and π·π‘ is a dummy variable for year t. All accounting information is scaled by the number of outstanding common share. This table shows the regression results of partitioning the sample by total assets of firms.Financial firms are classified into small or large firms based on the median value of total asset for the entire sample. The sum of observations in the two sub-samples (349 + 350 = 699) equals the total number of observations shown in Table 5.4.
182 Table 5. 9 Value relevance of fair values hierarchy for low versus high tier 1 banks
Panel A Banks with low Tier 1 Ratio Panel B Banks with High Tier 1 Ratio
VARIABLES Coeff. Test F-stat p-value VARIABLES Coeff. Test F-stat p-value
NFVNA π1 0.0829*** FVNA1= FVNA3 32.03 0.0001*** NFVNA π1 0.204*** FVNA1= FVNA3 0.31 0.5784
(0.0126) (0.0297)
FVNA1 π2 0.328*** FVNA1= FVNA2 0.41 0.5215 FVNA1 π2 0.342*** FVNA1= FVNA2 1.44 0.2317
(0.0496) (0.0347)
FVNA2 π3 0.367*** FVNA2= FVNA3 30.47 0.0001*** FVNA2 π3 0.260*** FVNA2= FVNA3 3.44 0.0649*
(0.0652) (0.0529)
FVNA3 π4 -0.0436 FVNA1= NFVNA 36.20 0.0001*** FVNA3 π4 0.376*** FVNA1= NFVNA 9.71 0.0020***
(0.0327) (0.0512)
EPS π5 1.082** FVNA2= NFVNA 32.98 0.0001*** EPS π5 2.561*** FVNA2= NFVNA 1.53 0.2172
(0.484) (0.889)
Constant π0 2.612 FVNA3= NFVNA 15.93 0.0001*** Constant π0 6.133* FVNA3= NFVNA 32.34 0.0001***
(1.819) (3.326)
Year dummy π·π‘ Yes Year dummy π·π‘ Yes
Observations 263 Observations 262
No of firms 69 No of firms 69
R-squared 0.768 R-squared 0.764
Notice:Robust standard errors in parentheses. *, **, *** Indicate statistical significance at the 0.10, 0.05, and 0.01 levels (two-tailed), respectively. The table reports the OLS estimation of the following equation πππ‘= π0+ π1ππΉπππ΄ππ‘ + π2πΉπππ΄1ππ‘ + π3πΉπππ΄2ππ‘+ π4πΉπππ΄3ππ‘+ π5πΈππππ‘ + Ξ΄π·π‘+ πππ‘. where πππ‘ is the market value per share of firm i three months following the end of fiscal year t. πΉπππ΄1ππ‘, πΉπππ΄2ππ‘ and πΉπππ΄3ππ‘are fair value of level 1, level 2 and level 3 net assets for firm i as reported at the end of fiscal year t. πΈππππ‘ is the reported net income of financial firm i for the fiscal year t and π·π‘ is a dummy variable for year t. All accounting information is scaled by the number of outstanding common share. This table provides the regression results of partitioning the banks in the sample by Tier 1 ratio.Banks are classified to either low or high Tier 1 ratio based on the median value of Tier 1 ratio for the entire sample. The sum of observations in the two sub-samples (263 + 262 = 525) is lower than the total number of observations (699) shown in Table 5.4, because the analysis in this table is restricted to financial firms whose primary business is to engage in traditional banking activities and have data on Tier 1 ratio in BankScope.
