url
25
Apr
2014

ON THE DETERMINANTS OF DERIVATIVE HEDGING BY INSURANCE COMPANIES: EMPIRICAL RESULTS

Unlike the above studies which use single-year observations, our research employs cross-sectional and panel data over three years in the life and non-life insurance sectors. The results obtained from the panel data estimation can then be compared with those from the cross-sectional estimation. Lagrange Multiplier (LM) and Hausman tests are carried out to determine the most appropriate model. The former is used to examine the relative efficiency of the heterogeneous panel data models (one-factor fixed/ random-effects models) against the homogeneous pooled OLS estimation. The latter is employed to determine which panel data model should be used in the study if the computed LM test statistic argues in favor of panel data models (Hausman, 1978).
A White test for heteroscedasticity is performed. If the null hypothesis of homoscedasticity is rejected at the 0.05 level, then heteroscedasticity-corrected estimates are reported.
As regards multicollinearity, we compute variation inflation factors (VIFs) for the independent variables. All the VIFs are less than 2. According to Studenmund (2001, p. 258), the multicollinearity is severe if the largest VIF value among all independent variables exceeds five. Based on this rule-of-thumb, it seems that multicollinearity is not a problem in this study.
Univariate Results
To examine whether there is any difference in the firm characteristics of derivative users and nonusers in the life and non-life sectors, we conduct equality tests of means, medians, and variances. For robustness, two tests are employed for each statistic. The results are shown in Table III. To simplify the exposition, we focus on results that are significant at the five percent level in both tests. There are greater differences between derivative participants and non-participants in the life sector than in the non-life sector. In the life sector, reinsurance dependence differs between derivative users and nonusers. In the tests for means and medians, both firm size and currency exposure are found to differ between the two groups. As to the non-life sector, the users and nonusers are different in firm size and mismatch of asset and liability durations, based on the tests for mean and variance. Significant results obtained from these univariate tests only indicate that there might be a difference between derivative users and nonusers as regards a particular firm characteristic, holding other firm characteristics constant. We now carry out a multivariate analysis. First, we reveal the current practices of off-balance-sheet transactions in the Taiwanese insurance industry. Second, we use panel data modeling to overcome certain data and research method-based limitations such as the inability to control for unobservable differences across individual insurers that might influence derivative use. Thus, this paper should be of value to academics, practitioners and financial regulators.
The remainder of the paper is organized as follows. Section II reviews the existing literature and develops testable hypotheses. Section III describes the data and methodology employed. Section IV presents our empirical findings and conducts robustness checks. The final section concludes the paper and provides some implications of our work. this

Table III Mean. Median and Variance Equality tests

m CSzarac й risrics Over all users Nonusers Equality tasts of median
JV Median JV Median JV Median Wlcoxon Mann*Whined tie-adj.) Г d-value i Adj. Med. Chi-sq>ire \p- ’-a lue]
ZE 72 15.8108 9 16.2152 63 15.6844 2.6222 [0.0087] 2.0317 [0.1540]
IN 72 174.4879 9 250.1076 63 173.9782 0.3065 [0.7592] 0.0000 [1.0000]
3SA.ST 74 0.0000 11 0.0000 63 0.0000 0.3028 [0.7621] 0.0179 [0.8936]
JSLIA 74 23.6491 11 59675 63 24.3298 0.1458 [0.8841] 0.0000 [1.0000]
ROWTH 74 0.3145 11 0.4108 63 0.3083 0.2973 [0.7662] 0.0000 [1.0000]
E3NS 74 479776 11 37.6366 63 48.4100 1.8089 [0.0705] 0.4271
LCF 74 0.0000 11 0.0000 63 0.0000 0.6729 [0.5010] 0.0589 [0.8083]
H 72 0.8344 9 0.8344 63 0.8344 0.1065 [0.9152] 0.0318 [0.8584]
H 72 0.0000 9 0.0000 63 0.0000 0.8282 [0.4076] 0.0000 [1.0000]
К 74 0.7749 11 29652 63 0.4059 2.3036 [0.0212] 1.7085 [0.1912]
t 72 0.0152 9 0.0193 63 0.0138 0.9024 [0.3668] 05079 [0.4760]
Et 71 4.1237 10 35401 61 4.1267 0.7194 [0.4719] 0.0000 [1.0000]
EEFC 71 0.43 T6 10 0.4816 61 0.4244 0.9475 [0.3434] 05100 [0.4751]

Unlike the above studies which use single-year observations, our research employs cross-sectional and panel data over three years in the life and non-life insurance sectors. The results obtained from the panel data estimation can then be compared with those from the cross-sectional estimation. Lagrange Multiplier (LM) and Hausman tests are carried out to determine the most appropriate model. The former is used to examine the relative efficiency of the heterogeneous panel data models (one-factor fixed/ random-effects models) against the homogeneous pooled OLS estimation. The latter is employed to determine which panel data model should be used in the study if the computed LM test statistic argues in favor of panel data models (Hausman, 1978). A White test for heteroscedasticity is

About The Author

Kevin J. Brandon

Home | Site Map | Contacts

Copyright © 2013 - 2019 Investment And Finance Online. All rights reserved