Monetary Policy & Asset Prices David Backus, Mikhail Chernov, and Stanley Zin NYU Macro Lunch | March 8, 2012 This version: March 5, 2012 Backus, Chernov, & Zin (NYU, LSE) Monetary policy 1 / 24 The problem Model: forward-looking difference equation yt = λEt yt+1 + x1t + x2t Questions I I I I How do we identify λ? Do we need to observe the shocks (x1 , x2 )? Are forecasts and asset prices helpful? Cross-equation restrictions? Connection to methods that identify shocks? Answers I I I I We need “enough” observables and structure We need to observe something, but one shock can be enough Forecasts, asset prices, and economic structure all help Identifying λ and x closely related Backus, Chernov, & Zin (NYU, LSE) Monetary policy 1 / 24 Example 1: independent shocks Example 1: independent shocks Independent moving average shocks Vary what we observe Apply to model with Taylor rule Backus, Chernov, & Zin (NYU, LSE) Monetary policy 2 / 24 Example 1: independent shocks Example 1: independent shocks Model yt = λEt yt+1 + xt xt = α(L)wt , wt ∼ NID(0, 1) [|λ| < 1, α one-sided and square-summable, w fundamental for x] Solution (agents observe everything) yt = β(L)wt , βj (α, λ) = αj + λαj+1 + λ2 αj+2 · · · What do we need to estimate λ? Backus, Chernov, & Zin (NYU, LSE) Monetary policy 3 / 24 Example 1: independent shocks Observable shock (x, y ) I I Joint process but degenerate (same innovations) Cross-equation restrictions on parameters (β depends on λ, α) Observe variable y (always) I Estimate β Observe shock x I I I Estimate α Use mapping β(α, λ) to infer λ λ overdetermined if x is MA(q) and q > 1 Backus, Chernov, & Zin (NYU, LSE) Monetary policy 4 / 24 Example 1: independent shocks Hidden shock (x, y ) still joint process, but we don’t observe x Observe variable y I I Estimate β But without x, we can’t distinguish roles of α and λ Do forecasts help? ftk = Et yt+k = [ ] β(L)/Lk wt + [Nope, depends only on β, can’t tell us about (λ, α)] Backus, Chernov, & Zin (NYU, LSE) Monetary policy 5 / 24 Example 1: independent shocks Partly observable shock Model and solution ⇒ yt = λEt yt+1 + x1t + x2t xit = αi (L)wit , yt = β1 (L)w1t + β2 (L)w2t (w1t , w2t ) ∼ NID(0, I ) Observe variable y and shock x1 — but not x2 I I Estimate α1 and β1 Use mapping β1 (α1 , λ) to infer λ Do forecasts help? I I They help us identify the wi ’s Then we can recover x2 from the difference equation Backus, Chernov, & Zin (NYU, LSE) Monetary policy 6 / 24 Example 1: independent shocks Model with Taylor rule Model (p = inflation, i = nominal interest rate) it = r + Et pt+1 + x1t (Euler equation) it = r + τ pt + x2t xit = αi (L)wit , (Taylor rule) (w1t , w2t ) ∼ NID(0, I ) Solution pt = β1 (L)w1t + β2 (L)w2t it = r + [β1 (L)/L]+ w1t + [β2 (L)/L]+ w2t + x1t What do we need to estimate λ = τ −1 ? Backus, Chernov, & Zin (NYU, LSE) Monetary policy 7 / 24 Example 1: independent shocks Taylor rule: observable shocks (x1 , x2 , p, i) I I Joint process but degenerate (two-dimensional innovation) Cross-equation restrictions on parameters Observe shocks (x1 , x2 ) I Estimate αi ’s Observe variables (p, i) I I Estimate βi ’s Use mapping βi (αi , λ) to infer λ = τ −1 Backus, Chernov, & Zin (NYU, LSE) Monetary policy 8 / 24 Example 1: independent shocks Taylor rule: hidden shock (“Cochrane version”) Shocks: x1 = 0, x2 hidden Observe variable p I I Estimate β2 (possible because shock is one dimensional) Can’t distinguish roles of α2 and λ Does extra variable i help? I No, contains no information not in p Do forecasts help? I No, they add no information beyond p Backus, Chernov, & Zin (NYU, LSE) Monetary policy 9 / 24 Example 1: independent shocks Taylor rule: partly observable shock (“Gertler version”) Observe variables (p, i) and shock x1 I I Estimate α1 and β1 Use mapping β1 (α1 , λ) to infer λ Key insight I i gives us an additional observable when x1 ̸= 0 How might we observe x1 ? I I Observe Et pt+1 directly — or derive from VAR in (p, i) Then back x1 out of Euler equation Backus, Chernov, & Zin (NYU, LSE) Monetary policy 10 / 24 Example 1: independent shocks Example 1: summary Identification may be possible with partial observability Observability facilitated by forecasts Independence needs further investigation Backus, Chernov, & Zin (NYU, LSE) Monetary policy 11 / 24 Example 2: correlated shocks Example 2: correlated shocks Correlated shocks in state-space model Vary what we observe Apply