ru kz en

Dina Urzhumova, actuary: "Essential deviations in long-term forecasts are inevitable"

Continuation, Part 2.
Dina Urzhumova, actuary: "Essential deviations in long-term forecasts are inevitable"

Errors in judgments of a Kazakhstani actuary are often due not to the low level of professionalism of an actuary, but to the fact that he, for objective reasons, is forced to make judgments based on limited, incomplete and sometimes poor-quality information, noted Dina URZHUMOVA, a professional actuary, CFA and experienced economist, Chairman of the Qualification Commission of the Society of Actuaries of Kazakhstan.


- Dina Stanislavovna, are there disagreements in judgments among actuaries? And what decision, in your opinion, should the court take if the matter concerns a dispute between the decisions of two actuaries?

- As our market is rigidly regulated, the big discrepancies in ordinary situations usually do not arise. If divergences are small, then both actuaries are right. I would like to repeat, estimation is not a point evaluation, but an interval of allowed values.

As I have already said, the quality of an actuary's judgment depends on the level of his professionalism, as well as on the quality and volume of information. In non-routine situations, especially with actuarial calculations made for the first time, the discrepancies potentially could be significant, especially if the estimates were based on different volume and quality of information with no strict calculation parameters. Judicial bodies are not competent enough to decide whose assessments are more adequate. In these cases, it is necessary to attract independent actuaries as experts, who do not have a conflict of interest with either party to the dispute. Such actuaries should have sufficient working experience to make such calculations and studies.

If the dispute is between an actuary and a third party, for example, a Customer of actuarial services, the Society may assist in protecting the interests of an actuary in court. It can hand out an independent actuarial conclusion and/or send its specialists to appear in trial on the side of an actuary.

The court should decide on the basis of an expert opinion.

- The actuary’s evaluations and calculations affect the amount of the company's pay-out for a long-term period, for example, 50 years in the case of pension annuities. In this case, can conservatism be considered as actuary’s protection?

- As I have already said, the solvency of insurance organizations and the financial well-being of policyholders depend on actuary's actions. As a matter of fact, insurance companies, beneficiaries, and supervisory authority are the actuary’s customers. The actuary always seeks to receive non-conservative, and not aggressive, but fair assessments, as only in this case the interests of all the actuary’s clients are fully satisfied. A fair estimate is the mathematical expectation of an estimated random value. The reasons why the actuary chooses conservative estimates are generally either incomplete and/or poor-quality data that do not allow obtaining fair estimates with a sufficient level of reliability or regulatory requirements enacting a conservative basis for assessments. The latter is true for the valuation of reserves. For long-term insurance products, the degree of economic and financial stability is also one of the factors influencing the choice of the actuary. If the actuary is sufficiently confident in the long-term forecast of the input parameters of the calculations, he will choose in favor of fair estimates.

Earlier I talked about an outmost instability of economic and financial situation in Kazakhstan. The level of complexity of actuarial decisions and protection of such decisions before management is best demonstrated by a hypothetical example. I have no information on historical dynamics of investment returns on the investment portfolio of insurance companies, so I will use historical factual data on pension funds as an example. From 1998 to 2002, the nominal return on investments exceeded 11% per annum.

Let us suppose, in 2003 an insurance organization, inspired by high profitability, decides to sell the product of a life-long, non-pension and therefore unregulated annuity with payments increasing by 5% per year. For simplicity, let us assume that the cost of such an annuity is zero. The same product is issued by other life insurance companies, and, due to high competition, sales depend on the size of payments, which in turn depend on the assumptions used in calculating payments, the more conservative the assumptions, the lower the payments are. The actuary makes a long-term forecast of investment returns for such product valuation, not having historical data prior to 1998. Since a number of data is insufficient to obtain a reliable long-term forecast, the actuary is forced to adhere to a conservative approach and assumes a long-term rate of 9%. Most likely, the management of the insurance organization would seem to find this approach too conservative. In 2003, the actual return on investment drops to 7-8%. The product is reassessed, and the actuary for the newly issued annuities in the calculations now uses a conservative rate of 6%, since a number of data are still insufficient to obtain a reliable long-term forecast. Again, the management of the insurance company will seem to find this choice too conservative. In 2008, the actual nominal rate of return on investment falls to a negative -1%. The year 2009 has come. The “killer” questions are: 1) Was the actuary excessively conservative or overly aggressive in past calculations?; 2) Should we continue issuing annuities with payments increasing by 5% per year?; and finally, 3) What long-term rate of return should an actuary use when assessing payments for newly issued annuity annuities?". These are questions to which the actuary should respond. I remind you that he is personally responsible for his decisions.

Now try to answer your own question, whether conservatism is actuary’s protection.

Now imagine that the actuary was unable to defend his decisions before the management either in 2002 or in 2003, and he used higher long-term rates of return in the calculations. Again, despite the fact that the decisions were made at the insistence of the management, he nevertheless bears personal responsibility for them.

