## 10. Filtered and observed demand factors

## Source details

**Canonical URL:** [10. Filtered and observed demand factors](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1466.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1466.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1466.pdf.json)

---

### I. Question and empirical approach
- Research question: How big is the impact of official demand (foreign officials and the Fed) on US yields?
- Model and data:
  - Structural arbitrage-free preferred-habitat model of Vayanos and Vila (2009).
  - Estimated on the term structure of monthly rates derived from US Treasury Inflation-Protected Securities (TIPS).
  - Sample period: 2001-2012.
  - Term structure determined by two factors: the short-term real interest rate and the (excess) demand factor.
  - Demand factor specified as a function of observable variables capturing:
    - the share of outstanding Treasury securities held by the Fed and foreign officials (official demand pressure), and
    - the relative maturity of the Fed’s Treasury portfolio.
- Caveat: analysis abstracts from “local” maturity-specific demand effects because of lack of maturity-specific data on official holdings.

### II. Model summary and estimation setup
- Continuous-time setup with zero-coupon bonds for maturities τ ∈ (0; T].
- Short-rate dynamics (Ornstein-Uhlenbeck):
  - dr_t = κ_r (r̄ - r_t) dt + σ_r dB_{r,t}
  - (r̄ ; κ_r ; σ_r) positive constants.
- Aggregate demand factor dynamics (Ornstein-Uhlenbeck):
  - dβ_τ = κ_β (β̄ - β_τ) dt + σ_β dB_{β,t}
- Demand intercept aggregation:
  - β_τ = θ_1 β_{τ,FOR} + θ_2 β_{τ,FED} (used conceptually; model estimates aggregate β_τ).
- Pricing implications: equilibrium spot rates affine in two risk factors (r_t, β_t); bond prices depend on functions A(τ), B(τ), C(τ) solved from linear ODEs.
- Data and maturities:
  - Zero-coupon equivalent US real rates from Fed’s TIPS-yields estimates.
  - Sample: Jan-2001 to Nov-2012 (two-year TIPS available only from Jan-2004).
  - Maturities: 2-, 3-, 4-, 5-, 7-, 10-, 15- and 20-yr.
- Observable demand measures D_t = [FO_t, Fed_t, MatFed_t, MatPub_t]′:
  - FO_t: fraction of foreign officials’ holdings of long-term Treasuries over total Treasury securities outstanding (adjusted).
  - Fed_t: fraction of Fed’s holdings of Treasury securities over total outstanding.
  - MatPub_t: log average maturity (in months) of Treasury securities held by the public.
  - MatFed_t: log average maturity (in months) of the Fed’s Treasury portfolio.
- Econometric approach:
  - Bayesian MCMC (Metropolis-Hastings within Gibbs).
  - State vector X_t = (r_t, β_t)′ discretized with monthly ∆ = 1/12.
  - Measurement errors assumed independent normal with common variance ߪ ల ଶ (notation preserved).

### III. Fit, factor dynamics, and term-premia decomposition
- Model fit:
  - Standard deviation of yield fitting errors: ߪ ల ଶ = 8.41.
  - Measurement errors rarely exceed 30 basis points in absolute value across fitted maturities; model underfits the 20-year maturity.
  - Measurement errors large during 2008–2009 crisis.
- Factor dynamics:
  - Estimated factor correlation: ρ = 0.24.
  - Mean half-lives:
    - First factor (short-rate): around 3 years.
    - Second factor (demand): around 5 years.
  - Product of arbitrageur risk aversion and demand elasticity: aα = 42.56.
- Factor loadings:
  - Short-rate factor loadings decline quickly with maturity and are negligible beyond 15 years.
  - Demand factor loadings: ≈ zero at very short maturities; peak around 10 years; ≈ unity for longer maturities (>10 years).
- Term-premia behavior:
  - Average term premia above 100 basis points up to 2004.
  - Fell below zero in 2005; around zero until crisis; then fell dramatically into negative territory after the crisis.
  - Correlation between the ten-year term premium and the demand factor: around 70 percent.

