Failure Threats of Insurance Companies: A Case Study of Financial Environments of Jordan

Insurance firms are known to have unique financial failure characteristics that affect the financial environment of the countries. Therefore, the purpose of this study is to assess the validity of the model used in predicting the financial failures of insurance companies. The model is believed to help in stabilizing the financial environment of the countries by predicting any collapses in the insurance sector. A discriminate regression technique was used to test 28 indicators chosen from 11 financial failure model parameters. 11 parameters of the model are the following: solvency, profitability, operational capabilities, structural soundness, capital expansion capacity, capital adequacy, reinsurance and actuarial issues, management soundness, capital expansion capacity, earnings and profitability, and liquidity. The results of the study proved that 22 variables from 11 parameters were significant; the study also validated the use of the financial failure model as a stable predictor of the financial failure of ASE insurance firms. The stability of the insurance industry is interpreted through the minimum deviation between the real and measured performances. The deviation was present in 3 out of 95 observations, and it affected only 3 firms out of 19, 1 firm out of 3 turned out to be affected by the risker deviation which is the type II error distorted observation. To conclude, the study by mentioning that insurance firms are not threatened by failure or distress and the financial failure model is a valid prediction model.


INTRODUCTION
It was reported in the 1930s that a few public Jordanian firms were listed in the counter market and were able to trade shares. The first corporate bond was issued in the 1960s, which was then followed up by the Amman Financial Market (AFM) launch in January 1978. It is an organized exchange that includes 66 listed firms. A major development then occurred in the Jordanian Capital Market on March 11, 1999, where three new institutions named Jordan Securities Commission (JSC), Amman Stock Exchange (ASE), and Securities Depository Center (SDC) replaced the AFM. The establishment of the Jordan Insurance Association for insurance companies boosted the Kingdom's economy. The Jordan Insurance Federation (JIF) was then established in 1989 following a Royal Decree, which comprises licensed 24-member insurance companies; they consist of one foreign insurance company, two 'takaful' operators, and 22 conventional insurers, 20 out of 24 companies are listed in the Amman Stock Exchange, which gives a clear indication that the market lacks the services of reinsurance companies. Consequently, the operations of the Jordanian insurers were reinsured by preserving a portion of the risk, which was ceded through the reminder of the Arab and foreign reinsurance companies.

