23 H2, we expect the coefficient b2 to be negative and b1 to be positive indicating that the
capitalised amount of RD is associated with less future economic benefits in more corrupt countries. We do not have an ex-ante prediction for b3 and b4.
In estimating all regressions equations 1 and 2, we add industry dummy variables based on the ICB Level 2 industry classification. Further, we also control for cross-sectional and
time series correlation by adding year fixed effects and clustering by firm. We winsorise all the continuous variables at the 1 percent level on both tails of the distribution. We report all
the variables employed in our models together with their source in Appendix II. Finally, for testing H4, we estimate the same regression across the sub-samples of low and
high internationalisation as described earlier. We then compare the size of the coefficients, and the significance of any difference, of the variable RDCapCorruption to infer the
differential influence of domestic corruption on the contribution of the capitalised development costs to future profitability across the two sub-samples.
4. Results
4.1 Descriptive statistics Table 2 shows descriptive statistics for corruption CPI, market development MrktDev and
level of enforcement Enforcement in the countries included in our sample. This reveals a range of values between all of the countries for corruption and in relation to the other country
relevant variables. For example, whilst focusing on countries with relatively large number of observations in the sample, Australia, Sweden, Denmark and Finland represented by 857
observations have the lowest levels of corruption ranging from 0.640 to 1.323. At the other extreme, Italy, South Africa, Greece, Poland and Spain have highest levels of corruption
ranging from 3.556 to 5.902 and show a total of 398 observations. Additionally, for the
24 same sets of countries, values of enforcement range from 32 to 52 for those countries with
low corruption levels and between 26 to 46 for the countries with higher levels of corruption depicting a lower level of enforcement. Similar variations exist in relation to market
development. TABLE 2 ABOUT HERE
Panel A of Table 3 shows descriptive statistics for all the variables used in our models for the full sample. These reveal, inter alia, that 46.6 of firm-year observations in our sample
capitalise some development costs while the remaining totally expense them in the income statement. While the capitalised development costs RDCap accounts for 1.3 of market
value on average, the expensed RD RDExp is around 5.5 of MV. The average firm-year observation in our sample has also a book value to market value of equity BM of 0.68,
shows a material RD intensity RDInt of 6.1, and has a a Leverage ratio of around 65.7.
TABLE 3 ABOUT HERE Panel B of Table 3 reports descriptive statistics for the dependent and independent variables
used in the multivariate analyses for the two sub-samples of Expensers and Capitalisers. The former has a mean median NI
of 1.087 0.911, while the latter exhibit a mean median future earnings of 1.35 1.02. T-test Mann-Whitney test indicates that the mean median
NI for Capitalisers is significantly higher p0.01 than the mean median for Expensers. The statistics are similar for NI2. The mean median RD expense RDExp for the
subsample of Expensers is 0.052 0.022, while the mean RDExp for Capitalisers is 0.059 0.020. T-test Mann-Whitney test shows that there is a strong statistically significant
difference across the two sub-samples. This may not be surprising given that a large number of firms appear to capitalise all development costs and expense zero amounts see Table 1.
25 The subsample of Capitalisers shows an average median RDInt of 0.059 0.021, while the
mean median of the corresponding amount for Expensers is 0.063 0.030 being significantly smaller based on a Mann-Whitney test p0.01. These descriptive statistics
indicate that, as expected, companies that capitalise some or all of the development costs are firms that are more RD intensive and have significantly higher future profitability.
Additionally, and as shown in prior literature e.g. Dinh et al., 2015; Oswald, 2008, T-test and Mann-Whitney test indicate that Capitalisers are smaller mean Size = 13.01 for
Expensers vs. mean Size = 12.614 for Capitalisers; p0.01, riskier mean Beta = 0.929 for Expensers vs. mean Beta = 0.980 for Capitalisers; p0.05 and more levered mean Leverage
= 0.591 for Expensers vs. mean Leverage = 0.733 for Capitalisers; p0.01.
