Banque nationale de Belgique SA · EURONEXT:BNB
Banque nationale de Belgique SA — Technical Presentation (October 2018)
76 slides from Banque nationale de Belgique SA's technical presentation, October 2018
Slide 1: Tractable Term Structure Models and the Zero Lower Bound
Tractable Term Structure Models and the Zero Lower Bound Bruno Feunou, Jean-Sébastien Fontaine, Anh Le, Chris Lundblad National Bank of Belgium October, 2018 Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 1 / 37
Slide 2: Motivation
Motivation One−month interest rate 0.02 0.018 0.016 0.014 0.012 0.01 0.008 0.006 0.004 0.002 0 200801 200901 201001 201101 201201 201301 Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 2 / 37
Slide 3: Motivation
Motivation Consider the Gaussian models: 0 − (δ0 + δ1 Xt ) − 12 λ0t λt −λ0t t+1 | {z } rt Mt+1 = e , λt = λ0 + λ1 Xt , Xt+1 = K0 + K1 Xt + Σt+1 , t+1 ∼ N(0, I ), Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 3 / 37
Slide 4: Motivation
Motivation Consider the Gaussian models: 0 − (δ0 + δ1 Xt ) − 12 λ0t λt −λ0t t+1 | {z } rt Mt+1 = e , λt = λ0 + λ1 Xt , Xt+1 = K0 + K1 Xt + Σt+1 , t+1 ∼ N(0, I ), are tractable: yn,t = An + Bn Xt ; Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 3 / 37
Slide 5: Motivation
Motivation Consider the Gaussian models: 0 − (δ0 + δ1 Xt ) − 12 λ0t λt −λ0t t+1 | {z } rt Mt+1 = e , λt = λ0 + λ1 Xt , Xt+1 = K0 + K1 Xt + Σt+1 , t+1 ∼ N(0, I ), are tractable: yn,t = An + Bn Xt ; but problematic: because when rt ≈ 0, model says 50% chance rt+1 will be negative. Feunou, Fontaine, Le
Slide 6: Motivation
Motivation Consider the Gaussian models: Consider the Black’s models: 0 0 − (δ0 + δ1 Xt ) − 12 λ0t λt −λ0t t+1 − max(δ0 + δ1 Xt , 0)− 12 λ0t λt −λ0t t+1 | {z } | {z } rt rt Mt+1 = e , Mt+1 = e λt = λ0 + λ1 Xt , λt = λ0 + λ1 Xt , Xt+1 = K0 + K1 Xt + Σt+1 , Xt+1 = K0 + K1 Xt + Σt+1 , t+1 ∼ N(0, I ), t
Slide 7: Motivation
Motivation Consider the Gaussian models: Consider the Black’s models: 0 0 − (δ0 + δ1 Xt ) − 12 λ0t λt −λ0t t+1 − max(δ0 + δ1 Xt , 0)− 12 λ0t λt −λ0t t+1 | {z } | {z } rt rt Mt+1 = e , Mt+1 = e λt = λ0 + λ1 Xt , λt = λ0 + λ1 Xt , Xt+1 = K0 + K1 Xt + Σt+1 , Xt+1 = K0 + K1 Xt + Σt+1 , t+1 ∼ N(0, I ), t
Slide 8: Motivation
Motivation Consider the Gaussian models: Consider the Black’s models: 0 0 − (δ0 + δ1 Xt ) − 12 λ0t λt −λ0t t+1 − max(δ0 + δ1 Xt , 0)− 12 λ0t λt −λ0t t+1 | {z } | {z } rt rt Mt+1 = e , Mt+1 = e λt = λ0 + λ1 Xt , λt = λ0 + λ1 Xt , Xt+1 = K0 + K1 Xt + Σt+1 , Xt+1 = K0 + K1 Xt + Σt+1 , t+1 ∼ N(0, I ), t
Slide 9: Motivation
Motivation More generally, in constructing no-arbitrage term structure models, we are often constrained by tractability considerations: I We tend to focus on a subset of sdfs Mt > 0 such that: P1,t = Et [Mt+1 ] is closed form, P2,t = Et [Mt+1 Mt+2 ] is closed form, ..., Pn,t = Et [Mt+1 Mt+2 ...Mt+n
Slide 10: Motivation
Motivation More generally, in constructing no-arbitrage term structure models, we are often constrained by tractability considerations: I We tend to focus on a subset of sdfs Mt > 0 such that: P1,t = Et [Mt+1 ] is closed form, P2,t = Et [Mt+1 Mt+2 ] is closed form, ..., Pn,t = Et [Mt+1 Mt+2 ...Mt+n
Slide 11: Summary of our approach
Summary of our approach We do not start from any explicit sdf Mt , Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 5 / 37
Slide 12: Summary of our approach
Summary of our approach We do not start from any explicit sdf Mt , Instead, we specify bond prices directly: P1,t = p1 (Xt ), for some analytical function p1 (.) P2,t = p2 (Xt ), for some analytical function p2 (.) ..., Pn,t = pn (Xt ), for some analytical function pn (.) With this, we are guarantee
Slide 13: Summary of our approach