LITERATURE REVIEW
In the last seven decays, scholars and researchers heavily discoursed the topic of the financial failure of companies that belongs to different exchanges sectors around the globe. In this literature review section, high-ranking journals are listed and summarized chronologically.
To begin with, Altman (1968) developed Z-Score model and Altman et al. (1977) developed ZETA credit risk model for assessing financial distress. Moreover, effective indicators and predictors variables of the corporate distress were specified and quantified. Altman et al. (2007) classified distress firms into original samples; it was shown that the discriminant-ratio model provided an early warning capacity up to four years before the financial distress. Both statistical analysis, artificial intelligence, and neural networks can be used to assess the financial distress probability. Bose and Pal (2006), Jardin and Severin (2012), Gestel et al. (2006), Lin et al. (2014), Carlos (1996), Chen and Du (2009), Gepp and Kumar (2015), Bae (2012), Wilson and Sharda (1994), Jo et al. (1997), and Li and Sun (2009) showed that the artificial intelligence and neural networks outperformed the traditional statistical techniques in predicting the financial distress around the globe. Kumar and Ghimire (2013) indicated that qualitative factors play a vital role in the financial soundness of insurance firms. However, they do not grasp the big picture of the insurer's financial health. It was also shown that the financial performance of each company and their aggregated overview fall under the following dimensions: capital adequacy, assets quality, reinsurance, and actuarial issues, management soundness, earnings and profitability, and liquidity. Halpern et al. (2009) observed that the firms that avoided financial distress or bankruptcy have a more critical influence on debt composition than the governance proposed changes. Tinoco and Wilson (2013) showed the utility of combining ac-counting, market, and macro-economic data for listed companies in financial distress prediction models. Dairui and Jia (2009) proved that there is a relationship between 13 variables (corporate governance variables, agency costs, and ownership structure, etc.) and the probability of financial distress. Zmijewski (1984) showed that the slack-based measure (SBM) data envelopment analysis (DEA) model has obvious advantages in predicting corporate financial stress, and categorized firms into safe, grey, and distress zones by proposing cut-off points. Das et al. (2003) proposed the compilations and usages of the key indicators in the surveillance of the insurance sector firms' financial soundness. Furthermore, there is other qualitative information that could be also used in the surveillance process such as the ownership arrangements.
Christidis and Gregory (2010) developed a new "dynamic logit" model (micro and macro variables) for prediction failures in the UK. Manzaneque et al. (2016) confirmed a negative relationship between board size and financial distress, while the ownership concentration does not have any significant impact on the Spanish financial distress. Amendola et al. (2015) found that micro-economic indicators and firm-specific factors influence the exit of Italian firms through bankruptcy, liquidation, and inactivity routes. Smajla (2014) showed that composite companies in Croatia have the best capital adequacy, retention ratio (due to reinsurance services), management, and profitability. Nonetheless, their liquidity still needs some improvements. Jahur and Quadir (2012) identified that rate adequacy, sales trends, indebtedness, management capability, financial planning, etc. of the Bangladesh SMEs are the cause of financial distress problems. It was also illustrated that fund management and resource crunch, poor accounting system, poor financial control, poor productivity and profitability, and management succession are the causes of financial distress. Geng et al. (2015) observed that financial performance and profitability indicators play an important role in the prediction of the profitability deterioration in China. Sevim et al. (2014) showed that Turkey's economy uses macroeconomic indicators to predict crises a year before it actually happens. Ntoiti (2013) indicated that financial management practices, human resource management practices, and corporate governance practices have a significantly negative relationship with financial distress, while government regulations have a positive relationship with the financial distress in Kenya. Simpson and Damaoh (2008) found that in Ghana, the CARMEL model was the most comprehensive compared to the other evaluation models and tools.
Following the literature review, this study develops a financial failure model that helps in achieving the purpose of the paper by evaluating the emerging markets' financial distress, testing the significance of the indicators, and examining the applicability of the model. The study is considered to be unique as there is little availability of similar research in this area of knowledge and it also contributes to the decision-making processes of ASE and Jordan Insurance Federation (JIF).

HYPOTHESES DEVELOPMENT
Since 28 indicators are being tested, the hypotheses apply for each one of the indicators H1: There is no significant effect of the independent variable on the financial failure score (solvency margin as a dummy variable (1 if solvency ratio > = 150%; 0, otherwise)).
H2: There is a significant effect of the independent variable on the financial failure score (solvency margin as a dummy variable (1 if solvency ratio > = 150%; 0, otherwise)).

METHODOLOGY
The sample used in this study is the annual data, which includes 19 financial statements of the listed companies. One company out of the 20 listed companies was excluded due to one year of miss-ing data. The data covers a period of 5 years from 2014 to 2018. Moreover, the statistical analysis used 95 panel data observations.
The parameters (indicators) of the financial failure model are solvency (total liabilities/total assets, total liabilities/total shareholders' equity, net operating cash flow/total liabilities); profitability (operating profit margin, earnings before income tax/total assets, earnings before income tax/ shareholder equity); structural soundness (fixed assets/total assets, current assets/total assets, shareholders' equity/fixed assets); business development capacity (total assets of this year/total assets of last year, net profit of this year/net profit of last year); capital expansion capacity (net profit/ number of ordinary shares at the end of year, net assets (total assets-total liabilities)/number of ordinary shares at the end of year, capital reserves/ number of ordinary shares at the end of year, net increase in cash and cash equivalents/number of ordinary shares at the end of year); earnings and profitability (loss ratio, expense ratio, return on asset, return on equity); operational capabilities (sales revenue/total assets, sales revenue/fixed assets); reinsurance and actuarial issues (retention ratio, net technical reserves/net claims paid at the end of the year, net technical reserves to net realized premiums); management soundness (asset per employee (total assets/number of employees)); capital adequacy (surplus/technical reserves ratio, solvency ratio (net written premium/total equity)); liquidity (Graham rating = (2* equity)/total liabilities). In addition, Table 1 also shows the measurements of 28 indicators and their references. This model is also used by other industries to deal with financial distress; it combines the parameters and indicators from the CARMEL model.

