1 Convexity Adjustment Calculation Interest Rate Futures Convexity is an attribute of an interest rate instrument which measures how the instruments' duration or sensitivity to rate movements (i.e. DV01) changes as rate levels change. Convexity is often referred to as the second derivative of price with respect to rate level. A Convexity adjustment is a general concept important in many Fixed-Income derivative models whereby the rate of the instrument is adjusted to reflect the applicable convexity for that instrument. The definition of this adjustment varies depending on the context of the derivative instrument and there is no one standard model. Most interest rate futures, such as Eurodollars, trade at a fixed value per basis point and thus have zero convexity. In contrast, a forward rate agreement or a swap containing a fixed-rate leg exhibits positive convexity similar to that of a bullet bond. Therefore, the rates implied by futures contracts are considered “expected rates” whereas FRA’s and swaps are consistent with executable forwards, which encompass convexity. The purpose of this document is to describe convexity adjustment whereby the interest rate future is converted into an equivalent forward rate agreement (FRA). This adjustment generates a lower rate for the futures contract, and becomes more pronounced the longer the expiry date of the futures contract. 2 Convexity Correction Formulas From a quantitative point of view the origin of the convexity correction can be expressed as the fact that futures and forwards are expected values of the forward rate but under different measures. The futures rate, due to the daily margining feature is the expected value of the forward rate under the risk neutral measure. This differs from the forward rate and is a model dependent quantity, that is, it depends on the dynamics of the interest rate. A common approach to compute convexity corrections is to use so called short rate models for the interest rate evolution. A popular choice corresponds to the use of the Hull-White model which postulates that the short rate under the risk neutral measure follows a normal diffusion process with volatility σ and speed of mean reversion a. The relation between (simple compounded) forward rate 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) from 𝑇1 to 𝑇2 and the observable (simple compounded) future rate 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) over the same period is: 1 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) ≈ 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) − [1 − 𝑒𝑥𝑝(−X(𝑇1 , 𝑇2 ))] [𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) + ] 𝛿 where 𝛿 = 𝑇2 − 𝑇1 X(𝑇1 , 𝑇2 ) = 𝐵(𝑡, 𝑇) = σ2 B(𝑇1 , 𝑇2 )(B(𝑇1 , 𝑇2 )(1 − e−2a𝑇1 ) + 𝑎B2 (0, 𝑇1 )) 2a 1 − e−a(T−t) 𝑎 Derivation of this formula is provided in the appendix. 