Summary of our approach We do not start from any explicit sdf Mt , Instead, we specify bond prices directly: P1,t = p1 (Xt ), for some analytical function p1 (.) P2,t = p2 (Xt ), for some analytical function p2 (.) ..., Pn,t = pn (Xt ), for some analytical function pn (.) With this, we are guarantee
Slide 14: Summary of our approach
Summary of our approach Our approach is highly flexible: researchers have complete freedom in specifying the one-period bond price: any P1,t = p1 (Xt ) will work Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 6 / 37
Slide 15: Summary of our approach
Summary of our approach Our approach is highly flexible: researchers have complete freedom in specifying the one-period bond price: any P1,t = p1 (Xt ) will work I Imposing a lower bound on the short rate is straightforward: choose P1,t = p1 (Xt ) < 1 for all Xt Feunou, Fontaine, Le, Lundblad Tracta
Slide 16: Summary of our approach
Summary of our approach Our approach is highly flexible: researchers have complete freedom in specifying the one-period bond price: any P1,t = p1 (Xt ) will work I Imposing a lower bound on the short rate is straightforward: choose P1,t = p1 (Xt ) < 1 for all Xt I We can generate a wide range of tra
Slide 17: Summary of our approach
Summary of our approach Despite potential nonlinearity, our models can be extremely fast to estimate Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 7 / 37
Slide 18: Summary of our approach
Summary of our approach Despite potential nonlinearity, our models can be extremely fast to estimate The key intuition: with certain parameterizations, we can ”undo” the nonlinearity and translate our models back to the linear space: Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models Oct
Slide 19: Summary of our approach
Summary of our approach Despite potential nonlinearity, our models can be extremely fast to estimate The key intuition: with certain parameterizations, we can ”undo” the nonlinearity and translate our models back to the linear space: I For example, for the sake of intuition: 1/3 fn,t = (An + Bn Xt )
Slide 20: Summary of our approach
Summary of our approach Despite potential nonlinearity, our models can be extremely fast to estimate The key intuition: with certain parameterizations, we can ”undo” the nonlinearity and translate our models back to the linear space: I For example, for the sake of intuition: 1/3 fn,t = (An + Bn Xt )
Slide 21: Outline
Outline 1 Our construction of bond prices 2 How close are we to no arbitrage? 3 Closed form yields and forwards 4 Examples 5 Time series dynamics 6 Empirical Illustrations Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 8 / 37
Slide 22: 1. Our construction of bond prices
1. Our construction of bond prices The n-period zero-coupon bond price Pn is given recursively by P0 (Xt ) ≡1, (1) Pn (Xt ) =Pn−1 (g (Xt )) × exp(−m(Xt )), (2) for some functions m(·) and g (·). Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 9 / 37
Slide 23: 1. Our construction of bond prices
1. Our construction of bond prices The n-period zero-coupon bond price Pn is given recursively by P0 (Xt ) ≡1, (1) Pn (Xt ) =Pn−1 (g (Xt )) × exp(−m(Xt )), (2) for some functions m(·) and g (·). Example 1: P1 (Xt ) = P0 (g (Xt )) × exp(−m(Xt )) = exp(−m(Xt )) (3) I m(·) gives the one-period rate Feu
Slide 24: 1. Our construction of bond prices
1. Our construction of bond prices The n-period zero-coupon bond price Pn is given recursively by P0 (Xt ) ≡1, (1) Pn (Xt ) =Pn−1 (g (Xt )) × exp(−m(Xt )), (2) for some functions m(·) and g (·). Example 1: P1 (Xt ) = P0 (g (Xt )) × exp(−m(Xt )) = exp(−m(Xt )) (3) I m(·) gives the one-period rate Exa
Slide 25: 2. How close are we to no arb?