4.2 Univariate analysis We present Pearson correlation coefficients between all variables in Table 4. The correlations
between the key variables of interest i.e., RDCAP, Corruption, NI and NI2 and other variables indicate the following. As hypothesized, the amount capitalised is positively and
significantly correlated with Corruption 0.072; p0.01. Additionally, as expected from evidence in prior literature, the amount capitalised is also positively and significantly
correlated with future earnings NI and NI2 exhibit a positive correlation at 1 level but also earnings benchmark beating PastBeat, ZeroBeat and BenchBeat, all exhibit a positive
correlation at 1 level, BM i.e., growth 0.257; p0.01, RDInt 0.236; p0.01, Leverage 0.046, p0.01, and Enforcement 0.094; p0.01.
Although these results depict a picture in line with our first hypothesis, they are unable to shed light on the remaining three hypotheses as they are based on a univariate correlation and
26 cannot bring into light the moderating effect of corruption and internationalisation. Thus,
results are further explored with multivariate analyses in the following section. TABLE 4 ABOUT HERE
4.3 Multivariate analysis Table 5 reports results for multivariate analysis testing the effect of corruption on the
magnitude of development costs capitalised. Models 1, 3 and 5 differ from 2, 4 and 6 respectively only for the measures used to proxy earnings benchmark beating.
Focusing on the full sample, the results support H1: firms in countries with higher levels of corruption capitalise higher amounts of development costs. In fact, the coefficient for
Corruption is positive as expected and statistically significant across both models 1 and 2 always at the 1 level. Confirming univariate analysis, our results also indicate that the
amount of capitalised development costs is positively influenced by RD intensity RDInt, growth BM, internationalisation IntSalesPerc and leverage Leverage. In addition, we
report that Size negatively affects the amount of development costs capitalised. Further, our multivariate analysis confirms that amount of development costs capitalised is significantly
affected by earnings management, as the coefficients for the measures derived from Dinh et al. 2015 are always positive and statistically significant.
Models 3 to 6 report the multivariate analyses testing H3. We report that both for firms with low and high international exposure, Corruption is positively and significantly
associated with the amount of development costs capitalised. However, tests comparing the size of the coefficients of the variable Corruption across all models indicate that these are
significantly higher for the sub-sample of firms with lower levels of internationalisation. This is in support of H3.
27 TABLE 5 ABOUT HERE
Table 6 reports results for multivariate analysis testing the moderating role of corruption in the relationship between RD and future earnings for the full sample but also across the sub-
samples of firms with lower and higher levels of internationalisation. These analyses test hypotheses H2 and H4. As previously, Models 7, 9 and 11 differ from 8, 10 and 12
respectively only for the measures used to proxy earnings benchmark beating. These multivariate analyses, firstly, confirm the rationale that capitalised development
costs are mirrored in future economic benefits as the coefficient for RDCap is positive and statistically significant across all model specifications. Moreover, this analysis illustrates the
moderating role of corruption in the relationship between RD and future earnings, supporting H2. In both models 7 and 8 the coefficient for the interaction between RDCap and
Corruption is negative statistically significant at the 5 level suggesting that the association between capitalised development costs and future earnings is lower in more
corrupt countries. Additionally, expensed RD RDExp is also positively correlated with future earnings again at the 1 level. The latter suggests that there is still an element in the
amounts expensed that contributes to future earnings. TABLE 6 ABOUT HERE
Secondly, the results across the sub-samples of firms with higher and lower levels of internationalisation provide a clearer picture of the role of domestic corruption. The tests
across Models 9 and 12 reveal that the contribution of capitalised development costs to future earnings when corruption is higher is present only for the sub-sample of firms with
lower levels of internationalisation. These results are in support of H4. The combined results of these tests and those presented in Table 5 indicate the following.
Although corruption does have an influence on the amount capitalised by companies, the
28 expected future earnings is not as high as expected when domestic corruption is high. This
finding is associated with the sub-sample of firms with low internationalisation. For firms with high internationalisation, domestic corruption does not impair the benefit from the
amounts capitalised, suggesting that these amounts genuinely represent future economic benefits which are mirrored in companies
’ future profitability. In fact, the positive influences of the expensed amounts revealed for the full sample is driven only by this sub-sample of
firms.