The more experienced an actuary is, the more likely he was to face similar situations in the past and, consequently, the more aces he has on his hands in order to justify his decisions reasonably.

Communication skills play an extremely important role in the work of an actuary. He needs to be able to explain in words or reflect in a report difficult things from a technical point of view in simple language, understandable for layman, to justify and defend his position.

- What is accuracy of calculations at the time of settlement?

- In order to evaluate whether the calculations were correct at the time of calculations, it is necessary to discard all the information that became known after the settlement date, and after that ask the same question, i.e. the correctness of calculations at the time of settlement is their correctness on the basis of the information that the actuary owned at that moment.

I will provide one vivid example from another field of activity, but conceptually identical to the situation in question. A few years ago, I worked for KPMG in the corporate finance department and was engaged in business valuation. Discounted cash flow method of business valuation is similar to the perspective method in actuarial calculations. I estimated the cost of one small private airport as of January 1, 2007. In accordance with business valuation standards, any events that occur after the date when the business is evaluated are not taken into account in valuation. In 2007, the financial crisis began; the problems with the Kazakh banks’ liquidity occurred in autumn. In addition, there have been adverse changes in the air transportation industry. All this negatively affected the financial position of the airport. Auditors conducting an audit of the airport in 2008, tried to challenge the assessment, referring to the described negative changes. After I introduced the standards of business valuation, they had to accept the assessment. The airport was really much more expensive on January 1, 2007, and a potential investor would be willing to pay such a price on that date, as the subsequent unfavorable events would not have been known to the investor.

- What is the world practice on actuary mistakes? If, for example, the error was found in 10-50 years, what is his responsibility? Are there precedents?

- To be honest, I do not like the word "mistake" in applying to long-term forecasts. It is applicable to incorrect results that can be avoided. And significant deviations in long-term forecasts are inevitable. I would prefer to use the word "error" in this case. In an unstable economy, even short-term forecasts are rarely accurate. And the longer the forecasting horizon is, the greater the error in estimating the expected values ​​of the estimated parameters.

Even in a stable economy, a static assessment of the mortality tables, correct at the moment of calculation, inevitably becomes inadequate, as mortality throughout the world is steadily declining due to rise in living standards and medicine development. The projected reduction in mortality should be taken into account in the evaluation of life insurance products and pension programs. Please note that in practice even such dynamic estimates have a significant error. I read in a research article by IMF experts, where they pointed out that the projections of mortality reduction rates used in calculating pension obligations in countries with the most developed actuarial profession, the United States, Great Britain, Canada, Australia, Japan and New Zealand, as a rule, are significantly underestimated. This leads to a spasmodic increase in pension liabilities at the moment of updating the mortality tables used to evaluate pension plans with fixed payments.

In our conditions, it is now impossible to obtain a reliable estimate of mortality reduction rates. 

In 2015, we updated the mortality tables to calculate retirement annuities. Together with the tables, we wanted to estimate the expected mortality reduction ratio. With this purpose in mind, we analyzed the data on mortality of the population for 2000-2014. In 2000-2006, the change in mortality was insignificant, with mortality rate growth in the working age, and its decline in the pension age. I have already said that with increase in level of economic activity in this period, the mortality rate from accidents at work increased. In 2007-2014, mortality declined by an average of 7% per year (!). This is an extraordinary high rate of decline. The main reason for such dramatic changes in mortality after 2007 was probably the implementation of the Program for Development of Cardiac and Cardiac Surgery Assistance, adopted in 2007. If the assumption is correct, then the mortality rate decline in future will slow down, as a further significant reduction will require more significant budgetary expenditure on health development. I have not analyzed mortality in Kazakhstan for the last 3 years. I believe that, given the deepening crisis caused by the fall in oil prices, mortality has not declined, but may have somewhat grown.

I quote all these details to make it clear why it is so difficult to get reliable estimates. There were no ordinary, stable periods that would make it possible to assess the long-term rates of mortality reduction in Kazakhstan with sufficient certainty! In such conditions, "mistakes," as you call them, are unavoidable.

In 2004, I was an intern in the research department of the American Society of Actuaries. This department evaluates the normative mortality tables on a regular basis. I asked the experienced department expert what he would do if he did not have enough data to get reliable estimates. He said that the usual practice in such cases was to use zero mortality until the data was accumulated.

Due to the small number of insurers, the accumulation of insurance companies' data for a reliable estimate of the mortality rate of pension annuity payment recipients is not expected in the foreseeable future, and in case with disable people mortality rate assessment it is even worse.