2. How close are we to no arb? Two relevant concepts: 1 No dominant trading strategies: a portfolio with strictly positive payoffs must have a strictly positive price Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 10 / 37
Slide 26: 2. How close are we to no arb?
2. How close are we to no arb? Two relevant concepts: 1 No dominant trading strategies: a portfolio with strictly positive payoffs must have a strictly positive price 2 No arbitrage opportunities: a portfolio with non-negative payoffs must have a strictly positive price I non-negative payoffs ≡ stri
Slide 27: 2. How close are we to no arb?
2. How close are we to no arb? Theorem 1: Our bond price construction allows no dominant trading strategies Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 11 / 37
Slide 28: 2. How close are we to no arb?
2. How close are we to no arb? Theorem 1: Our bond price construction allows no dominant trading strategies Proof. 1 Consider a portfolio: investment in n-period bond is wn (in face value) Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 11 / 37
Slide 29: 2. How close are we to no arb?
2. How close are we to no arb? Theorem 1: Our bond price construction allows no dominant trading strategies Proof. 1 Consider a portfolio: investment in n-period bond is wn (in face value) P I the price this period: n wn Pn (Xt ), Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models Octobe
Slide 30: 2. How close are we to no arb?
2. How close are we to no arb? Theorem 1: Our bond price construction allows no dominant trading strategies Proof. 1 Consider a portfolio: investment in n-period bond is wn (in face value) P I the price this period: Pn wn Pn (Xt ), n wn Pn−1 (Xt+1 ) > 0 for all Xt+1 ∈ X I the payoff next period: 2 T
Slide 31: 2. How close are we to no arb?
2. How close are we to no arb? Theorem 1: Our bond price construction allows no dominant trading strategies Proof. 1 Consider a portfolio: investment in n-period bond is wn (in face value) P I the price this period: Pn wn Pn (Xt ), n wn Pn−1 (Xt+1 ) > 0 for all Xt+1 ∈ X I the payoff next period: 2 T
Slide 32: 2. How close we are to ruling out no arb opportunities?
