thema beleggingen by Willem van Ruitenburg and David Schrager HEDGING COMPLEX CROSS-GAMMA EXPOSURES: AN ELEGANT VANILLA ALTERNATIVE W. van Ruitenburg (boven) is Insurance Risk Officer bij ING Reinsurance. D. Schrager PhD is directeur Pricing & Product Hedging bij Nationale-Nederlanden. Abstract: Over the last decennia many insurance companies have sold insurance contracts, either in the pension or retail markets, with embedded guarantees on the performance of markets and/or investment funds. More and more insurance companies will report P&L on these books using Market Consistent valuation assumptions. Hedge programs are initiated to reduce P&L volatility. As a first step linear hedges, using e.g. futures and forward contracts, are often considered to eliminate the directional exposure towards market movements. More advanced hedge programs will include vanilla options to (partly) protect against bigger market returns and/or higher return volatility of single asset classes, i.e. so-called gamma/convexity matching, and to hedge for movements in market implied volatilities affecting the (Market Consistent) reserves of the embedded options in the insurance contracts. The investments underlying the guarantees often contain exposures in multiple equity indices, exchange rates and potentially bond funds. Insurance companies are therefore exposed to simultaneous movements across the different markets. The option risk is multivariate, rather than univariate in nature. In this article we will discuss an elegant approach to hedge these so-called cross-gamma exposures using vanilla option hedge strategies. Explanation Liability Movement in Greeks: Guarantees underlying e.g. variable annuity contracts and separate account guarantees in pension contracts are often on the performance of a basket of assets potentially including a variety of equity, exchange rates and bond exposures. [Eq 1]: dBSK (X1,..,XN,t ) = 兺i wi dXi,t The market value of the embedded guarantee (MVL) on a guarantee level K can be expressed as: [Eq 2]: MVL(BSK;K,T,Ω) = ΕQ 关 DFT ∙ Max { K - BSKT } 兴 The parameter Ω is the implied covariance matrix for the basket elements. This matrix is typically calibrated using historically observed correlations and implied volatilities for the individual asset classes, if available. First generation hedge programs target the change in MVL due to a change in the assets in the basket or Delta risks. Delta risks can be hedged using futures, forwards or the asset itself. Once Delta risks are hedged specific option, or non-linear, risks remain. Consider that when hedging an option one has to adjust the Delta hedge upon a move in the assets. This is due to convexity of the MVL as a function of BSK. In other words, MVL is not a linear function of BSK and hence a Delta hedge of a guarantee is not static but requires rebalancing. This article focusses on these non-linear risks, specifically Gamma. Just as Gamma in practice is defined as the change in notional Delta of the underlying due to a change in the underlying, Cross-Gamma is defined as the change in notional Delta of an underlying due to a change in another underlying. We will now provide more mathematical background to this explanation of CrossGamma. The market value movement of the MVL only as a function of immediate return on the basket can be approximated by a 2nd order Taylor approximation like: 34 de actuaris september 2014 thema beleggingen [Eq 3]: dMVL(dBSK) = ∂MVL ∂BSK 1 ∙ BSK ∙ dBSK + 2 ∙ ∂ 2 MVL ∙ ∂BSK 2 (BSK∙dBSK )2 + Ο-3 Equation [3] is more often displayed in terms of Greeks as: [Eq 4]: dMVL(dBSK) = ΔBSK ∙ dBSK + 21 ∙ ΓBSK ∙ (dBSK) 2 + Ο-3 Substituting [Eq 1] into [Eq 3] results in: [Eq 5]: dMVL(dX 1 ,..,X N ) = 兺i 共 ∂MVL ∂xi ∙ Xi ∙ dXi 兲 + 21 ∙ 关 兺i 兺j ∂∂xMVL ∙ ( Xi Xj ∙ dXi dXj ) 兴 + Ο-3 ixj 2 In terms of Greeks [Eq 5] translates into: [Eq 6]: dMVL(dX1,..,XN ) = ⌊兺i Δ Xi ∙ dXi ⌋+ 21 ∙⌊兺i ΓXi ∙ (dXi ) 2 ⌋+ 12 ∙ 关 兺i 兺j<>i ΓXiXj ∙ dXi dXj 兴 + Ο-3 The movement of the liability option value caused by immediate returns in the underlying assets contains a delta return, a gamma return and a cross-gamma return. For continuously rebalanced baskets it can be derived using the chain-rule of calculus that: Δ Xi = ΔBSK ∙ wi ΓXi = ΓBSK ∙ wi 2 ΓXi ,Xj = ΓBSK ∙ wiwj Following above relations the more elements the basket contains the greater the sum of all the cross-gamma exposures is relative to the single asset gamma exposures. For example in an equally weighted basket of N underlying components the single asset gammas only represent 1/N part of all the 2nd order exposures in the basket option. Hedging programs that focus on hedging the linear exposures will have a short position in gamma. This means options have to be bought to manage the risks in these portfolios. As these dynamically delta hedged positions will always end up in 'buying high and selling low' the delta hedge will typically lose money on a daily base. Typically the more futures/forwards need to be traded to remain delta neutral the higher the realized gamma loss. From Figure 1 it can be seen how short gamma positions leave hedge programs exposed to the size of the market return or better said to the level of realized market return volatility1. Gamma Liability Net Result Value Linear Hedge -20% 1 – It should be noted that these gamma losses are offset by releases of the market value reserve over time as time value ("Theta") runs out. -15% -10% -5% 0% 5% 10% 15% 20% Figure 1: Example of typical pay-off profile of linear hedge program Equivalently the expected result of open cross-gamma exposures is a function of the