The use of zero mortality is, in fact, a "punishment" of recipients of pension annuity payments, as in this case their payments are substantially below fair values. In case of compulsory insurance of the employer's liability for causing harm to the employee (CAI), enterprises are "punished" because they have to pay unjustly high premiums in order to provide payment for lost income replacement. Therefore, it is better to use some compromise conservative option to partially reduce the degree of "punishment". I emphasize that compulsory accident insurance (CAI) is a mandatory type of insurance; the policyholder cannot vote with "feet", i.e. to refuse to purchase an insurance product if he considers the insurance rates to be too high.

Such a compromise table of mortality inevitably has a significant error. But such an error cannot be called a "mistake".

Therefore, without understanding the situation, it is not worthwhile to accuse the actuaries that their judgments were erroneous. Errors in judgments are often explained not by a low level of professionalism of an actuary, but by the fact that he, for objective reasons, has to make judgments based on limited, incomplete and, at times, poor-quality information. And once again, the actuary bears personal (!) responsibility for an error as a professional participant in the insurance market. I hope that the cycle of interviews with actuaries on your information portal will raise awareness and understanding of the problems that actuaries face in society.

And I would very much like to hope that the attitude of customers to the work of actuaries evaluating long-term products and programs will change over time. Customers, as a rule, put unrealistic, tight deadlines, but, however, require a comprehensive and thorough analysis in obtaining reliable estimates, without understanding that with such a volume of work it is either quality or timing. I worked on a number of pension projects, and from my experience I can say that the preliminary stage alone, gathering information, checking data, finding out the causes of errors and inconsistencies in the data and adjusting them, takes at least 2-3 months! And if the data are "ineradicable", then the alternative information source search, the analysis of the applicability of the data, and analysis of these data can take at least another month. If certain actuarial calculations are made for the first time, building a model also requires a considerable amount of time, depending on its complexity.

- The insurers shared their discontent with the mortality tables of disabled people, which was developed in Kazakhstan. They say that these tables do not work: no matter how cynical, but fewer people die than it is indicated in the calculations.

- As far as I remember, the evaluation of the mortality tables of disabled people was in 2009. Actuaries of the Society did not participate in their development; therefore I do not know all the details. But I can explain why they were overestimated. I have already mentioned that the number of people receiving payments from insurance organizations in connection with their professional capacity loss is extremely small, and therefore it is impossible to estimate the mortality rate on the basis of insurance data. For table construction the data of the State Social Insurance Fund (GFSS) were used. Since compulsory social insurance was introduced in Kazakhstan in 2005, it was only data on disabled people who had lost their ability to work from 2005 to 2008. The reliability of the estimates obtained on the basis of this small sample was extremely low.

Besides, the estimates obtained reflected so-called selective mortality. Please note that disability, especially in case of a high degree loss, is a shock for the body. Afterwards, the body adapts and stabilizes. Therefore, the mortality of disabled people depends not only on age, but also on the duration of disability. The higher mortality in the first years after the disability is called selective, and at the end of the adaptation period it is called “final”. Ideally, people with disabilities need to build two-dimensional tables of mortality. Since the SSIF data contained information only about the mortality rate of people with the disability period of no more than four years, based on such data, minus adjustment is required.

A more appropriate source would be data from the SCPP on the mortality of disabled people in the social security system. First, it is more than 300 thousand people in comparison with a small sample of the State Social Insurance Fund for the years of 2005-2008. Secondly, the social security system for disabled people exists since the Soviet era, and, based on the SCPP data it is possible to obtain estimates of the final mortality rate. I do not know why the SSIF data was selected.

- You also made a mortality table for pensioners in Kazakhstan, twice with a major interval. What is the conceptual difference?

- The first time in 2004, the second - in 2015. In both cases it was based on limited information, and quite a low reliability of the estimates.

The complexity of mortality estimate of pension annuity recipients is explained by extremely small number of the group, only around 2% of retirees. Retirement annuities can only be acquired by persons who have had high incomes during the entire accumulation period from 1998 to retirement, since the criterion for buying an annuity is the sufficiency of savings to receive payments not lower than the statutory minimum pension at the date of purchase of the annuity. Besides, due to the short history of the issue of pension annuities, data for people older than 70 years of age are not available.

In 2004, with complete absence of insurance companies' data, we had to use foreign statistics on the mortality rate of the population in terms of income level as well as Kazakhstani statistics on the mortality rate of the population in terms of education as benchmarks. We also made a comparative analysis of the mortality of the population and the mortality rate of policyholders who receive payments for lifetime pension annuities in those countries where this insurance product is voluntary.

Compared to 2004, in 2015 we were able to slightly improve the initial estimates, as insurance companies have accumulated data on the mortality rate of recipients of pension annuities. These data were not enough to build complete mortality tables because of a low number of policyholders, and yet they gave us additional supporting points and allowed us to conclude that the initial estimates required adjustments downward. Thus, in 2004 we were not sufficiently conservative. Time will tell how correctly we estimated the mortality this time. If there are no adjustments made within 10 years, I assure you, the problems will accumulate again.

