Explaining Your Financial Results Attribution Analysis and Forecasting Using Replicated Stratified Sampling

Transcription

1 Insights October 2012 Financial Modeling Explaining Your Financial Results Attribution Analysis and Forecasting Using Replicated Stratified Sampling Delivering an effective message is only possible when you have accurate and timely information. New techniques such as Replicated Stratified Sampling are improving companies ability to explain their financial performance. Today, leading an insurance company is like walking a tightrope on a windy day. Market conditions can leave senior management wondering whether its products will face a gentle breeze or a strong headwind. The challenges are compounded when risk management meets accounting, even in a principle-based environment. The consequences of a misstep may be severe: A missed earnings forecast or missed capital ratio expectations could dampen the enthusiasm of analysts and investors. An insurer could be punished even if the economics of a product line are in great shape, but volatility in earnings or surplus is not understood or communicated well enough. Senior management must be able to explain its risk management techniques to regulators, shareholders and rating agencies, and describe any potential disconnects between short-term earnings and longterm economics. But delivering this message requires reliable and timely information, which is only possible with robust and efficient measurement techniques. Attribution Analysis Attribution analysis is a tested way to gather reliable information. By breaking down a given result say, a change in liabilities into the fundamental drivers of the change, senior management can understand why that result occurred and how it can be improved. In turn, that information is fed back into the forecasting process and allows management to feel more confident when delivering guidance to the market. There is ample evidence that the complexity of both products and the regulatory environment demands better tools in this area: Product risk. Variable annuities (VAs) and universal life (UL) products are particularly vulnerable to fluctuations in the economic environment. The multitude of different designs for the guaranteed death and living benefits written with VAs has resulted in sensitivity to both levels and volatility of the equity market, compounded by policyholder behavior risk. The cost of the guarantees is directly related to the level of interest rates via the typical valuation approaches associated with hedge programs. UL products can bear significant interest rate risk if the performance of the general account assets is not able to support the level of interest rate guarantee provided. Again, policyholder behavior compounds this risk. An indexed UL policy would also have equity risk associated with the market performance of the particular index the contract tracks, potentially increasing the cost of any guaranteed benefits. Secondary guarantees on these products add further complexity.

2 Regulatory changes. Recent and pending changes in regulation increase the need for a more accurate assessment of earnings and balance sheet risk. For instance, principle-based requirements for VAs and other similar contracts under Actuarial Guideline (AG) 43 and RBC C-3 Phase II were intended to establish a more holistic approach to reserve and solvency requirements so that these amounts are directly related to the risks written by each individual company. However, with a few years of experience including the period of the global financial crisis many would argue that this framework has delivered counterintuitive outcomes. Of particular concern for users of this information is the potential for sound risk management techniques such as hedging or reinsurance to have adverse or unexpected impact on the required reserves or capital levels. Regulators are also closely scrutinizing reserving practices for UL policies with secondary guarantees. Recent changes to AG 38 now require the use of principle-based techniques in addition to traditional formula-based methods. The new requirements not only apply to new business but are retroactive to business dating back to as early as Adequately explaining the drivers of results for this business is often further complicated by internal and external reinsurance arrangements designed to alleviate the excess reserve strain. How It Works The ultimate goal of attribution analysis is to explain all the material drivers of change in a company s balance sheet over a given period. Ideally, this analysis is embedded in the production process itself to provide real-time validation of the reported results. Identifying the material drivers of earnings allows a battery of standard sensitivity tests to be designed and their results fed to senior management, which should also improve earnings or capital ratio forecasts. Companies will never be able to predict exactly how the market will play out over a given period, but a comprehensive stable of scenarios developed ahead of time can help avoid surprises. The ultimate goal of attribution analysis is to explain all the material drivers of change in a company s balance sheet over a given period. Relevant Reserving Guidance Over the past decade, as products such as VAs and UL products have become ever more complex, insurance regulators have recognized the need to implement methodologies that better reflect the diversity of product risks. Several pieces of guidance recently developed by the National Association of Insurance Commissioners reflect this awareness. AG 38. The Application of the Valuation of Life Insurance Policies Model Regulation provides guidance for the calculation of reserves for UL policies with secondary guarantees (ULSG). ULSG products have received close scrutiny from regulators over a period of years as product designs have changed, creating reserving ambiguities and different interpretations of requirements, as well as a significant demand for structured solutions (such as securitizations or captive reinsurance arrangements) to alleviate the reserve strain. In order to address concerns that some insurers are not holding adequate reserves for secondary guarantees, regulators have very recently adopted changes to AG 38 that will require the use of principle-based techniques for new business as well as many in-force policies. AG 43. The purpose of AG 43 is to interpret the standards for the valuation of reserves for VA and other contracts involving certain guaranteed benefits similar to those offered with VAs. AG 43 became effective on December 31, Some are urging that the guideline be reviewed in light of the experience during the recent global financial crisis. The review would also look at RBC C-3 Phase II, the capital guideline for these products. Some argue that the level and sensitivity of the reserves and capital, particularly after hedging, do not provide accurate or actionable data on the true risks assumed by companies writing VAs. RBC C-3 Phase II. This capital requirement was adopted in 2005 and was developed to quantify market risks for VAs and certain other products with similar guarantees. Like AG 43, the calculation requires a stochastic analysis of the material asset/liability risks retained by the company as well as a more prescriptive calculation called the Standard Scenario Amount. Due to the total balance sheet nature of the calculation, it is possible to have a zero C-3 capital requirement if the AG 43 reserves are larger than the Total Asset Requirement calculated under C-3 Phase II. 2 towerswatson.com

