Insights October 2011 Variable Annuity Hedging Practices in North America Selected Results From the 2011 Towers Watson Variable Annuity Hedging Survey Introduction Hedging programs have risen to prominence among life insurers seeking to manage capital market risks associated with their products, especially variable annuities (VAs). These programs rely upon the ability to acquire effective positions that offset risk across meaningful time frames. It is generally believed they offer advantages over more traditional risk management techniques such as passive diversification strategies, which offer insufficient capital market risk protection, or reinsurance options, which have limited availability or are too expensive. The importance of hedging in managing VA capital market risks prompted Towers Watson to conduct two surveys to explore insurers use of VA hedging practices the first in early 2009 and the second in early July of 2011. This article presents selected highlights from the 2011 Towers Watson VA Hedging Practices Survey ( the survey ). Both the survey and this article focus on: Program overview Hedging strategies Modeling Systems Recent program changes ranges from about three years to 15 years. However, a frequency distribution of hedging program age indicates significant program development immediately following the tech bubble collapse and the advent of guaranteed minimum withdrawal benefits (GMWB) early in the last decade. Approximately 6% of respondent hedge programs are between six and eight years old. All living benefit writers participating in the survey hedge at least some of the risks associated with these guarantees. However, guaranteed minimum death benefit (GMDB) and guaranteed minimum income benefit (GMIB) writers have a more varied treatment compared with guaranteed minimum accumulation benefit (GMAB) and guaranteed minimum withdrawal benefit (GMWB) writers. The elections made by respondents to hedge specific guarantees are summarized in Figure 1. Notably, some GMDB and GMIB writers are electing not to hedge these benefit types, with accounting mismatch issues the likely motivation (see more on this below). In addition, select respondents apply a macro hedge to cover certain guarantee types, which were counted as hedged for the purposes of this question. Figure 1. Guarantees hedged by number of participants This survey targeted 21 top North American VA writers as potential participants. Seventeen responded to the survey, including one company that said it primarily used reinsurance solutions for VA risk management. Therefore the numerical and graphical results in this article assume a universe of 16 respondents. For example, a reference to half the respondents would signify eight survey participants. Program Overview Participants in the hedging survey cumulatively account for over $700 billion of VA account value, and the average across participants is approximately $45 billion. The age of hedging programs covered by the survey 20 15 10 5 0 10 6 0 GMDB 8 4 4 GMIB 1 52 15 0 0 1 GMAB GMWB Hedged Not hedged Not written
Just as survey responses provided insights into the variability of the basic decision to hedge a specific guarantee, they also provided insights into the nature of the hedging strategies being pursued, by both writer and guarantee type, once the decision to hedge is made. Hedging Strategies Within the broad area of managing capital market risks, hedging strategies are largely defined by the specific strategic objectives under which they operate. These can range from economic to regulatory as indicated in Figure 2. Respondents said the most typical strategic hedging objective for all guarantee types is economic, and the least typical is AG 4, with more response variability exhibited for GMDB and GMIB exposures. As shown in the figure, select writers use a combination of SFAS-1/157 and SOP 0-1 for living benefit objectives. Accounting mismatch issues for GMDB and GMIB riders continue to pose challenges for VA writers. U.S. GAAP-based reserves for GMDBs and GMIBs are developed from real-world-based valuations and accrual methodologies, and do not track well with the marked-to-market derivative contracts used to hedge these risks as markets move. Thus many writers are leaving some of these riders unhedged, or hedged at percentages notably less than target. Other respondents indicated use of macro-hedging techniques for these riders. These techniques typically involve use of out-of-the-money derivatives to provide protection against severe market movements. For a given strategic objective, the computation of the risk offsetting positions required by a hedging strategy relies critically on the valuation of the guarantee embedded in a VA contract, and, in particular, the sensitivities of the hedged guarantee to specific changes in market parameters. These sensitivities, referred to as Greeks, are defined as the rates of change, or derivatives, of the guarantee value with respect to the relevant market parameters. The Greeks are of central importance within hedging programs because neutralizing exposure