Both in 2004 and in 2015, we wrote in the reports that the estimates obtained are of low reliability, and in the future, an actual monitoring of the actual experience, the data analysis and periodic table correction are required. To do this, it is necessary to collect and analyze the statistics annually and, if necessary, make adjustments so that there will be no accumulation of losses of insurance organizations (in case of overvaluation) or policyholders (in case of underestimation).

Let us talk about the actuarial control cycle, which includes the following stages: product development, actuarial evaluation, implementation, monitoring of actual experience, and finally, making adjustments. Continuous repetition of this cycle is necessary even when reliable estimates are available, since there are systemic changes in economic and demographic indicators. This is especially important when the reliability of these estimates is low. Unfortunately, the Society of Actuaries does not have the necessary financial resources to carry out such work, and the National Bank of Kazakhstan initiates the table update only when the inadequacy of mortality indicators becomes too tangible for insurance organizations. If the initial estimates of mortality were underestimated, it is likely that the table update would not been initiated at all, and this would negatively affect the beneficiaries.

The period of 10 years for updating the mortality tables is unacceptably long. Unfortunately, relatively accurate data can only be obtained for the last 2-3 years. Based on my own experience, I can say that there are mistakes and inconsistencies in the data, as a rule. Only actuaries can identify the errors, find out from the source of information the causes of such errors and make the necessary adjustments; it is possible to do without delay, i.e. in regard to the data for the last year only. To our knowledge, the National Bank of Kazakhstan intends to organize insurance statistics collection in the Credit bureau. Obviously, this will not solve the problem of poor data quality and, as a consequence, low reliability of estimates, since regular data analysis is not planned. In case of mortality tables of disabled people, the data needs to be collected from the SCPP and the Disability Determination Services (DDS), and the State Social Insurance Fund. The credit bureau will not deal with this. I would like to note that the SCPP completely or partially overwrites the historical data, and for that reason, the data accumulation is very slow, if the information collection takes place with a period of 10 years.

The Society of Actuaries could collect and analyze data, and update the mortality tables on a regular basis. But for this a source of financing must be found.

In the United States and Great Britain, for example, professional actuarial associations have special research departments that regularly collect and analyze insurance statistics on mortality of insured persons for various types of insurance on a regular basis. The mortality table update is also carried out by the working group of the professional association of actuaries. But in these countries associations have several thousand members, and their incomes amount millions of dollars. As for the social security system, there is a constantly working group of actuaries in the relevant state body that updates the mortality tables and assesses the solvency of the social security system in the short, medium and long term.

In my reports on various projects of the Ministry of Labor I have many times noted the need to create such a group within the Ministry of Labor or other state authority. With constant monitoring and analysis, it will be possible to develop an adequate model of the social security system and balance its parameters in such a way that it remains solvent in the long run and, at the same time, provides adequate income to the beneficiaries. Only in this case will the pension system not be frustrated by sporadic chaotic changes, designed to solve one problem in the short term, and at the same time fraught with long-term problems and negative side effects. Sometimes innovations cancel changes made earlier. Long-term consequences and side effects are often not analyzed, as a result we "fix one thing and break another". A vivid example is liquidation of private pension funds and consolidation of all pension assets in the Single Accumulated Pension Fund (SAPF). For comparison, in the UK, the decision to introduce a pension reform was taken only after actuaries working in the State Actuary office for 10 years (!) carefully analyzed the consequences of such a reform and "tuned" the parameters of the future pension system.

In conclusion, I want to say that in any field where actuarial services are required, the actuary's work, ideally, should be permanent, with a continuous application of actuarial control cycle. This will allow making timely adjustments and preventing the accumulation of losses if the actual indicators differ significantly from the forecast values, especially in an unstable economy, as well as in cases where judgments, forecasts and conclusions, for objective reasons, are based on limited, incomplete information.

- Dina Stanislavovna, thank you very much for the interview!

Note: SCPP was renamed as NJSC Government for Citizens. For brevity, the old abbreviation is used.

Link to the first part of the interview:

read also
Freedom Finance insurance companies recommenced financial literacy classes for schoolchildren
Freedom Finance insurance companies recommenced financial literacy classes for schoolchildren

The Freedom Finance Camp project was created with the support of the Agency of the Re...

Some LICs in US stopped signing agreements with older people
Some LICs in US stopped signing agreements with older people

Some US life insurance companies suspended applications from older people during the ...

Dina Urzhumova, an actuary: "Point estimate is zero information, this is a 100% error probability"
Dina Urzhumova, an actuary: "Point estimate is zero information, this is a 100% error probability"

Dina Urzhumova is an actuary, CFA and economist with extensive working experience, an...

The most interesting materials of the site you have in the mail! Subscribe to the newsletter.

Submit your application