### IV. Transmission channels of LSAP/MEP and institutional context
- Primary channels:
  - Duration channel: decreases the bond risk premium component and has greater effects on longer-maturity bonds.
  - Scarcity channel: changes the scarcity price of ultra-safe, highly liquid Treasury assets.
  - Signaling channel: lowers expected policy rates and thus the expectations component of long-term yields.
- Additional channel:
  - Inflation channel: large LSAPs could increase inflation risk and inflation expectations, potentially pushing long-term interest rates up; effectiveness depends on monetary policy credibility and anchored inflation expectations.
- Interaction with Treasury debt management:
  - Treasury increased relative supply of longer-term securities during LSAPs, increasing average maturity of Treasury debt outstanding and partly softening demand pressures.
- Institutional magnitudes and facts:
  - September 2009: non-Americans held >65 percent of all US government notes and bonds outstanding; foreign official investors held roughly 37 percent of total Treasury supply.
  - 2004: foreign official net purchases of US Treasury notes and bonds estimated at 201 billion USD, amounting to 70percent of total net issuance of US Treasury notes and bonds in that period.
  - By 2012: Fed held about 18percent of total Treasury securities outstanding and about 35percent of the segment with maturity over ten years.
- Fed unconventional programs (select chronology and amounts):
  - TSLF started March 2008 (Treasury securities lent from SOMA).
  - LSAP1 (March 2009–March 2010): FOMC committed to purchase $300 billion of longer-term Treasury securities and $850 billion of agency securities in addition to the $600 billion in MBS and agency debt announced in November 2008.
  - LSAP2 (announced November 2010): $600 billion in additional purchases of longer-term Treasury securities to be completed by mid-2011.
  - MEP (started September 2011): Fed sold $667 billion of shorter-term Treasury securities and used proceeds to purchase longer-term Treasury securities.
  - FOMC announced a $400 billion program in September 2011 to be completed by end-June 2012; in June 2012 continued program through end-2012, resulting in purchase, sale and redemption of an additional $267 billion in Treasury securities.

### V. Quantitative impacts of official demand (filtered vs observed)
- Foreign officials:
  - By August 2008 (2001–08 period), foreign purchases exerted a cumulative negative impact of around 81 basis points on long-term US Treasury yields.
  - During LSAP1 and LSAP2 periods foreign-official effect negligible (offset by coincident increases in Treasury supply).
  - 2004–05 drop due to foreign demand more contained, consistent with "less than 50 basis points" assessments.
- Federal Reserve (size and maturity decomposition):
  - LSAP1 (Mar-09 to Nov-09) on 10-year real yield:
    - Panel A (size alone): -63 basis points.
    - Panel C (size + maturity): -41 basis points (maturity effects offset some size effect).
    - Comparable studies cited: D'Amico and King (2013) -45 bps; Bomfim and Meyer (2010) -60 bps; Gagnon et al. (2011) -58 to -91 bps.
  - LSAP2 estimated impact: authors estimate yields would have been at least 111 basis points higher in absence of LSAP2 (paper focuses on real rates; other studies focus on nominal rates).
  - Maturity Extension Program (MEP):
    - Reduction in yields during MEP attributed mainly to changes in Fed portfolio maturity relative to public’s portfolio rather than portfolio size.
    - Estimated impact of MEP on 10-year real yield: around -47 basis points.
    - Comparisons: Swansson (2011) -15 basis points; Meaning and Zhu (2012) -80 basis points.
- Aggregate Fed counterfactual:
  - Without Fed purchases, 10-year real yields would have been up to 140 basis points higher (authors’ concluding statement).

### VI. Counterfactual experiments and timing of impacts
- Counterfactuals fix foreign official demand pressure or Fed demand pressure at Jan-2001 levels and compute model-implied rates.
- Results:
  - Fed asset purchase programs materially contributed to lower five- and ten-year real yields from 2010 onwards; without Fed purchases real yields would have been significantly higher since 2010.
  - Foreign officials’ cumulative impact concentrated during the conundrum period (mid-2000s) and negligible during the crisis due to offsetting supply increases.

### VII. Residual demand drivers and robustness
- Residual between filtered β_t and fitted observed demand remains; plausible contributors:
  - Financial intermediaries’ hedging and portfolio rebalancing (e.g., MBS hedging).
  - Market sentiment and macroeconomic uncertainty (US variance risk premium, VRP).
  - Institutional investors (pension funds, life insurers) and mutual funds: around 11 percent of US government debt holdings as of 2012.
  - Commercial bank hoarding of Treasuries during crisis periods (2008–2010).
- Reduced-form regressions (equation (19), Table 6):
  - VRP, MBS duration, and commercial bank holdings are qualitatively consistent with additional demand pressures.
  - In univariate regressions, commercial banks' activity delivers the highest R-square.
  - All three variables remain significant when included jointly.