to a given market risk is tantamount to creating positions, through hedging transactions, that have little aggregate sensitivity to that risk, or (near) zero net value of the associated Greek. In practice, it is not possible to maintain aggregate Greeks that are identically zero so that hedging strategies will typically specify trading tolerances, or risk limits, to bound the size of aggregate Greeks. The key Greeks that are typically considered for hedging are: Equity Delta and Gamma. First- and second-order derivatives, respectively, of guarantee value with respect to equity price level Rho and Convexity. First- and second-order derivatives, respectively, of guarantee value with respect to interest rates Figure 2. Primary hedging strategic objectives 10 8 6 4 6 8 5 9 9 8 2 0 N/A 1 2 2 1 0 Economic SFAS 1 220 15 0 1 SFAS 157 1 2 SOP 01 11 0 1 1 1 1 1 AG 4 Other 12 GMDB GMIB GMAB GMWB 2 towerswatson.com
Vega. First-order derivative of guarantee value with respect to (equity) volatility Foreign Exchange (FX) Delta and FX Gamma. Defined similarly to the analogous equity quantities, but for foreign exchange rates these are less often utilized in the context of VA hedging Decisions regarding which specific Greeks to hedge, and the frequency with which hedging transactions occur in order to maintain stipulated risk limits, depend upon the risk appetite of the writer, and are properly made with a consideration of the relative costs and benefits associated with hedging the particular risks associated with a specific guarantee. Certain trends were observed in the survey responses for each guarantee type: For GMWB, the most widely offered VA guarantee, all respondents that write the guarantee hedge Delta, most typically either intraday or once per day. More than three-quarters of the survey participants hedge Rho, most typically intraday or once per day, but hedging also occurs weekly or with another frequency. About a third of respondents hedge Vega and Gamma. Among these, specified frequencies include intraday, daily, weekly, monthly and other. A little more than a third of respondents hedge FX Delta, with most of these hedging intraday. For GMAB, all respondents that write the guarantee hedge Delta, with the typical frequency being intraday or once per day. About two-thirds of respondents hedge Rho, with frequencies ranging from intraday to weekly, though other frequencies also occur. Most survey participants do not hedge Vega, but those that do hedge intraday or daily. About a third of the respondents hedge Gamma and FX Delta, with frequencies ranging from intraday to monthly and other periods. For GMIB, a little more than half of respondents hedge Delta, typically intraday or once per day. About a third of respondents hedge Rho and FX Delta, with specified frequencies being intraday, monthly or other. Most survey respondents do not hedge Vega. Among those that do, specified frequencies include daily, monthly and other. For GMDB, most respondents hedge Delta, and this is most often done intraday or once per day. A little less than half of respondents hedge Rho, and more than half hedge intraday, with others hedging weekly or with another frequency. About a third of the respondents hedge FX Delta, with half of these hedging intraday. Most respondents do not hedge Vega or other sensitivities. Modeling Valuation of the guarantees embedded in a VA contract for hedging purposes is typically accomplished within a Monte Carlo framework in which risk-neutral economic scenarios are generated using parameters appropriate to current market conditions. For example, parameters used to model equities typically include spot levels and volatilities, while those used for interest rates would generally include the current yield curve and perhaps additional parameters such as mean reversion rates and volatilities. Though FX risk is not frequently explicitly modeled in VAs, the market variables generally used to parameterize this risk include FX spot levels and volatilities. Once these economic scenarios are constructed, they are used with a detailed model of the structural features of the guarantee and policyholder information to value the guarantee as the expected value of the present value of its future cash flows. Greeks associated with the guarantee are then calculated using a finite difference approximation. This involves valuation of the guarantee at closely spaced values of the market parameters defining Macro hedging techniques typically involve use of out-of-the-money derivatives to provide protection against severe market movements. Respondents said the most typical strategic hedging objective for all guarantee types is economic, and the least typical is AG 4, with more response variability exhibited for GMDB and GMIB exposures. towerswatson.com