Financial failure model
Financial failure score (solvency margin is a dummy variable (1,0)) = α +β 1 L/A +β 2 L/E +β 3 OCF/ L+β 4 OPM +β 5 EBT/A +β 6 EBT/E +β 7 R /A +β 8 R/ FA +β 9 A 1 /A 0 +β 10 P 1 /P 0 + β 11 FA/TA +β 12  The dependent variable for the financial failure model is a dummy variable, which takes value 0 when a company is in financial distress and 1, otherwise, a company is considered as "distressed" when its solvency margin is less than 150% in each year according to Internationally Active Insurance Groups & Reinsurance Standards Instructions and Amendments.
Solvency: a performance indicator variable that indicates the insufficient funds of fulfilling stakeholders' debts in the long and short run (Geng et al., 2015).
1) L/A: total liabilities/total assets (since all the firms are exposed to the same market conditions the ratio should be kept within the industry levels; there is the tradeoff between the tax shield of debt and the equity of shareholders, which should be kept at a safety margin). Capital adequacy: it is considered as an ultimate risk to the financial stability of an insurer stems, because of writing business that is intense in volume, volatile or undetermined result (Das et al., 2003). Liquidity: it is a term that refers to the loss of confidence in an insurer, usually results in a cancel over, return of unexpired premium, or seeking insurance elsewhere.

Statistical techniques
The technique used in this study is the discriminant analysis; it is an analysis that constructs predictive models for group memberships. Based on the linear combinations of the predictor variables, groups are discriminated at their best, and discriminant functions are composed (or, for more than two groups, a set of discriminant functions). The known sample of cases for the group memberships is used at the beginning to generate the functions. Later on, the functions could be used for new cases that have unknown group memberships and known measurement of the predictor variables.

RESULTS
The testing results of the financial distress model showed that there are 95 valid cases without missing or out-of-range group codes and/or at least one missing discriminating variable. Table 1 includes 76 cases at a good level and 19 cases at a poor level. It also reveals that the mean of the successful level is higher than the mean of the distress level for the following parameters: According to Table 2 the main null hypothesis can be rejected; there is no significant effect for the following variables: L/A, L/E, OCF/L, EBT/A, EBT/E, FA/TA, CA/TA, EPS, NA/OS, CR/OS, SR/ TR, SOL, RET, A/NEM, and GRAH (Significant Wilks' Lambda at 1% (P-value < 1%)), R/A , ROA (Significant Wilks' Lambda at 5% (P-value < 5%)) and OPM, TR/C, E/FA, LOSR, and ROE (Significant Wilks' Lambda at 10% (P-value < 10%)). Thus, alternative hypothesis can be accepted. On the other hand, in regards of the other variables: R/FA, A1/A0, P1/P0, NIC/OS, TR/RP, and EXPR (insignificant Wilks' Lambda (P-value > 10%)), we cannot reject the null hypothesis. Therefore, there is no significant effect on financial failure score. Table 3 shows that there is no correlation between 28 variables. Only 44 correlations coefficients between the variables were either above + 0.5 or below -0.5, and the rest of 378 correlation coefficients aren't highly correlated with one another.     Table 4 shows the log determinate values, it was calculated for the successful and total but not for distress because it had only 19 observations, which also lead to undetermined equality of population covariance matrices values for distress, successful and total. Unfortunately, that comes at a disadvantage to the result claims. Moreover, the values of the successful and total were almost the same. Therefore, there is not clear evidence that the results will not be affected.  Table 5 is divided into multiple parts. To begin with, in Part A, an Eigenvalue of 3.804 was calculated, which is an indication that the function explains only 3.804 of the variances of the company's performance (dependent variable). It is also a good indication of the fitness of the model (high Eigenvalue is associated with better fitness). Furthermore, it was also tabulated that the canonical correlation has a value of 0.89, which is a great value since it has a high effect size (0.89^2 = 0.796).

