Bayesian inference for discretely observed diffusion processes Moritz Schauer with Frank van der Meulen and Harry van Zanten Delft University of Technology, University of Amsterdam Van Dantzig Seminar 1 / 25
Estimating parameters of a discretely observed diffusion process Diffusion process X d X t = b θ ( t, X t ) d t + σ θ ( t, X t ) d W t , X 0 = u, with transition densities p ( s, x ; t, y ) Discrete observations X t i = x i , 0 = t 0 < t 1 < · · · < t n . ◮ Bayesian estimate for parameter θ with prior π 0 ( θ ) . ◮ Likelihood is intractable (product of transition densities) ◮ Continuous time likelihood known in closed form (Girsanov’s theorem) 2 / 25
Computational approach Data Augmentation (DA): Sample from the joint posterior of missing data and parameter. 1. Sample diffusion bridges conditional on { X t i = x i } and θ (this gives “complete”, latent data); 2. Sample from θ conditional on the complete data. Can use an accept/reject or Metropolis-Hastings step. Rough outline : ◮ Simulation of diffusion bridges ◮ If unknown parameters are in the diffusion coefficient, DA does not work ◮ Example ◮ When and how to discretize 3 / 25
Computational approach Data Augmentation (DA): Sample from the joint posterior of missing data and parameter. 1. Sample diffusion bridges conditional on { X t i = x i } and θ (this gives “complete”, latent data); 2. Sample from θ conditional on the complete data. Can use an accept/reject or Metropolis-Hastings step. Rough outline : ◮ Simulation of diffusion bridges ◮ If unknown parameters are in the diffusion coefficient, DA does not work ◮ Example ◮ When and how to discretize 3 / 25
Computational approach Data Augmentation (DA): Sample from the joint posterior of missing data and parameter. 1. Sample diffusion bridges conditional on { X t i = x i } and θ (this gives “complete”, latent data); 2. Sample from θ conditional on the complete data. Can use an accept/reject or Metropolis-Hastings step. Rough outline : ◮ Simulation of diffusion bridges ◮ If unknown parameters are in the diffusion coefficient, DA does not work ◮ Example ◮ When and how to discretize 3 / 25
Computational approach Data Augmentation (DA): Sample from the joint posterior of missing data and parameter. 1. Sample diffusion bridges conditional on { X t i = x i } and θ (this gives “complete”, latent data); 2. Sample from θ conditional on the complete data. Can use an accept/reject or Metropolis-Hastings step. Rough outline : ◮ Simulation of diffusion bridges ◮ If unknown parameters are in the diffusion coefficient, DA does not work ◮ Example ◮ When and how to discretize 3 / 25
Examples: Butane dihedral angle, Pokern (2007) 5 4 angle 3 2 1 0.0 0.2 0.4 0.6 0.8 1.0 time J � d X t = θ i ψ i ( X t ) d t + d W t i =1 4 / 25
Chemical reaction network, Golightly and Wilkinson (2010) 20 15 variable DNA # 10 P P2 5 RNA 0 0 10 20 30 40 50 t � d X t = Sh θ ( X t ) d t + S diag( h θ ( X t )) d W t 5 / 25
Intuition: Diffusion bridge Two processes with equivalent distributions P and W ◮ Diffusion process X with σ ≡ 1 starting in u ◮ Brownian motion W starting in u Brownian motion W conditional on W T = v : Brownian bridge. The two conditional distributions P ⋆ and W ⋆ given X T = v resp. W T = v are equivalent d P ⋆ d W = p (0 , u ; T, v ) d P d W ⋆ φ (0 , u ; T, v ) with p and φ denoting the transition densities. Works only if σ is constant. More general bridge proposals X ◦ are needed, d P ⋆ d P ◦ ( X ◦ ) = C Ψ( X ◦ ) 6 / 25
Intuition: Diffusion bridge Two processes with equivalent distributions P and W ◮ Diffusion process X with σ ≡ 1 starting in u ◮ Brownian motion W starting in u Brownian motion W conditional on W T = v : Brownian bridge. The two conditional distributions P ⋆ and W ⋆ given X T = v resp. W T = v are equivalent d P ⋆ d W = p (0 , u ; T, v ) d P d W ⋆ φ (0 , u ; T, v ) with p and φ denoting the transition densities. Works only if σ is constant. More general bridge proposals X ◦ are needed, d P ⋆ d P ◦ ( X ◦ ) = C Ψ( X ◦ ) 6 / 25
Intuition: Diffusion bridge Two processes with equivalent distributions P and W ◮ Diffusion process X with σ ≡ 1 starting in u ◮ Brownian motion W starting in u Brownian motion W conditional on W T = v : Brownian bridge. The two conditional distributions P ⋆ and W ⋆ given X T = v resp. W T = v are equivalent d P ⋆ d W = p (0 , u ; T, v ) d P d W ⋆ φ (0 , u ; T, v ) with p and φ denoting the transition densities. Works only if σ is constant. More general bridge proposals X ◦ are needed, d P ⋆ d P ◦ ( X ◦ ) = C Ψ( X ◦ ) 6 / 25