### VIII. Main numerical findings and summary magnitudes
- Two principal components explain >99.76percent of real yields (PC analysis).
- Estimated correlation between factor shocks: ρ = 0.24.
- Product aα = 42.56.
- Mean half-lives: ~3 years (short-rate), ~5 years (demand).
- Standard deviation of yield fitting errors: ߪ ల ଶ = 8.41.
- Term premia:
  - Averaged above 100 basis points up to 2004; fell below zero in 2005; large negative term premia post-crisis.
- Cumulative impacts:
  - Foreign official purchases: around -81 basis points on long-term yields over 2001–08.
  - Fed asset purchases: absence would leave 10-year real yields up to 140 basis points higher.
  - LSAP1 10-year real yield impact: Panel A -63 basis points; Panel C -41 basis points (size + maturity).
  - LSAP2 absence implied at least +111 basis points on 10-year real yields (authors’ estimate).
  - MEP impact on 10-year real yield: around -47 basis points.

### IX. Policy-relevant implications, limitations, and extensions
- Policy implications:
  - Official demand pressures (foreign officials and the Fed) materially affected the term structure of US real interest rates, mainly through term-premium channels.
  - Fed asset purchase programs were effective in lowering medium- and long-term real yields.
  - The “conundrum” (2004–05) can be rationalized by rising demand lowering term premia while expected short rates increased.
- Model limitations and suggested extensions:
  - Two-factor representation is parsimonious; richer models could introduce additional large players (pension funds, regulated commercial banks) or model multiple individual demand factors.
  - To assess impacts on nominal yields, a joint nominal-real term-structure model is required because nominal effects include changes in inflation expectations and inflation risk premia.
  - Framework suitable to analyze subsequent events (e.g., Fed tapering in May 2013) to study interaction of official demand changes and forward guidance.

*Source: _wp1466 - "10. Filtered and observed demand factors" (IMF Working Paper).*

### References .............................................................................................................

### _wp1466 - References

### Tables
- 1. Yield summary statistics and PC analysis .........................................................................16
- 2. Estimated parameters of the model ....................................................................................21
- 3. Demand factor parameters .................................................................................................27
- 4. Impact of foreign officials’ demand for treasuries on the term structure of real rates ......29
- 5. Impact of Federal Reserve asset purchase policies on the term structure of real rates ......30
- 6. Explaining differences between official observed and filtered demand factors ................35

### Figures
- 1. Estimated ownership of US Treasury securities ..................................................................6
- 2. Fed Treasury holdings across different maturity segments .................................................8
- 3. US Treasury debt outstanding ............................................................................................10
- 4. Measures of demand pressures in the US Treasury market ...............................................17
- 5. Observed yields and measurement errors ..........................................................................22
- 6. Estimated factors ................................................................................................................23
- 7. Factor loading ....................................................................................................................24
- 8. Real interest rates decomposition ......................................................................................26
- 9. The cumulative effects of demand pressures on real rates ................................................32

*Source: _wp1466 - References.*

### 10. Filtered and observed demand factors ...............................................................................

### 10. Filtered and observed demand factors

### I. Introduction: question and approach
- Research question: How big is the impact of official demand (foreign officials and the Fed) on US yields?
- Model and data:
  - Structural arbitrage-free preferred-habitat model of Vayanos and Vila (2009).
  - Estimated on the term structure of monthly rates derived from US Treasury Inflation-Protected Securities (TIPS).
  - Sample period: 2001-2012.
  - The term structure is determined by two factors: the short-term real interest rate and the (excess) demand factor.
  - The demand factor is specified as a function of observable variables capturing:
    - the share of outstanding Treasury securities held by the Fed and foreign officials (official demand pressure), and
    - the relative maturity of the Fed’s Treasury portfolio.
- Key modeling caveat: the analysis abstracts from “local” maturity-specific demand effects because of the lack of maturity-specific data on official holdings.