L/A L/E OCF/L OPM EBT/A EBT/E R/A R/FA
In Part B, it is proved that the prediction model is statistically significant since the P-value is < 1%. In Part C, the predictors are ranked from the best to the according to the standardized canonical discriminant function coefficients (TR/RP, SR/TR .302*** x NA/OS (*** means significance at 1%, ** means significance at 5%, * means significance at 10%). Significant variables are ranked from best predictor to the worst one.
To further elaborate, 22 significant indicators in predicting the financial failure, which are shown in the formula above, are the following: profitability (operating profit margin, earnings before income tax/total assets, earnings before income tax/ shareholder equity); operational capabilities (sales revenue/total assets, sales revenue/fixed assets); solvency (total liabilities/total assets, total liabilities/total shareholders' equity, net operating cash flow/total liabilities); structural soundness (fixed assets/total assets, current assets/total assets); and capital expansion capacity (net assets (total assets-total liabilities)/number of ordinary shares at the end of year, capital reserves/number of ordinary shares at the end of year). While the significant CARMEL parameters were: reinsurance and actuarial issues (RET. ratio, net technical reserves/ net claims paid); management soundness (asset per employee); earnings and profitability (loss ratio, ROA and ROE); capital adequacy (surplus/ technical reserves ratio and solvency ratio (net written premium/total equity); capital expansion capacity (EPS); and liquidity (Graham rating).
Finally, in part F it is shown that the mean for distress performance score is -0.965, and the mean for successful performance score is 3.859.     Note: *** Significance at 1%; ** Significance at 5%; * Significance at 10%.
Electronic copy available at: https://ssrn.com/abstract=3902486 positive (92.1% of the prediction attempts of the successful performance companies are successful and 7.9% are distress). In addition, it also shows specificity of 89.5%, which means that 10.5% are false-positive (89.5% of the prediction attempts of the distress performance companies are distress and 10.5% are successful). This is also an indication that the model has high sensitivity and specificity. Finally, the original group cases were 96.8% correctly classified, and 91.6% were correctly classified in the cross-validated group cases.

DISCUSSION
The results presented in this study are consistent and contradicting to a certain degree with Altman et al.

CONCLUSIONS
The financial failure of insurance firms in any exchange of the sector services is tremendously important due to the vital role it plays in stabilizing the country's financial environment accompanied with the other financial industries. Herein, the novelty in the development of the financial failure model is the key factor that differentiates this paper. The model consisted of 28 indicators selected from 11 parameters; it was derived from the other financial failure models regardless of the sector services that were mentioned in the literature review.
The results showed that the financial failure model is a good fit and statistically significant. The model even did a better job explaining the variances of the company's performance. There were a few contradictions between some results, which did not support the fitness of the model. Nevertheless, the model had high sensitivity and specificity in the original and cross-validated group cases. Moreover, the minimum deviation between the real and predicted performance is an indication of the insurance industry stability, 22 indicators out of 28 were found to be significant.
In conclusion, this paper summarizes a couple of important outcomes. To begin with, few deviations between the real and predicted performance are an indication that none of the insurance firms are threatened by failure or distress. This statement also supports the fact that the financial failure model is sustainable in predicting the financial failure of the insurance firms in ASE. It also reinforces the stability of ASE exchange and the entire financial environment in Jordan. Finally, it is recommended for future studies to assess the model by adding macro approach indicators and using artificial intelligence and neural networks as an analysis technique.