Diffusion bridges Bridge from (0 , u ) to ( T, v ) d X ⋆ t = b ⋆ ( t, X ⋆ t ) d t + σ ( t, X ⋆ X ⋆ t ) d W t , 0 = u with drift ( a = σσ ′ ) b ⋆ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ◮ Delyon & Hu, Durham & Gallant : Proposals X ◦ of the form � � t ) + v − X ◦ t dX ◦ λb ( t, X ◦ d t + σ ( t, X ◦ X ◦ t = t ) d W t , 0 = u. T − t λ ∈ { 0 , 1 } . ◮ Beskos & Roberts : rejection sampling algorithm for obtaining bridges without discretisation error. 7 / 25
Diffusion bridges Bridge from (0 , u ) to ( T, v ) d X ⋆ t = b ⋆ ( t, X ⋆ t ) d t + σ ( t, X ⋆ X ⋆ t ) d W t , 0 = u with drift ( a = σσ ′ ) b ⋆ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ◮ Delyon & Hu, Durham & Gallant : Proposals X ◦ of the form � � t ) + v − X ◦ t dX ◦ λb ( t, X ◦ d t + σ ( t, X ◦ X ◦ t = t ) d W t , 0 = u. T − t λ ∈ { 0 , 1 } . ◮ Beskos & Roberts : rejection sampling algorithm for obtaining bridges without discretisation error. 7 / 25
Diffusion bridges Bridge from (0 , u ) to ( T, v ) d X ⋆ t = b ⋆ ( t, X ⋆ t ) d t + σ ( t, X ⋆ X ⋆ t ) d W t , 0 = u with drift ( a = σσ ′ ) b ⋆ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ◮ Delyon & Hu, Durham & Gallant : Proposals X ◦ of the form � � t ) + v − X ◦ t dX ◦ λb ( t, X ◦ d t + σ ( t, X ◦ X ◦ t = t ) d W t , 0 = u. T − t λ ∈ { 0 , 1 } . ◮ Beskos & Roberts : rejection sampling algorithm for obtaining bridges without discretisation error. 7 / 25
Diffusion bridge proposals Bridge from (0 , u ) to ( T, v ) d X ⋆ t = b ⋆ ( t, X ⋆ t ) d t + σ ( t, X ⋆ X ⋆ t ) d W t , 0 = u with drift b ⋆ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) Bridge from (0 , u ) to ( T, v ) d X ◦ t = b ◦ ( t, X ◦ t ) d t + σ ( t, X ◦ X ◦ t ) d W t , 0 = u with drift b ◦ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log ˜ p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ˜ Take ˜ p the transition density of � � d ˜ β ( t ) + ˜ ˜ B ( t ) ˜ X t = X t d t + ˜ σ ( t ) d W t . If ˜ a ( T ) = a ( T, v ) (and a few more conditions), then d P ⋆ d P ◦ ( X ◦ ) = ˜ p (0 , u ; T, v ) p (0 , u ; T, v )Ψ( X ◦ ) 8 / 25
Diffusion bridge proposals Bridge from (0 , u ) to ( T, v ) d X ◦ t = b ◦ ( t, X ◦ t ) d t + σ ( t, X ◦ X ◦ t ) d W t , 0 = u with drift b ◦ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log ˜ p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ˜ Take ˜ p the transition density of � � ˜ d ˜ β ( t ) + ˜ B ( t ) ˜ X t = X t d t + ˜ σ ( t ) d W t . If ˜ a ( T ) = a ( T, v ) (and a few more conditions), then d P ⋆ d P ◦ ( X ◦ ) = ˜ p (0 , u ; T, v ) p (0 , u ; T, v )Ψ( X ◦ ) where Ψ is tractable. 8 / 25
Diffusion bridge proposals Bridge from (0 , u ) to ( T, v ) d X ◦ t = b ◦ ( t, X ◦ t ) d t + σ ( t, X ◦ X ◦ t ) d W t , 0 = u with drift b ◦ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log ˜ p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ˜ Take ˜ p the transition density of � � ˜ d ˜ β ( t ) + ˜ B ( t ) ˜ X t = X t d t + ˜ σ ( t ) d W t . If ˜ a ( T ) = a ( T, v ) (and a few more conditions), then d P ⋆ d P ◦ ( X ◦ ) = ˜ p (0 , u ; T, v ) p (0 , u ; T, v )Ψ( X ◦ ) where Ψ is tractable. 8 / 25
Diffusion bridge proposals Bridge from (0 , u ) to ( T, v ) d X ◦ t = b ◦ ( t, X ◦ t ) d t + σ ( t, X ◦ X ◦ t ) d W t , 0 = u with drift b ◦ ( t, x ) = b ( t, x ) + a ( t, x ) ∇ x log ˜ p ( t, x ; T, v ) . � �� � r ( t, x ; T, v ) ˜ Take ˜ p the transition density of � � ˜ d ˜ β ( t ) + ˜ B ( t ) ˜ X t = X t d t + ˜ σ ( t ) d W t . If ˜ a ( T ) = a ( T, v ) (and a few more conditions), then d P ⋆ d P ◦ ( X ◦ ) = ˜ p (0 , u ; T, v ) p (0 , u ; T, v )Ψ( X ◦ ) where Ψ is tractable. 8 / 25
An example Example: Simulate X given that X 0 = 0 and X 1 = π/ 2 . d X t = (2 − 2 sin(8 X t )) d t + 1 2 d W t True MBB Delyon−Hu Guided Guided proposal from X t = 1 . 34 d t + 1 d ˜ 2 d W t . yielding � � t ) + π/ 2 − X ◦ d t + 1 t d X ◦ 2 − 2 sin(8 X ◦ X ◦ t = − 1 . 34 2 d W t , 0 = 0 . 1 − t 9 / 25
Finding good proposals ◮ Cross entropy method (previous example) ◮ Local linearizations (chemical reaction network example) ◮ Substituting space dependence for time dependence (next slide) 10 / 25
Substituting space dependence for time dependence d X t = − sin( X t ) d t, X 0 = π/ 2 d ˜ ˜ X t = − sech( t ) d t, X 0 = π/ 2 11 / 25
Recommend
More recommend