3 Even when the need for better attribution and forecasting abilities is well recognized, execution is hampered by the sheer scale of the exercise many dynamic factors need to be evaluated, including interaction among those factors: changes in equity markets, interest rates, implied volatility, as well as mortality, morbidity and various facets of policyholder behavior. In addition, for many business lines, analyzing both the change in liabilities and change in assets is crucial to explaining the overall performance. Quantifying the impact of each change requires running what is typically a very large model multiple times at each reporting date. Until recently, the run time associated with completing even a small fraction of the tasks required for a complete attribution analysis made this virtually impossible. But now, Replicated Stratified Sampling (RSS), a relatively new technique developed by Towers Watson that uses statistical sampling to accurately measure changes in risk metrics, makes developing the complete picture feasible. The use of RSS produces dramatic reductions in run time. Illustrative Case Study A More Robust AG 43 Attribution Analysis Using RSS In order to manage to a five-day production time frame, a company uses traditional cell compression techniques to reduce its model population by more than 75%. Even so, it is only able to complete the baseline run plus five sensitivity tests for the AG 43 stochastic calculations. These five sensitivity steps were used to create a basic attribution analysis (Figure 1). The company estimates that its compression techniques introduce roughly a 5% error in the results compared to a full seriatim run. To reduce this error factor, one option would be to run seriatim for the baseline and use grouping for sensitivities only. However, in order to meet production timelines, the trade-off would mean running fewer sensitivities. The company does run additional sensitivity tests when results change significantly, but that work has to be completed after the production cycle, when results are already finalized and messaging to its constituents has already begun. Until recently, the run time associated with completing even a small fraction of the tasks required for a complete attribution analysis made this virtually impossible. Figure 1. This basic attribution analysis is similar to what many companies produce today. It reflects a few key drivers of change, but production time frames and long run times hinder a more robust analysis. AG 43 reserve Assets Net gain (loss) Starting values (1,948.4) (194.8) (2,143.3) Expected impact from capital markets Change Equity indices 1.7% (119.2) 98.7 Interest rates 0.32% 12.3 (27.6) (15.4) Volatility Cross Greeks (0.5) (0.4) (0.9) Total capital markets (146.9) 82.8 Impact from new in-force file New business (2.2) (2.2) Unexplained (5.0) Total other impact (7.2) Closing values (1,726.0) (330.7) (2,056.7) Gain (loss) (135.9) towerswatson.com