### II. Main findings on official demand impacts
- Cumulative impacts estimated:
  - By 2008, foreign purchases of US Treasuries exerted a cumulative negative impact of around 80 basis points on long-term US Treasury yields.
  - Quantified effect of Fed purchases (QE): QE depressed real 10-year yields by around 140 basis points (total subsequent impact).
  - During the Maturity Extension Program (MEP) estimated depressions:
    - 5-year yields depressed by around 48 basis points.
    - 10-year yields depressed by around 61 basis points.
- Other empirical findings:
  - Both the amount of Treasuries held by the Fed and the maturity pressures (duration) exerted by Fed purchases matter for yields.
  - Official demand pressure is defined as official policy demand relative to the Treasury supply; hence the market impact of official purchases varies with Treasury supply.

### III. Institutional and market context (descriptive statistics and historical developments)
- Foreign official and private holdings (selected facts from the text):
  - In September 2009 non-Americans held more than 65 percent of all US government notes and bonds outstanding, with foreign official investors holding roughly 37 percent of the total Treasury supply.
  - In 2004 foreign official institutions are estimated to have purchased US Treasury notes and bonds quantifiable in a net face value of 201 billion USD, amounting to 70percent of the total net issuance of US Treasury notes and bonds during that period.
- Fed holdings and maturity concentration:
  - By 2012 the Fed held about 18percent of the total Treasury securities outstanding.
  - The Fed held about 35percent of the segment of the Treasury market with maturity over ten years.
- Fed unconventional programs and magnitudes:
  - Term Securities Lending Facility (TSLF) started March 2008 (Treasury securities lent from SOMA).
  - LSAP1 (March 2009–March 2010): FOMC committed to purchase $300 billion of longer-term Treasury securities and $850 billion of agency securities in addition to the $600 billion in MBS and agency debt announced in November 2008.
  - LSAP2 (announced November 2010): $600 billion in additional purchases of longer-term Treasury securities to be completed by mid-2011.
  - MEP (started September 2011): aimed to increase average maturity of Fed’s Treasury portfolio without expanding balance sheet; under MEP the Federal Reserve sold a total of $667 billion of shorter-term Treasury securities and used the proceeds to purchase longer-term Treasury securities.
  - FOMC announced a $400 billion program in September 2011 to be completed by end-June 2012; in June 2012 continued the program through end-2012, resulting in the purchase, sale and redemption of an additional $267 billion in Treasury securities.

### IV. Transmission channels of unconventional monetary policy (LSAP/MEP)
- Primary channels identified:
  - Duration channel: decreases the bond risk premium component of Treasury yields and has greater effects on longer-maturity bonds.
  - Scarcity channel: affects the term premium by changing the scarcity price of ultra-safe, highly liquid Treasury assets.
  - Signaling channel: lowers expected policy rates and thus the expectations component of long-term yields.
- Additional (by-product) channel:
  - Inflation channel: large LSAPs could increase inflation risk and inflation expectations, which could push long-term interest rates up; effectiveness of LSAPs depends on monetary policy credibility and anchored inflation expectations.
- Interaction with Treasury debt management:
  - Treasury increased relative supply of longer-term securities in the lower interest rate environment created by LSAPs, increasing the average maturity of Treasury debt outstanding and partly softening demand pressures for longer-term bonds created by Fed programs.

### V. Literature positioning and contribution
- Existing evidence:
  - Studies find qualitatively downward effects of official interventions; quantitative estimates vary.
  - Range of estimated nominal ten-year Treasury yield impacts of LSAP-style operations: from 35 to more than 160 basis points (as summarized in the literature review).
  - Many studies find Fed impact on nominal yields is mostly felt through the real term premium component.
- Contribution of this paper:
  - Jointly analyzes foreign and domestic (Fed) official demand pressures within a single structural preferred-habitat framework.
  - First structural estimation of the preferred-habitat model including observable measures of demand pressures (to the authors’ knowledge).
  - Allows separating quantity pressures from duration pressures to better understand transmission channels of Fed policies.
  - Provides consistent estimation of official demand impacts across maturities and over time.