4.4 Discussion of findings On reflection of our hypotheses and informed by prior literature, the following inferences can
be drawn from the results presented. Whilst the level of development cost capitalisation is positively associated with a number of firm-level characteristics, for instance, RD intensity,
growth and leverage it is also influenced by corruption as a country characteristic. More specifically, we show a strong significant association between country corruption and levels
of development cost capitalisation H1 and, further in such cases, the lower the contribution of the capitalised development costs to future earnings H2.
Such findings are broadly consistent with that strand of the accounting literature prior to the mandatory of IFRS that finds that accounting choices, such as capitalisation, are
associated with earnings management, for instance to avoid reporting a loss, to achieve earnings smoothing or benchmark beating Cazavan-Jeny Jeanjean, 2006; Cazavan-Jeny et
al., 2011; Dinh et al., 2015; Markarian et al., 2008. For example, Cazavan-Jeny et al. 2011, p. 146 contend that their
‘...results are therefore consistent with managers potentially capitalising RD expenditures to achieve certain financia
l reporting objectives’. Moreover, more corrupt environments are associated with lower quality of accounting Doupnik, 2008;
29 Han et al., 2008; Nabar Boonlert-U-Thai, 2007 conducive to earnings management Fan et
al., 2014; Picur, 2004; Riahi-Belkaoui, 2004 such as the aggressive use of capitalisation. It is due to concerns of such accounting, and the potential manipulation of earnings, that US
GAAP requires all RD costs to be expensed SFAS 2. Inherent with capitalisation is the signalling effect, based upon hitherto proprietary
information, regarding the strength of future earnings. Overall, the literature is more mixed as to the affect of capitalised costs on future earnings. As Ahmed and Falk 2006 summarise,
‘when a firm capitalises expenditure and reports the amount as an asset in its financial statements, it signals good news. The capitalisation decision suggests that, in the reporting
entity’s judgment, the capitalised expenditure is expected to yield benefits to the entity…likely to be realized in the foreseeable future’ p. 232 see Shah et al., 2013; Wolfe,
2012. However this is in contrast to the more negative findings regarding the relationship between capitalisation and future earnings for instance Cazavan-Jeny and Jeanjean 2006
for instance Cazavan-Jeny Jeanjean, 2006. Our findings reveal a consistency of over- capitalisation, associated with high levels of corruption, and its impaired ability to generate
stronger future earnings, effectively sending noisy or misleading signals regarding future earnings.
Furthermore, as a firm becomes more international, levels of home country risk factors, such as that associated with high corrupt environments, may diminish as the firm becomes
more exposed to international norms and levels of scrutiny reducing potential managerial discretion in accounting choice Sandholtz Gray, 2003 Our findings in relation to H3
confirm that as a firm becomes more international, the influence of home country corruption on capitalisation is mitigated. Such a finding is consistent with the literature on domestic and
international accounting choices such as earnings management. For instance, Prencipe 2012,
30 p. 693 finds that
‘multinationals tend to carry out less income-increasing earnings management than domes
tic firms’. Further, Lang, Lins, et al. 2003 and Lang, Raedy, et al. 2003 find that international firms, through cross-listing, are more transparent, and less
aggressive in accounting choices, than domestic firms. It follows from H3, and the preceding findings of H1 and H2, that the influence of home country corruption on the contribution of
capitalised costs to future earnings is mitigated with levels of internationalisation. These findings, examining corruption as a country factor influencing accounting choice
provide new insights into the accounting literature, where such research is sparse Houqe Monem, 2016 despite the recognized link between corruption and earnings opacity. Further,
whilst Dinh et al. 2015examined RD capitalisation, that study did not shed any light on the determinants of capitalisation from an international context, nor did it consider the
potential and now revealed significant influence of corruption.
5. Sensitivity analyses