2. How close we are to ruling out no arb opportunities? Theorem 2: Our bond price construction ensures that bond portfolios with strictly non-negative payoffs cannot admit strictly negative prices. ( 0 Required by NA [ 0 Implied by our models Figure: Prices of portfolios with strictly non-negative p
Slide 33: 3. Closed form yields and forwards
3. Closed form yields and forwards The n-period yields and forward rates are given by n−1 X yn,t =(1/n) m(g ◦i (Xt )) (8) i=0 fn,t ≡(n + 1)yn+1,t − nyn,t = m(g ◦n (Xt )). (9) I g ◦n (Xt ) denotes g (g (....g (Xt )...)) (n times), Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October
Slide 34: 3. Closed form yields and forwards
3. Closed form yields and forwards The n-period yields and forward rates are given by n−1 X yn,t =(1/n) m(g ◦i (Xt )) (8) i=0 fn,t ≡(n + 1)yn+1,t − nyn,t = m(g ◦n (Xt )). (9) I g ◦n (Xt ) denotes g (g (....g (Xt )...)) (n times), I For example, if g (Xt ) = K1 Xt , then g ◦n (Xt ) = K1n Xt . With th
Slide 35: 4. Examples: the linear case
4. Examples: the linear case g (X ) = KX m(X ) = δ0 + δ10 X , (10) Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 14 / 37
Slide 36: 4. Examples: the linear case
4. Examples: the linear case g (X ) = KX m(X ) = δ0 + δ10 X , (10) ⇒ yn,t = An + (Bn /n)Xt , with An =δ0 Bn =Bn−1 K + δ10 . (11) Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 14 / 37
Slide 37: 4. Examples: the linear case
4. Examples: the linear case g (X ) = KX m(X ) = δ0 + δ10 X , (10) ⇒ yn,t = An + (Bn /n)Xt , with An =δ0 Bn =Bn−1 K + δ10 . (11) This is (essentially) the Nelson-Siegel model! Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 14 / 37
Slide 38: 4. Examples: the linear case
4. Examples: the linear case g (X ) = KX m(X ) = δ0 + δ10 X , (10) ⇒ yn,t = An + (Bn /n)Xt , with An =δ0 Bn =Bn−1 K + δ10 . (11) This is (essentially) the Nelson-Siegel model! Compare with the standard Gaussian no-arbitrage DTSM: 1 0 nAn =(n − 1)An−1 + δ0 − Bn−1 ΣBn−1 2 Bn =Bn−1 K1Q + δ10 . (12) Feu
Slide 39: 4. Examples: the Nelson-Siegel model
4. Examples: the Nelson-Siegel model Proposition 1: Suppose Xt ∈ R3 , with m(·) and g (·) given by h i −λ 1−e −λ m(Xt ) = 1 1−eλ −e −λ Xt , (13) λ 1 0 0 g (Xt ) = 0 e −λ λe −λ Xt , (14) 0 0 e −λ then, our bond prices construction implies yields with Nelson-Siegel loadings. Proof. Pn−1 ◦i ) yi
Slide 40: 4. The Nelson-Siegel model
4. The Nelson-Siegel model Bjork and Christensen (1999) and Filipovic (1999) show theoretically that the N-S model does not preclude all arbitrage opportunities. Despite this short-coming, the N-S model has been highly popular! Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2
Slide 41: 4. The Nelson-Siegel model
4. The Nelson-Siegel model Bjork and Christensen (1999) and Filipovic (1999) show theoretically that the N-S model does not preclude all arbitrage opportunities. Despite this short-coming, the N-S model has been highly popular! Christensen, Diebold and Rudebusch (2010) argue that the N-S model is “a
Slide 42: 4. The Nelson-Siegel model
4. The Nelson-Siegel model Bjork and Christensen (1999) and Filipovic (1999) show theoretically that the N-S model does not preclude all arbitrage opportunities. Despite this short-coming, the N-S model has been highly popular! Christensen, Diebold and Rudebusch (2010) argue that the N-S model is “a
Slide 43: 4. Examples: the linear-quadratic case
4. Examples: the linear-quadratic case g (X ) = KX m(X ) =δ0 + δ10 X + X 0 δ2 X (15) Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 17 / 37
Slide 44: 4. Examples: the linear-quadratic case