### VI. Model summary
- Framework: limited-arbitrage preferred-habitat model (Vayanos and Vila, 2009).
- Agent types:
  - Preferred-habitat investors: strong preferences for specific maturities (e.g., official investors).
  - Risk-averse arbitrageurs: trade bonds across maturities for return considerations and enforce no-arbitrage.
- Role of arbitrageurs: smooth local demand and supply pressures; because arbitrageurs are risk-averse, preferred-habitat demand still influences equilibrium yields.
- Empirical implementation: demand factor is linked to observable measures of Fed and foreign official shares and Fed portfolio average maturity; model estimated on TIPS term structure, 2001–2012.

*Italic: Content summarized from chapter "10. Filtered and observed demand factors" of the supplied IMF PDF.*

### Section introduces the main elements of the model, while more details are given in Appendix I.

### Section introduces the main elements of the model, while more details are given in Appendix I.

### Model setup and key state variables
- The model is set in continuous time with a continuum of zero-coupon bonds having maturities τ in the interval (0; T].
- The risk free instantaneous rate r_t (the limit of the spot rate of maturity τ, ܴ_{τ,t}, as τ → 0) follows the Ornstein-Uhlenbeck process:
  - dr_t = κ_r (r̄ - r_t) dt + σ_r dB_{r,t}    (1)
  - (r̄ ; κ_r ; σ_r) are positive constants and B_{r,t} is a Brownian motion.
- Preferred-habitat investors form maturity clienteles; the clientele for maturity τ only buys the bond with the same maturity.
- The demand for the bond with maturity τ is a linear function of the bond’s yield ܴ_{τ,t}:
  - y_{τ,t} = α(τ) (τ ܴ_{τ,t} - β_τ)    (2)
  - α(τ) is a positive function of maturity. Increases in β_τ are associated with a decreasing excess demand y_{τ,t}.

### Specification of the demand intercept and aggregation
- The intercept β_τ is assumed to take the form:
  - β_τ = ∑_{i=1}^{N} β_{τ,i}    (3)
  - where β_{τ,i} are various demand factors capturing changes in preferred-habitat investor needs, investor pool size/composition, or available supply of government bonds.
- Hypothesis: these demand factors are influenced by strong accumulation of reserves by foreign officials and large scale bond purchases by the Fed. Official demand for long-term US Treasury securities is high and not fully elastic (as documented in Section 2 of the source).
- The model assumes β_τ composes of a foreign-specific demand factor β_{τ,FOR} and a Fed-specific demand factor β_{τ,FED}, with weights θ_1 and θ_2:
  - β_τ = θ_1 β_{τ,FOR} + θ_2 β_{τ,FED}    (4)
- Instead of modeling the components separately, the aggregate demand factor β_τ follows an Ornstein-Uhlenbeck process:
  - dβ_τ = κ_β (β̄ - β_τ) dt + σ_β dB_{β,t}    (5)
  - β̄ is the unconditional mean; (κ_β, σ_β) are positive constants and B_{β,t} is a Brownian motion.

### Modeling choices, tractability, and implications for bond pricing
- The law of motion is specified for the aggregate demand factor (β_τ) and not its components to limit the number of parameters entering equilibrium bond-yield pricing and preserve model tractability.
- As a result, bond pricing depends on parameters describing the dynamics of two factors:
  - the short rate (equation (1)), and
  - the aggregate demand factor (equation (5)).
- The parameters θ_1 and θ_2 can affect the estimate of β_τ but do not enter directly into the bond pricing recursions (presented in Appendix I).
- Introducing a third state equation to model β_{τ,FOR} and β_{τ,FED} separately would make bond prices depend on a system of nine non-linear difference equations (increasing model complexity).

### Additional modeling notes (from source footnotes)
- The original version of Vayanos and Vila (2009) used local demand effects β_m = ∑ θ_j β_{m,j} with m denoting maturity; the present model abstracts from local demand effects due to lack of maturity-specific data on official holdings (foreign in particular).
- Analysis of local supply effects of QE on bond-by-bond data is noted as provided elsewhere (D’Amico and King (2013) referenced in source).

*Source: _wp1466 - Section introduces the main elements of the model, while more details are given in Appendix I.*

### introduction of a third factor, would come at the cost of solving five additional equations that would seriously comprom

### _wp1466 - introduction of a third factor, would come at the cost of solving five additional equations that would seriously compromise the possibility to take the model to the data.