3 Convexity Adjustment Settings Mean Reversion Speed The mean reversion speed a represents the speed with which the short term rate reverts to its long term mean. The typical range of values for mean reversion is 0.001 for negligible effects to 0.1, which is a relatively high mean reversion. Its inverse 1/a is the time scale on which mean reversion takes place. Rate Volatility The value used as default for the short rate volatility σ is chosen as the approximate normal (Bachelier) 10 year cap volatility obtained by multiplying the 10 year Black volatility by the corresponding swap rate. Rate volatility = (10-yr swap rate {USSWAP10}) * (10-yr cap volatility {USCV10}) /100 For each currency, different tenors may be used for rate volatility. The choice of which tenors is based on the unique set of market conditions (i.e. liquidity) for each swap curve and individual tenors. One recent market condition impacting determination of what market data to use is the continued existence of negative rates. As a result, IR VOL market players have been migrating to normal volatilities for trading; therefore, EUR and CHF was recently enhanced to use normal volatility. For other currencies (besides EUR and CHF) if the swap rate becomes negative then the forward swap rate and Swaption volatility is used for rate volatility. Below we list several currencies and which tenors are used: USD: USCV10 * USSWAP10 EUR: EUNS06 BVOL GBP: BPNS06 BVOL JPY: JYCV5 * JYSWAP5 CHF: SFNS03 BBIR CAD: CDCV03* CDSW3 AUD: ADCV5* ADSWAP5 SEK: SKSN055 BBIR SGD: SDCV2* SDSW2 HKD: HDCV2* HDSW2 NZD: NDSN011 BBIR 4 Bloomberg Screens SWDF saves the user default on Rate Volatility and Mean Rev Speed. 5 EDSF and EDS always assumes 0.03 as default for Mean Rev Speed, and use cap volatility for the currency. EDS<go> select 1 For more information, please press the <HELP> key twice on the BLOOMBERG PROFESSIONAL® service. 6 Appendix: Derivation of Convexity Correction Formulas In the Hull-White model with mean reversion a and constant volatility σ, 𝑑𝑟(𝑡) = (𝜃(𝑡) − 𝑎 𝑟(𝑡)) 𝑑𝑡 + 𝜎 𝑑𝑊(𝑡) where 𝜃(𝑡) is chosen to fit the interest rate term structure. The solution to this SDE is: t 𝑟(𝑡) = α(t) + σ ∫ e−a(t−u) dW(u) 0 where α(t) = f M (0, t) + σ2 (1 − e−a t )2 2a2 and 𝑓 𝑀 (0, 𝑡) is the market instantaneous forward rate at time 0 for the maturity T. Some notation before proceeding to the derivation of convexity correction. Denote the coverage period 𝛿 of the forward rate from 𝑇1 to 𝑇2 and the function 𝐵(𝑡, 𝑇) as: 𝛿 ≝ 𝑇2 − 𝑇1 1 − e−a(T−t) 𝑎 𝐵(𝑡, 𝑇) ≝ The time 0 value of a zero coupon bond maturity at T can be expressed as following under the risk neutral measure Q: 𝑇 𝑃(0, 𝑇) = 𝐸 𝑄 [𝑒𝑥𝑝 (− ∫ 𝑟(𝑡)𝑑𝑡)] 0 𝑇 𝑇 t = 𝐸 𝑄 [𝑒𝑥𝑝 (− ∫ α(t)dt − σ ∫ ∫ e−a(t−u) dWu 