4. Examples: the linear-quadratic case g (X ) = KX m(X ) =δ0 + δ10 X + X 0 δ2 X (15) ⇒ quadratic yield coefficients: yn,t = An + (Bn /n)Xt + Xt0 (Cn /n)Xt An =δ0 Bn =Bn−1 K + δ10 Cn =K 0 Cn−1 K + δ2 , (16) Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 17 / 37
Slide 45: 4. Examples: the linear-quadratic case
4. Examples: the linear-quadratic case g (X ) = KX m(X ) =δ0 + δ10 X + X 0 δ2 X (15) ⇒ quadratic yield coefficients: yn,t = An + (Bn /n)Xt + Xt0 (Cn /n)Xt An =δ0 Bn =Bn−1 K + δ10 Cn =K 0 Cn−1 K + δ2 , (16) Compare with the standard Gaussian-quadratic no-arbitrage DTSM: 1 1 nAn =(n − 1)An−1 + δ0 − lo
Slide 46: 4. Examples: Black-style models
4. Examples: Black-style models Recall that the main idea here is to guarantee the positivity of yields In Black’s models, this is achieved by rt = max(0, δ0 + δ10 Xt ) Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 18 / 37
Slide 47: 4. Examples: Black-style models
4. Examples: Black-style models Recall that the main idea here is to guarantee the positivity of yields In Black’s models, this is achieved by rt = max(0, δ0 + δ10 Xt ) Within our framework: We choose the same g (X ) = KX We choose a generalized logistic transformation for the short rate function: δ
Slide 48: 4. Examples: Black-style models
4. Examples: Black-style models Our choice of m(X ) captures the spirit of the max(0, δ0 + δ10 Xt ) transformation (to guarantee positivity) in Black’s models: log(1+exp(s)) 6 5 4 3 2 1 0 ï4 ï3 ï2 ï1 0 1 2 3 4 5 6 s Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 19 / 37
Slide 49: 4. Examples: Black-style models
4. Examples: Black-style models We deliver analytical yields/ forwards: fn,t = u(θ, δ0 + δ10 K n Xt ) (18) where u(θ, s) captures the logistic transformation: u(θ, s) = θlog (1 + exp(s/θ)). This means that we can work with transformed forwards f˜n,t ≡ u −1 (θ, fn,t ) = δ0 + δ10 K n Xt . (19) and we
Slide 50: 5. Time series dynamics
5. Time series dynamics We have not touched time series dynamics This means that virtually any time series dynamics is acceptable in our framework and will not affect any of our proofs earlier. Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 21 / 37
Slide 51: 5. Time series dynamics
5. Time series dynamics We have not touched time series dynamics This means that virtually any time series dynamics is acceptable in our framework and will not affect any of our proofs earlier. Thus, we are flexible enough to accommodate: I GARCH-like or stochastic volatility I Unspanned factors I L
Slide 52: 5. Time series dynamics
5. Time series dynamics We have not touched time series dynamics This means that virtually any time series dynamics is acceptable in our framework and will not affect any of our proofs earlier. Thus, we are flexible enough to accommodate: I GARCH-like or stochastic volatility I Unspanned factors I L
Slide 53: 6. Empirical Illustrations
6. Empirical Illustrations We focus on the no-dominance (ND) versions of two Black’s style models: 1 BlackND : A model in which the states follow a Gaussian VAR(1). This is a close analogue to the traditional Black’s model. Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018
Slide 54: 6. Empirical Illustrations
6. Empirical Illustrations We focus on the no-dominance (ND) versions of two Black’s style models: 1 BlackND : A model in which the states follow a Gaussian VAR(1). This is a close analogue to the traditional Black’s model. 2 SV-BlackND : A model in which the states dynamics features stochastic vola
Slide 55: 6. Empirical Illustrations