to model with Taylor rule Backus, Chernov, & Zin (NYU, LSE) Monetary policy 12 / 24 Example 2: correlated shocks Example 2: correlated shocks Model yt = λEt yt+1 + e ⊤ xt xt+1 = Axt + Cwt+1 , {wt } ∼ NID(0, I ) [|λ| < 1, A stable & “regular”] Solution yt = e ⊤ (I − λA)−1 xt = β ⊤ xt What do we need to estimate λ? Backus, Chernov, & Zin (NYU, LSE) Monetary policy 13 / 24 Example 2: correlated shocks Observable state and shock Observe state x and vector e (hence shock e ⊤ x) I Estimate A Observe variable y I Estimate β, use A to infer λ λβ ⊤ = (β − e)⊤ A−1 [λ typically overdetermined] Forecasts redundant ftk = Et yt+k Backus, Chernov, & Zin (NYU, LSE) = β ⊤ Ak xt Monetary policy 14 / 24 Example 2: correlated shocks Digression on hidden shock & state What’s hidden, state x or shock e ⊤ x? One line of thought (Hansen-Sargent) I I Only part of state observed → may not be able to back out innovations [??] We suggest: observe the state, not the shock I I Lots of econometric approaches Or: build it from forecasts of y ft = [yt , ft1 , · · · , ftn−1 ]⊤ ft spans state x if A is regular and n = dim(x) Backus, Chernov, & Zin (NYU, LSE) Monetary policy 15 / 24 Example 2: correlated shocks Hidden shock Shock e ⊤ x not observed (e not known) Observe variable y , state x I I I From y : estimate β From x: estimate A But since we don’t know e, can’t infer λ Do forecasts help? I They span the state, not otherwise helpful Backus, Chernov, & Zin (NYU, LSE) Monetary policy 16 / 24 Example 2: correlated shocks Partly observable shock Model ⇒ yt = λEt yt+1 + e1⊤ xt + e2⊤ xt yt = (e1 + e2 )⊤ (I − λA)−1 xt = β ⊤ xt Observe variable y , state x I I From y : estimate β From x: estimate A Is it enough to observe shock e1⊤ x? I Can’t easily disentangle effects of e1 and e2 Backus, Chernov, & Zin (NYU, LSE) Monetary policy 17 / 24 Example 2: correlated shocks Partly observable shock (continued) Reminder: solution is β ⊤ = (e1 + e2 )⊤ (I − λA)−1 What we know: β (n knowns) What we don’t know: e1 , λ (n + 1 unknowns) Needed: one or more restrictions I I I Uncorrelated shocks Other information that restricts the shocks Tight economic structure Backus, Chernov, & Zin (NYU, LSE) Monetary policy 18 / 24 Example 2: correlated shocks Model with Taylor rule Model (p = inflation, i = nominal interest rate) it = r + Et pt+1 + e1⊤ xt it = r + τ pt + (Euler equation) e2⊤ xt xt+1 = Axt + Cwt+1 , (Taylor rule) {wt } ∼ NID(0, I ) Solution: pt = β ⊤ xt β ⊤ = (e1 − e2 )⊤ (τ I − A)−1 What do we need to estimate λ = τ −1 ? One restriction on e2 Backus, Chernov, & Zin (NYU, LSE) Monetary policy 19 / 24 Example 2: correlated shocks Example 2: summary Forecasts helpful in observing state Identification then possible with I I Partial observability of shocks Restriction(s) on hidden (monetary policy) shock Backus, Chernov, & Zin (NYU, LSE) Monetary policy 20 / 24 Example 3: bond pricing models Representative agent model Model it = − log Et [mt+1 exp(−pt+1 )] e2⊤ xt it = r + τ pt + log mt+1 = −δ − α log gt+1 (Euler equation) (Taylor rule) (pricing kernel) log gt+1 = g + e1⊤ xt xt+1 = Axt + Cwt+1 , {wt } ∼ NID(0, I ) Solution: pt = β ⊤ xt β ⊤ = (αe1 − e2 )⊤ (τ I − A)−1 r = δ + αg − (αe1 + β)⊤ CC ⊤ (αe1 + β)/2 Backus, Chernov, & Zin (NYU, LSE) Monetary policy 21 / 24 Example 3: bond pricing models Representative agent model: identification What do we need to estimate λ = τ −1 ? Reminder: solution is β ⊤ = (αe1 − e2 )⊤ (τ I − A)−1 r = δ + αg − (αe1 + β)⊤ CC ⊤ (αe1 + β)/2 Identification strategies I I I I I Observe log gt , e1 Long bond yields span state Risk aversion α: match equity premium One restriction on e2 Additional information from mean bond yields Backus, Chernov, & Zin (NYU, LSE) Monetary policy 22 / 24 Summary and conclusion Identification never a free lunch We suggest I I I Forecasts and bond yields aid observability of state Additional restrictions can identify shock and “Taylor rule” Representative agent bond pricing models can also help The biggest challenge is not identification, but fit Backus, Chernov, & Zin (NYU, LSE) Monetary policy 23 / 24 Related work (some of it) Identifying monetary policy shocks I Christiano-Eichenbaum-Evans, Cochrane, Hansen-Sargent, Leeper-Sims-Zha Factor models I Sargent-Sims, Stock-Watson Macro bond pricing models I Gallmeyer-Hollifield-Zin, Piazzesi-Schneider, Rudebusch-Swanson, Smith-Taylor ... Backus, Chernov, & Zin (NYU, LSE) Monetary policy 24 / 24