covariance among the pairs of underlying asset classes. On a daily base such positions will lose money whenever the different assets show big market movements in the same direction. de actuaris september 2014 35 thema beleggingen Increases in covariance can either be caused by 1) increased volatilities of individual assets or 2) increased correlation between the asset classes. The following regression will be performed: One of the better known hedge instruments to replicate these complex liability exposures are basket options. Basket options are however rather illiquid instruments for which the pricing is not transparent. This illiquidity typically increases when less standard equity indices and/or bonds are part of the basket. Traders who sell these baskets will typically hedge their short positions by going long options on the individual asset classes and will charge the buyer of the basket options generously for the open position/risk that the trader faces. The fitted coefficient âi can be seen as the gamma that has been realized. Realized Gamma approach: From [Eq 3] and [Eq 4] it is seen that for an option with price P on single underlying S the delta and gamma in money terms are defined by: ΔS = ∂P ∙ S ∂S 2 ΓS = ∂ P2 ∙ S2 ∂S Gamma exposures change on a daily base as markets move and the moneyness of the guarantee changes. This implies that the regression should be based on a limited interval of recent historical observations and that this window is being rolled forward. In Figure 2 the outcomes of conducted analyses are shown. The green dots are the actually observed corrected delta movements, i.e. dΔX t - ΔX ∙ dXi,t . The red dots represent the delta movements i, i,t explained by the individual gamma exposure, i.e. ΓX ∙ dXi,t and the i,t blue points show the fitted delta movements, ci + ai ∙ Xi,t Effective Gamma Example (2) Daily Change Delta Index X As traders will hedge their short cross-gamma positions using vanilla hedge strategies insurance companies could take a similar tactic to replicate cross-gamma exposure. In this article the 'Realized Gamma' approach will be further explained. [Eq 9]: dΔX t - ΔX ∙ dXi,t = ci + ai ∙ dXi,t + εi,t i, i,t -3% Via some calculus using the chain formula it's straightforward to derive that the changes in the notional- or money-deltas can be approximated by: [Eq 7]: dΔS ≈ ( ΓS+ΔS ) ∙ dS or dΔS - ΔS ∙ dS ≈ ΓS ∙ dS The above relationship is very sound: the change in the delta of the option is driven by the gamma of the option and the return of the underlying2. Replacing S with the basket BSK and substituting Eq[1] shows the explanation of option liability delta movement dΔX as: i 0% 1% 2% 3% 4% 5% In the analyses that have been conducted it has been observed that the correlation between the actual delta movements and the fitted delta movements is typically 70%-80% for baskets containing equity and exchange rates exposures. In Figure 3 it is shown that for embedded insurance guarantees on a basket of assets the force that drives the delta changes (and therefore trading in/out of linear hedges) may be a multiple of the theoretically calculated gamma for that single component as calculated by the liability model in case of average positive correlations. Effective Gamma Example (1) i The Realized Gamma approach is a regression based approach where the observed changes of the individual asset deltas are regressed on the individual asset returns. -1% Figure 2: Example of observed versus fitted delta movements [Eq 8]: dΔX = wi ∙ dΔBSK ≈ wi ∙ (ΓBSK + ΔBSK ) ∙ 兺 i wi dXi The change in delta for asset component Xi is driven by the return of the basket as a whole and not by the return of the individual exposures. -2% Value Gamma Fitted Realized Gamma Single assed-factor Gamma Time Figure 3: Fitted short realized versus calculated single asset gamma exposures over time 2 – It should be noted that in practice the absolute gamma in money terms is a much bigger than the absolute money-delta exposure. 36 de actuaris september 2014 thema beleggingen Insurance companies that are currently hedging the single asset gamma exposures can improve their hedge program by focusing hedging the realized gamma rather than the effective gamma. Advantages of this approach are: – Cross-gamma exposures can be hedged using simple vanilla options, no need to buy expensive and illiquid basket options. – Easy to implement, no need to explicitly calculate cross-gammas. Calculation time efficient as it re-uses Delta hedge program output. – Easy to understand concept, can be explained to management. Can be introduced as simple risk measure alongside use as explicit hedging tool. – Method relies on recent observed correlations among the riskdrivers and hence is an up-to-date measure of non-linear risks (see also disadvantages). Some disadvantages are: – Method relies on historically observed correlations among the riskdrivers, no guarantee that this correlation will remain in-tact. – Movements in (aggregated) liability delta can also be caused by e.g.: assumption changes, changes in in-force etc. This may either distort the validity of the observed fitted gamma or may require the time series to be cleaned-up. Conclusion Insurance companies write or have written guarantees on multiple underlying assets or asset classes. In this article we discuss a new technique for risk management of these products. The proposed 'Realized Gamma' approach may be the best solution for companies seeking (computational) efficient risk measurement and a hedge based on simple instruments. @ Reacties op dit artikel graag naar redactie.actuaris@ag-ai.nl