6. Empirical Illustrations We focus on the no-dominance (ND) versions of two Black’s style models: 1 BlackND : A model in which the states follow a Gaussian VAR(1). This is a close analogue to the traditional Black’s model. 2 SV-BlackND : A model in which the states dynamics features stochastic vola
Slide 56: 6. Empirical Illustrations
6. Empirical Illustrations We focus on the no-dominance (ND) versions of two Black’s style models: 1 BlackND : A model in which the states follow a Gaussian VAR(1). This is a close analogue to the traditional Black’s model. 2 SV-BlackND : A model in which the states dynamics features stochastic vola
Slide 57: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment We use the linear rational model of Filipovic, Larson, and Trolle (JF 2017) as a DGP to generate 100 samples of yields data that exhibit: Feunou, Fontaine, Le, Lundblad Tractable Term Structure Models October 2018 23 / 37
Slide 58: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment We use the linear rational model of Filipovic, Larson, and Trolle (JF 2017) as a DGP to generate 100 samples of yields data that exhibit: I salient features of the yield data in the U.S., such as bond returns predictability, the shapes of the yiel
Slide 59: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment We use the linear rational model of Filipovic, Larson, and Trolle (JF 2017) as a DGP to generate 100 samples of yields data that exhibit: I salient features of the yield data in the U.S., such as bond returns predictability, the shapes of the yiel
Slide 60: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment We use the linear rational model of Filipovic, Larson, and Trolle (JF 2017) as a DGP to generate 100 samples of yields data that exhibit: I salient features of the yield data in the U.S., such as bond returns predictability, the shapes of the yiel
Slide 61: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment We use the linear rational model of Filipovic, Larson, and Trolle (JF 2017) as a DGP to generate 100 samples of yields data that exhibit: I salient features of the yield data in the U.S., such as bond returns predictability, the shapes of the yiel
Slide 62: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment 1-m mean 5-yr vol 10-yr 400 1800 1600 380 1400 360 1200 basis points 1000 340 800 320 600 400 300 200 280 0 0 100 200 300 400 500 600 700 0 20 40 60 80 100 120 months maturity (in months) (a) Simulated yields (b) Average Means and Vols of Simulate
Slide 63: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment Full-Sample h mat GaussianNA BlackNA BlackND SV-BlackND 1-m 10.8 0.91 0.96 0.75∗ 2-yr 8.0 0.96 0.93 0.81∗ 3-m 5-yr 7.9 1.00 0.90 0.82∗ 10-yr 7.8 1.00 0.97 0.75∗ 1-m 27.5 0.94 0.92 0.82∗ 2-yr 26.1 0.97 0.91 0.77∗ 1-yr 5-yr 27.8 1.00 0.91 0.72∗ 10-y
Slide 64: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment ZLB Sample h mat GaussianNA BlackNA BlackND SV-BlackND ∗ 1-m 7.8 0.69 0.58 0.69 2-yr 7.5 0.76 0.49∗ 0.61 3-m 5-yr 7.2 1.01 0.74 0.62∗ 10-yr 6.8 0.91 0.91 0.65∗ 1-m 24.0 0.67 0.47∗ 0.74 2-yr 23.4 0.87 0.66 0.54∗ 1-yr 5-yr 24.1 0.97 0.81 0.61∗ 10-yr
Slide 65: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment Full-Sample ZLB Sample h mat GNA BNA BND SV-BND GNA BNA BND SV-BND 1-m 54.3 0.30 0.51 0.29∗ 68.9 0.15 0.14∗ 0.15 2-yr 30.9 0.50 0.63 0.34∗ 36.8 0.34 0.18∗ 0.20 3-m 5-yr 19.4 0.76 0.74 0.48∗ 25.2 0.42 0.42 0.29∗ 10-yr 15.0 0.62 0.94 0.59∗ 20.1 0.28