Figure. Stochastic processes used for equity index modeling 1% 1% 6% 68% 68% Lognormal 1% Mixed 1% Stochasitc volatilty 6% Other Figure 4. Interest rate modeling approach 6% 19% 1% 6% 56% 56% Deterministic 6% Stochastic, one-factor model 19% Stochasitc, two-factor model 6% Stochastic, multifactor (greater than two) model 1% Other the relevant rate of change, differencing these guarantee values and dividing by the parametric spacing to approximate the appropriate derivative. Because of the innate complexity associated with the valuation of VA guarantees, there are a number of modeling decisions that writers must make regarding the trade-off between fidelity and simplicity. The nature of the stochastic process used to model equity risk is a case in point. The range of choices made by respondents is shown in Figure. As seen in the figure, respondents predominantly use a lognormal process for modeling equity risk. Lognormal models are easy to implement and offer operational tractability, but cannot capture certain aspects of the empirically observed behavior of equity derivative markets. In particular, the wellknown dependence of equity implied volatility on an option strike, referred to as the volatility smile (or skew ), cannot be captured by such models. Hence the choice of a lognormal model favors a trade-off of simplicity over fidelity. A quarter of respondents use more complex models such as mixed lognormal or stochastic volatility (typically Heston) to capture volatility skews. As depicted in Figure 4, interest rate modeling for VA hedging bears certain qualitative similarities to equity modeling. The simplest choice that can be made in this case is the use of a deterministic interest rate model, and this choice is in fact made by a bit over half of the survey respondents. Obviously, the observed random character of interest rates and the attendant complexities of interest rate dynamics cannot be captured by such models. The decision to use deterministic interest rate models offers another example of favoring simplicity over fidelity, with the loss of fidelity arguably more significant than when lognormal models in equities are used. Perhaps this accounts for the fact that the proportion of respondents choosing to make use of more complex interest rate models, with the ability to achieve greater fidelity, is somewhat greater for interest rate than for equity models. It should be noted that more than 1% of survey participants say they use stochastic interest rate models, while other respondents cited Hull- White, Black-Karasinski, Hull-White II, Libor Market Model and not applicable (use of a stat-based strategy). 4 towerswatson.com
In both cases, it is important for the insurer to have a clear understanding of the risks unaccounted for in a choice of an equity or interest rate model, and quantification of the potential impact on hedge targets and Greeks if there is a switch to a more sophisticated model. Systems Due to the computational intensity of the Monte Carlo calculations typically used for VA hedging, and the need to perform these calculations in a capitalmarket-driven operational environment, hedging valuation systems tend to be quite specialized. Notably, almost all survey respondents identified use of a modeling solution specifically designed for hedging, whether developed in-house or acquired from an external vendor, rather than generalpurpose actuarial modeling software to support their hedging operations. It is noteworthy that no survey participants outsourced the calculation process itself, supporting the supposition that outsourcing of the calculation function, to the extent that it occurs, may be limited to small to midsize market participants. Figure 5. Techniques used to accelerate hedging calculations 0 6 9 12 IT-distributed processing techniques Unique scenario generation by policy Model point compression 5 Advanced math techniques 2 Other Not applicable 1 10 11 In response to the great demands for processing speed, hedging programs use a wide range of techniques to accelerate the processing speed of hedge ratio calculations, as indicated in Figure 5. In this figure, responses indicate that unique scenario generation by policy rivaled distributed processing. Much has been written about the use of advanced mathematical techniques such as accelerated Monte Carlo or replicating portfolio techniques to improve computational efficiency. But survey responses suggest that, so far, these methods have not gained traction in the industry. While the majority of survey responses indicate at least some use of distributed processing networks to achieve greater processing speed in hedging calculations, a closer look at the survey results suggests a noticeable size polarity in these responses. A little over 56% of respondents used over 250 computing nodes to support their hedging programs. However, a little over a quarter of the respondents use 25 or fewer nodes. towerswatson.com 5