𝑑𝑡)] 𝑇 0 0 𝑄 0 𝑇 t = 𝑒𝑥𝑝 (− ∫ α(t)dt) 𝐸 [𝑒𝑥𝑝 (−σ ∫ ∫ e−a(t−u) dWu 𝑑𝑡)] 0 𝑇 0 0 𝑇 T 0 𝑇 0 u 𝑇 = 𝑒𝑥𝑝 (− ∫ α(t)dt) 𝐸 𝑄 [𝑒𝑥𝑝 (−σ ∫ ∫ e−a(t−u) dt dWu )] = 𝑒𝑥𝑝 (− ∫ α(t)dt) 𝐸 𝑄 [𝑒𝑥𝑝 (−σ ∫ B(u, T) dWu )] 0 𝑇 = 𝑒𝑥𝑝 (− ∫ α(t)dt + 0 2 𝑇 0 σ ∫ B2 (t, T) 𝑑𝑡) 2 0 7 The time zero (simple compounded) forward rate is related to the bond price as: 1 + 𝛿 ∗ 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) = 𝑃(0, 𝑇1 ) 𝑃(0, 𝑇2 ) 𝑇2 σ2 𝑇1 2 σ2 𝑇2 2 = 𝑒𝑥𝑝 (∫ α(t)dt + ∫ B (t, 𝑇1 ) 𝑑𝑡 − ∫ B (t, 𝑇2 ) 𝑑𝑡) 2 0 2 0 𝑇1 As described above, the time zero (simple compounded) future rate under the risk neutral measure is given by: = 1 + 𝛿 ∗ 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) 𝑇2 𝑄 = 𝐸 [𝑒𝑥𝑝 (∫ 𝑟(𝑡)𝑑𝑡)] 𝑇1 𝑇2 𝑇2 t = 𝐸 𝑄 [𝑒𝑥𝑝 (∫ α(t)dt + σ ∫ ∫ e−a(t−u) dWu 𝑑𝑡)] 𝑇1 𝑇2 𝑇1 0 𝑇2 t = 𝑒𝑥𝑝 (∫ α(t)dt) 𝐸 𝑄 [𝑒𝑥𝑝 (σ ∫ ∫ e−a(t−u) dWu 𝑑𝑡)] 𝑇1 𝑇1 𝑇2 𝑇1 0 𝑇2 𝑇2 𝑇2 = 𝑒𝑥𝑝 (∫ α(t)dt) 𝐸 𝑄 [𝑒𝑥𝑝 (σ ∫ ∫ e−a(t−u) dt 𝑑dWu + σ ∫ ∫ e−a(t−u) dt 𝑑dWu )] 𝑇1 0 𝑇2 𝑇1 𝑇1 𝑇1 u 𝑇2 = 𝑒𝑥𝑝 (∫ α(t)dt) 𝐸 𝑄 [𝑒𝑥𝑝 (σ ∫ B(u, 𝑇2 ) − B(u, 𝑇1 ) dWu + σ ∫ B(u, 𝑇2 )dWu )] 𝑇1 0 𝑇2 = 𝑒𝑥𝑝 (∫ α(t)dt + 𝑇1 𝑇1 σ2 𝑇1 σ2 𝑇2 2 ∫ (B(t, 𝑇2 ) − B(t, 𝑇1 )) 𝑑𝑡 + ∫ B 2 (t, 𝑇2 ) 𝑑𝑡) 2 0 2 𝑇1 The ratio of the future rate and forward rate expressions is: = 1 + 𝛿 ∗ 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) 1 + 𝛿 ∗ 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) = 𝑒𝑥𝑝 ( 𝑇1 𝑇2 𝑇1 𝑇2 σ2 2 (∫ (B(t, 𝑇2 ) − B(t, 𝑇1 )) 𝑑𝑡 + ∫ B 2 (t, 𝑇2 ) 𝑑𝑡 − ∫ B 2 (t, 𝑇1 ) 𝑑𝑡 + ∫ B 2 (t, 𝑇2 ) 𝑑𝑡)) 2 0 𝑇1 0 0 𝑇1 𝑇2 = 𝑒𝑥𝑝 (σ2 ∫ B 2 (t, 𝑇2 ) − B(t, 𝑇1 )B(t, 𝑇2 ) 𝑑𝑡 ) 𝑒𝑥𝑝 (σ2 ∫ B 2 (t, 𝑇2 ) 𝑑𝑡) 0 𝑇1 𝑇 We approximate this by dropping the second exponential 𝑒𝑥𝑝 (σ2 ∫𝑇 2 B2 (t, 𝑇2 ) 𝑑𝑡). 1 In the case of large 𝑇1 , this term would be small compare to the term in the first 𝑇 exponential 𝑒𝑥𝑝 (σ2 ∫0 1 B2 (t, 𝑇2 ) 𝑑𝑡). In the case of small 𝑇1 and 𝑇2 , the overall convexity adjustment would be small anyway and the effect of omitting this term would be negligible. Thus, we have the following approximation: 8 = 𝑇1 1 + 𝛿 ∗ 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) ≈ 𝑒𝑥𝑝 (σ2 ∫ B 2 (t, 𝑇2 ) − B(t, 𝑇1 )B(t, 𝑇2 ) 𝑑𝑡) 1 + 𝛿 ∗ 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) 0 The expression inside this exponential function is equivalent to the following function: 𝑇1 X(𝑇1 , 𝑇2 ) ≝ σ2 ∫ B 2 (t, 𝑇2 ) − B(u, 𝑇1 )B(u, 𝑇2 ) 𝑑𝑡 0 σ2 = B(𝑇1 , 𝑇2 )(B(𝑇1 , 𝑇2 )(1 − e−2a𝑇1 ) + 𝑎B 2 (0, 𝑇1 )) 2a With further arrangement of above equation, we have the following relationship between the (simple compounded) forward rate and the observable (simple compounded) future rate: 1 + 𝛿 ∗ 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) ≈ 𝑒𝑥𝑝(X(𝑇1 , 𝑇2 )) 1 + 𝛿 ∗ 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) 1 𝑓𝑤𝑑0 (𝑇1 , 𝑇2 ) ≈ 𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) − [1 − 𝑒𝑥𝑝(−X(𝑇1 , 𝑇2 ))] [𝑓𝑢𝑡0 (𝑇1 , 𝑇2 ) + ] 𝛿