4 A More Robust Analysis Using RSS, the company is able to make several enhancements to its AG 43 process. First, the company uses traditional cell compression techniques to produce its baseline result but introduces RSS for the five sensitivity tests. The company is able to complete all of these runs overnight, and with an additional day to compile the results, it has completed the same analysis as usual in two days instead of five. The error associated with the sensitivity tests using RSS is less than 1%, based on the calibration of the RSS parameters. The company then begins to take advantage of these time savings. It implements a full seriatim run for the baseline result, eliminating the roughly 5% error introduced by cell compression. This adds an additional day to the process, which now stands at three days instead of the original five. The refined baseline result is now more precise than ever. The company is able to use the remaining two days of production time to increase the number of sensitivity tests to over 25, allowing for a more robust attribution of several interest rate, equity and interaction effects. This allows the company to produce a more complete and timely attribution, ensuring the key drivers of the change can be clearly communicated to management along with the reported results (Figure 2). Using the expanded analysis in Figure 2, the company has a much clearer picture of what drove results during the period in question and is able to significantly reduce the magnitude of the unexplained change. The summary factors in Figure 2 could easily be expanded to reflect even more individual effects. For example, basis risk could be expanded to monitor key exposures to unhedged factors, as well as correlation effects. Interest rate exposure could be split into key rate components. Total decrements could have several subcomponents, including lapses, excess partial withdrawals, benefit elections and mortality. Separate attribution results should also be developed for the standard scenario and the stochastic results to fully explain the drivers of change and the environment in which each would dominate. Replicated Stratified Sampling The Towers Watson RSS technique uses a stratified sampling statistical approach to estimate the impact that a factor will have on the entire population by observing its impact on a sample of the population. The key to the process is the technique of drawing multiple samples from the underlying population. Its power comes from the speed of convergence and your ability to fine-tune the degree of precision that you need. Other efficient modeling methods have limitations, whether in terms of the ability to control the precision of the result or the need for frequent recalibration, or because of the applicability to just one calculation type. RSS covers all of these aspects very well, providing speed, accuracy, minimal need for recalibration and broad applicability. The RSS process starts with determining the metric that is to be tested (reserve, values, etc.). Then we use a calibration process to select the optimal mix of strata and sample sizes to produce the desired level of precision in the shortest amount of run time. The underlying population data i.e., the in-force contracts are sampled and run through the existing actuarial models. The output is combined as part of the RSS technique, producing the sensitivity of the desired metric to a given change. RSS offers several benefits over other smart modeling techniques: It is easy to understand because it is based on statistical techniques that are widely used and proven. The user has full control over the parameters and the ability to calibrate to the desired level of accuracy. The technique is software agnostic, meaning it relies on your existing financial modeling software there is no need to invest in new software. There is little need for maintenance or recalibration. New samples are drawn each time the process is used (e.g., monthly, quarterly), and the parameters do not change frequently. 4 towerswatson.com

5 As a next step, the analysis could be expanded to include capital projections, allowing a total balance sheet view of the impact that market changes will have on the full economics for the business line. As market conditions change and the business matures, continued refinement of the analysis would be expected. These refinements would also be fed back into the company s forecast model to help facilitate strategic decision making for this line of business. Figure 2. This is an example of a more robust attribution analysis. Producing this level of detail is possible only when using a smart modeling technique, such as RSS. AG 43 reserve Assets Net gain (loss) Starting values (1,948.4) (194.8) (2,143.3) Impact from capital markets Change Equity S&P % 93.1 (119.2) (26.1) BOND Index 1.3% (34.0) (34.0) Russell % EAFE 1.2% Cross delta (4.9) (4.9) Fund basis risk equity Subtotal: Equity index (119.2) Interest rates One year 0.18% Five year 0.62% (0.3) (5.1) (5.4) 10 year 0.59% 10.9 (23.3) (12.4) 20 year 0.43% 30 year 0.38% Fund basis risk bond Subtotal: Interest rates 14.7 (27.6) (13.0) Equity volatility One year 1.9% Five year 0.8% Subtotal: Volatility Cross Greeks (0.5) (0.4) (0.9) Time (25.6) 10.4 (15.2) Total impact from capital markets (136.5) 83.1 Impact from actuarial assumptions Total decrements Other impact New business (2.2) (2.2) Unexplained (2.5) 0.6 (1.9) Total other impact (4.7) 0.6 (4.1) Closing values (1,726.0) (330.7) (2,056.7) Gain (loss) (135.9) towerswatson.com

6 Conclusion Current market volatility and complex reporting requirements make running an insurance company difficult even for the most skilled senior management teams. Analysts and investors may cool to a company if it misses an earnings prediction or capital ratio expectations. This possibility makes it imperative that management can explain changes in company financials with confidence and deliver credible guidance to the market as to future expectations. But this message can be delivered only if reliable information is available, and reliable information is available only when accurate and efficient measurement techniques are in play. With thorough attribution analysis, management can create a holistic picture of how different risks will impact the company. This more complete understanding of potential risks to company earnings and balance sheets can help senior management to better prepare for market volatility and set realistic earnings expectations. RSS is one way that management can ensure these analyses deliver on that promise. Kendrick Lombardo is a consultant with Towers Watson, specializing in pricing, risk management, financial modeling and actuarial appraisals. Kendrick focuses on annuity products and other products with investment guarantees. He has advised companies on many aspects of product design, policyholder behavior, market-consistent methods, capital management and in-force management. Cheryl Tibbits is a director with Towers Watson, specializing in financial modeling, mergers and acquisitions, and capital management. Cheryl is the head of the Atlanta life insurance consulting practice and leads the Software and Financial Modeling initiative for the Americas Life practice. She is also a leading expert in the modeling and management of VA products. About Towers Watson Towers Watson is a leading global professional services company that helps organizations improve performance through effective people, risk and financial management. With 14,000 associates around the world, we offer solutions in the areas of employee benefits, talent management, rewards, and risk and capital management. Copyright 2012 Towers Watson. All rights reserved. TW-NA towerswatson.com