183
As a sensitivity test, this study re-estimates the main model to test the value relevance of fair hierarchy and scales all the variables using lagged total assets in lieu of the number of shares outstanding. A number of accounting studies use lagged total assets to mitigate the potential scale related-effect of the price model (see, for example, Marquardt and Wiedman, 2004; O'Hanlon and Taylor, 2007; Manganaris et al., 2015). As shown in Table 5.10, both non-fair value and fair value amounts are value relevant to investors since their estimated coefficients are significantly different from 0. Furthermore, the valuation coefficients on FVNA1 and FVNA2 are statistically different from that FVNA3 (F-statistics: 4.58, F-statistics: 4.80, respectively). In other words, the empirical analysis after scaling all the variables by the total assets indicates that the value relevance of level 3 fair values, based on managerial projections and other unobservable inputs, is significantly lower than that on level 1 and level 2 fair values, measured using observable inputs. The valuation of level 1 fair values is not different from that of level 3 fair value net assets. Unlike the main model results, the valuation coefficients on fair value amounts appear to not significantly differ from that on non-fair values. Overall, Table 5.10 shows that the main results hold to a large extent after scaling the variables by lagged total assets in the model employed. Finally, the present study runs the regression to test whether investors place differential weights across the three levels of fair value using the market value of equity six months after the end of the fiscal year. Some prior studies use stock prices six months after fiscal year-end as the dependent variable in the price model (e.g. Liu et al., 2011; Barth et al., 2012). Table 5.11, the main findings of the empirical analysis are not substantially altered after employing the market value six months, rather than three month, following the end of fiscal year.
184
Table 5. 10 Value relevance of fair values hierarchy using alternative scaling method
Pane A Panel B
VARIABLES Coeff. Test F-stat p-value
NFVNA π1 0.807*** FVNA1= FVNA3 4.58 0.0326**
(0.152)
FVNA1 π2 0.986*** FVNA1= FVNA2 0.78 0.3788
(0.158)
FVNA2 π3 0.856*** FVNA2= FVNA3 4.80 0.0288**
(0.162)
FVNA3 π4 0.475** FVNA1= NFVNA 2.61 0.1069
(0.206)
EPS π5 3.846*** FVNA2= NFVNA 0.49 0.4849
(1.453)
Constant π0 -1.448 FVNA3= NFVNA 3.60 0.0582*
(52.76)
Year Dummy Yes
Observations 699
No of firms 185
R-squared 0.713
Notes: Robust standard errors in parentheses. *, **, *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels (two- tailed), respectively. The table reports the OLS estimation of the following equation πππ‘= π0+ π1ππΉπππ΄ππ‘ + π2πΉπππ΄1ππ‘ +
π3πΉπππ΄2ππ‘+ π4πΉπππ΄3ππ‘+ π5πΈππππ‘ + Ξ΄π·π‘+ πππ‘. where πππ‘ is the market value of firm i three months following the end of fiscal year t. πΉπππ΄1ππ‘, πΉπππ΄2ππ‘ and πΉπππ΄3ππ‘are fair value of level 1, level 2 and level 3 net assets for firm i as reported at the end of fiscal year t. πΈππππ‘ is the reported net income of financial firm i for the fiscal year t and π·π‘ is a dummy variable for year t. In this form of the model all the market and accounting variables are scaled by lagged total assets (i.e. total assets for firm i as reported at the end of fiscal year t-1). While Panel A shows the results of the regression, Panel B offers F-statistics testing the differences between the estimation coefficients.
In particular, both non-fair values and the three levels of fair values are value relevant to investors. The valuation coefficient on level 1 fair values is significantly greater than that on level 3 fair values. Again, this indicates that investors place less valuation weight on level 3 fair values, which is subject to potential managerial manipulation and measurement error in comparison to level 1 fair values measured based on observable inputs from active markets of identical assets or liabilities. The valuation coefficient on level 2 fair value is greater than that on level 3 fair values, yet the difference is not statistically significant. The valuation coefficient on each of three fair value levels is higher in magnitude than that on non-fair value net assets, suggesting that they are more value relevant than non-fair value amounts.
185
In sum, the results are largely unaffected when the estimation of the value relevance of fair value hierarchy employs the market value of equity six months after the end of the fiscal year as a dependent variable.