Figure 6. GFC Strategy for managing market/credit events 0 2 4 6 8 10 We have already introduced a new, or strengthened an existing, macro-hedging program 5 We have already introduced a new, or strengthened an existing, counterparty credit risk management program We are planning to introduce a new, or strengthen an existing, macro-hedging program over the next year but have not yet done so We are planning to introduce a new, or strengthen an existing, counerparty credit risk management program over the next year 0 Other Recent Program Changes The survey focused on program changes prompted by the global financial crisis (GFC) and the adoption of AG-4 for VA statutory reserving. In general, the results, as expected, found that the GFC prompted more programmatic changes than the adoption of AG-4. The most dramatic changes from the GFC occurred on the product side. Traditional approaches (fee increases, limiting fund access, reducing benefit richness) were adopted by almost all survey participants. However, newer product-based risk management techniques also emerged and continue to gain traction in the marketplace (CPPI, target volatility funds, among others). With respect to hedging, the GFC highlighted the potential for problems in dynamic hedging programs associated with large adverse market movements occurring over a short time period. These arise in connection with abrogation of the small change assumption. The assumption underlies the use of first-order Greeks to neutralize market risks. These 4 7 risks can lead to hedge slippages and liquidity issues that can impair or prevent execution of hedging transactions. Macro-hedging techniques, which typically pre-position out of the money hedging assets, attracted more attention post GFC as companies looked to add new, or bolster existing, capabilities. Half of the survey respondents have introduced, or plan to introduce or strengthen, a macro-hedging program in response to the GFC. The need for effective counterparty credit risk management was emphasized during the GFC by a number of conspicuous counterparty defaults, or near defaults. Survey responses indicated some programmatic impact related to this experience, which is highlighted in Figure 6. Other notable post-gfc hedging program modifications include modifying the instruments used to hedge (indicated by 8% to 56% of participants, varying by guarantee) and hedging more Greeks (indicated by 1% to 19% of participants, varying by guarantee). Programmatic changes resulting from AG-4 were generally more muted. Two of the more notable changes evident in the survey results were: Some reduction in the number of Greeks hedged (indicated by 0% to 1% of participants, varying by guarantee) due to the deleterious effects some multiple Greek hedging strategies had on statutory reserve requirements A modification of instruments used in hedging (indicated by 1% to 19% of participants, again varying by guarantee) AG-4 responses found a mixed treatment of hedging within the statutory reserve valuation. Almost half the respondents did not recognize their hedging strategy within the valuation. Some of these companies strategies did not qualify as a clearly defined hedging strategy (CDHS) under the AG-4 requirements, while some companies strategies did qualify as a CDHS but were still not recognized within the valuation. 6 towerswatson.com
Towers Watson s View The GFC presented VA hedging programs with significant challenges. As a result of the crisis, VA providers have a more seasoned perspective regarding their risk management protocols, while program stakeholders expectations of both the day-to-day and longerterm effectiveness of these programs may be more accurately calibrated than in the past. The survey suggests an increased level of sophistication employed by hedging programs, including, but not limited to: The increased interest in and/or use of macro-hedging techniques Some increased focus on counterparty credit risk management The expansion of Greeks hedged in certain instances VA writers continue to take positive product development steps to develop new or tailor existing designs that better incorporate risk management-related considerations. These include target volatility funds and CPPI-like fund transfer dynamics. already difficult accounting mismatch problem. While a large number of writers continue to use simplistic interest rate and equity models for valuation, some have moved to more sophisticated models, and this may emerge as a trend for the future. Demands on hedging programs continue to be great, and we view best-in-class programs as having the operational discipline in place to promote continual program improvement and to establish tools and analytics to help confront these challenges head on. Further information The full report of survey findings is available to participants only. For help with questions about this article or the survey, please contact: Dave Czernicki +1 12 201 568 dave.czernicki@towerswatson.com David J. Maloof +1 212 09 764 david.maloof@towerswatson.com Challenges with respect to hedging still remain. The deleterious effects hedging has on statutory reserving requirements add complexity to an towerswatson.com 7
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. The information in this publication is of general interest and guidance. Action should not be taken on the basis of any article without seeking specific advice. Copyright 2011 Towers Watson. All rights reserved. TW-EU-2011-22047. October 2011. towerswatson.com