Slide 66: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment Full-Sample ZLB Sample h mat GNA BNA BND SV-BND GNA BNA BND SV-BND 6-m 0.2 1.62 1.27 0.84∗ 0.3 0.96 0.60∗ 0.73 2-yr 0.5 1.59 0.98 0.75∗ 0.6 1.06 0.51 0.51∗ 3-m 5-yr 0.8 1.42 0.93 0.72∗ 1.0 1.13 0.68 0.46∗ 9-yr 1.2 1.24 0.93 0.72∗ 1.3 1.01 0.82 0.6
Slide 67: 6. Empirical Illustrations – a Simulated Environment
6. Empirical Illustrations – a Simulated Environment Full-Sample ZLB Sample h mat GNA BNA BND SV-BND GNA BNA BND SV-BND 6m-9y 0.5 1.63 1.01 0.77∗ 0.7 1.04 0.49 0.45∗ 3-m 6m-4y 0.8 1.46 0.98 0.74∗ 1.0 1.10 0.73 0.49∗ 5y-9y 1.0 1.33 0.93 0.71∗ 1.2 1.04 0.76 0.52∗ 6m-9y 0.4 1.61 1.01 0.76∗ 0.4 1.39 0.7
Slide 68: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data 1-month rate - Data 1-month shadow rate - Black na 1500 1-month shadow rate - Black nd 1-month shadow rate - SV-Black nd 1000 Basis points 500 0 -500 1970Jan 1980Jan 1990Jan 2000Jan 2010Jan 2020Jan time Figure: U.S. 1-month rate and 1-month shadow ra
Slide 69: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data 1-month rate - Data 1-month shadow rate - Black na 200 1-month shadow rate - Black nd 1-month shadow rate - SV-Black nd 100 0 Basis points -100 -200 -300 -400 2008Jan 2009Jan 2010Jan 2011Jan 2012Jan 2013Jan 2014Jan 2015Jan 2016Jan time Figure: U.S. 1
Slide 70: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data 200 Gaussian Black na Black nd 150 SV-Black nd Basis points 100 50 0 2008Jan 2009Jan 2010Jan 2011Jan 2012Jan 2013Jan 2014Jan 2015Jan 2016Jan (a) 3-month forecast horizon Gaussian 200 Black na Black nd 150 Basis points SV-Black nd 100 50 0 -50 2008Jan
Slide 71: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data RV Black na 150 Black nd SV-Black nd Basis points 100 50 0 1970Jan 1975Jan 1980Jan 1985Jan 1990Jan 1995Jan 2000Jan 2005Jan 2010Jan 2015Jan 2020Jan (a) 3-month forecast horizon 350 RV Black na 300 Black nd 250 Basis points SV-Black nd 200 150 100 50 0
Slide 72: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data Entire Sample ZLB Sample BNA BND SV-BND BNA BND SV-BND 3-m Average 25.02 0.75 0.58∗ 11.61 1.19 0.64∗ 12-m Average 45.81 0.81 0.67∗ 28.84 1.04 0.78∗ Table: RMSE (in basis points) btw RV and volatility forecasts Feunou, Fontaine, Le, Lundblad Tractable
Slide 73: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data Entire Sample ZLB Sample BNA BND SV-BND BNA BND SV-BND 3-m Average 0.26 0.25 0.79∗ 0.67 0.62 0.80∗ 12-m Average 0.25 0.24 0.79∗ 0.59 0.58 0.80∗ Table: Correlation between volatility forecasts and RV Feunou, Fontaine, Le, Lundblad Tractable Term Struc
Slide 74: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data Gaussianna Black na Black nd 1000 SV-Black nd 800 Basis points 600 400 200 0 1970Jan 1980Jan 1990Jan 2000Jan 2010Jan 2020Jan time Figure: EH component of 120-m yields over sample period 1970:Jan - 2015:Dec Feunou, Fontaine, Le, Lundblad Tractable Ter
Slide 75: 6. Empirical Illustrations – historical U.S. data
6. Empirical Illustrations – historical U.S. data Black na 100 Black nd SV-Black nd 80 60 Basis points 40 20 0 -20 1970Jan 1980Jan 1990Jan 2000Jan 2010Jan 2020Jan time Figure: EH component of 120-m yields, relative to Gaussian model, over sample period 1970:Jan - 2015:Dec Feunou, Fontaine, Le, Lundb
Slide 76: Conclusion
Conclusion Propose an alternative approach to designing new tractable term structure models: 1 We specify bond prices directly without going through any sdf. 2 We then verify that these prices are free of dominant trading strategies. Imposition of lower bounds on yields is straightforward yet mainta
Banque Nationale De Belgique — Technical Presentation (October 2018) on 6ix