Table 5. 11 Value relevance of fair values hierarchy using six months market value
Panel A Panel B
VARIABLES Coeff. Test F-stat p-value
NFVNA π1 0.0584*** FVNA1= FVNA3 19.18 0.0001***
(0.0115)
FVNA1 π2 0.190*** FVNA1= FVNA2 3.91 0.0484**
(0.0173)
FVNA2 π3 0.144*** FVNA2= FVNA3 2.22 0.1365
(0.0249)
FVNA3 π4 0.0979*** FVNA1= NFVNA 77.33 0.0001***
(0.0220)
EPS π5 1.694** FVNA2= NFVNA 14.13 0.0002***
(0.708)
Constant π0 5.778*** FVNA3= NFVNA 8.07 0.0046***
(1.246)
Year Dummy Yes
Observations 699
No of firms 185
R-squared 0.574
Notes: Robust standard errors in parentheses. *, **, *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels (two- tailed), respectively. . The table reports the OLS estimation of the following equation πππ‘= π0+ π1ππΉπππ΄ππ‘ + π2πΉπππ΄1ππ‘ +
π3πΉπππ΄2ππ‘+ π4πΉπππ΄3ππ‘+ π5πΈππππ‘ + Ξ΄π·π‘+ πππ‘. where πππ‘ is the market value per share of firm i six months following the end of fiscal year t. πΉπππ΄1ππ‘, πΉπππ΄2ππ‘ and πΉπππ΄3ππ‘are fair value of level 1, level 2 and level 3 net assets for firm i as reported at the end of fiscal year t. πΈππππ‘ is the reported net income of financial firm i for the fiscal year t and π·π‘ is a dummy variable for year t. All accounting information is scaled by the number of outstanding common share. The sample includes 699 firm-year of 185 distinct firms in the European Economic Area (EEA) and Switzerland over the period 2009-2012. While Panel A shows the results of the regression, Panel B offers F-statistics testing the differences between the estimation coefficients.
5.6 Conclusion
The second empirical part of this thesis examines the value relevance of the three levels of fair values as disclosed under IFRS 7. In particular, it examines whether there are variations in the valuation weight placed by investors on fair value net assets across the three levels of fair value hierarchy. In a further analysis, it investigates whether the value relevance of the three levels of fair value hierarchy depends on the country-level institutional environment and firm-level corporate governance mechanisms.
186
The results of this study suggest that investors place a higher valuation weight on mark-to-market fair values net assets, based on observable inputs, relative to mark-to-model fair values, estimated based on unobservable inputs. In particular, the valuation coefficient on level 3 fair value is lower in magnitude than that on level 1 and level 2 fair values; yet the difference is statistically significant only for level 1. Overall, these results are comparable to the empirical findings of prior accounting studies focusing on US markets (Kolev, 2009; Song et al., 2010; Goh et al., 2015). Furthermore, the value relevance of fair value hierarchy is examined as a function of the institutional environment and corporate governance. The results reveal that the value relevance of level 3 fair values tend to be lower for firms domiciled in countries characterised by weak institutional environment and for firms with weak corporate governance mechanisms. The impact of the institutional environment and corporate governance on the value relevance of level 1 and level 2 fair values appear to be statistically insignificant. These findings provide support to the view that both the institutional environment and corporate governance play a significant role in mitigating information asymmetry problem associated with fair value amounts estimated based on managerial projections and private information.
As a further analysis, the entire sample is split using the sample median of total assets into two groups: small and large firms. The results for small firms show that level 3 fair values are not value relevant to investors, and the valuations of both level 1 and level 2 fair value net assets are significantly higher than that on level 3 fair values. For the group of large firms, the results of the whole sample tend to be confirmed; the valuations on level 1 and level 2 fair value net assets are greater than that on level 3 fair values, the difference is significant only for level 1.
Additionally, using the sample mean of capital Tier 1 ratio, the banks in the sample are classified into: low Tier 1 versus high Tier 1 ratio banks. The results for banks with lower Tier 1 ratios show that the valuation coefficient on level 3 fair values is not statistically significant and significantly lower than those on level 1 and level 2 fair values. Interestingly, the valuations on
187
the three levels of fair values hierarchy are not significantly different in magnitude for the sub-