### Model structure and asset-pricing implications
- Arbitrageurs trade across maturities and demand risk premia for taking maturity-differentiated bond risk; their portfolio choice follows mean-variance optimization (objective in equation (6)) with risk-aversion coefficient ܽ, dollar investments ݔ
௧,ఛ
, and time-t wealth ܹ
௧
.
- Arbitrageurs’ budget constraint is given by equation (7) with bond price ܲ
௧,ఛ
 paying $1 at time t+τ.
- Equilibrium spot rates are assumed affine in two risk factors ݎ
௧
 and β
௧
 (equation (8)); the A(τ), B(τ), C(τ) functions are solved via a system of linear ODEs imposing equilibrium x
୲,த
ൌ y
୲,த
.
- Expected excess return of inflation-indexed Treasury bond over the risk-free rate at maturity ߬ is linear in factor sensitivities (equation (9)), with factor risk premia defined in equations (10) and (11). Factor risk premia depend on factor demand and factor correlation ρ.

Key technical implications:
- Returns in excess of the risk-free rate are a linear function of the bond’s sensitivities to the risk factors (general consequence of no-arbitrage).
- Factor risk premia directly relate to demand for bonds at each maturity, allowing the model to capture scarcity and duration channels through prices of risk.

### Estimation — data
- Data: zero-coupon equivalent US real rates from the Fed’s TIPS-yields estimates.
- Sample period: Jan-2001 to Nov-2012 (note: two-year TIPS available only from Jan-2004).
- Maturities used: 2-, 3-, 4-, 5-, 7-, 10-, 15- and 20-yr maturities.
- Empirical observations:
  - Average real yield curve is upward sloping; long-term real rates fell substantially 2001–2005 and later experienced dramatic swings during the 2007–2009 financial crisis.
  - Around end-2008, real rates spiked and at some point exceeded nominal rates (coinciding with deflation and repo-market liquidity tensions).
  - After 2011 and up to end-2012, medium- to long-term real interest rates turned negative.
  - Short-term yields are more volatile than long-term yields; volatility decreases with maturity.
- Principal component analysis (bottom panel of Table 1): two PCs explain more than 99.76percent of the real yields data. First PC loads strongly on short maturities; second PC behaves like slope factor.

Observable demand measures used to proxy preferred-habitat demand D_t = [FO_t, Fed_t, MatFed_t, MatPub_t]′:
- FO_t: fraction of foreign officials’ holdings of long-term Treasuries over total Treasury securities outstanding (adjusted series from Bertaut and Tryon (2007)).
- Fed_t: fraction of Fed’s holdings of Treasury securities over total outstanding.
- MatPub_t: log average maturity (in months) of Treasury securities held by the public.
- MatFed_t: log average maturity (in months) of the Fed’s Treasury portfolio.
- Notes on measurement caveats:
  - TIC data limitations and need for adjustments (indirect transactions). Bertaut and Tryon (2007) adjustments used.
  - TIC includes both Treasuries and TIPS; disaggregated foreign ownership of TIPS not available, but foreign demand for TIPS has averaged around 39percent.

### Econometric methodology
- Model: yields are affine functions of normally distributed factors; estimation via Bayesian MCMC (Metropolis-Hastings within Gibbs).
- State-space setup:
  - State vector X_t = (r_t, β_t)′ evolves under objective measure (discretized via Euler scheme with monthly ∆ = 1/12).
  - Measurement equation maps factors into observed real yields for selected maturities; measurement errors assumed independent normal with common variance ߪ
ఌ
ଶ
.
  - Additional measurement equation links the dynamics of β_t to observed demand factors via coefficients ߛ (empirical counterpart to equation (4) — equation (14)).
- Identification and priors:
  - α(τ) ≡ α exp{-δτ} with δ fixed to 0.1.
  - Unconditional mean of the short rate r̄ is fixed to 2percent.
  - Product of demand elasticity and arbitrageur risk aversion (αܽ) is estimated as a single parameter because α and ܽ are not separately identified.
  - Priors largely uninformative; constraints imposed (stationarity of factors; non-negative arbitrageur risk aversion).

Rationale for Bayesian approach:
- Likelihood highly nonlinear because bond prices are complex nonlinear functions of parameters.
- MCMC draws allow quantification of posterior uncertainty for derived objects (loadings, term premia, regressions).
- Priors and parameter constraints are easily implemented.

### Estimation results — model fit, factors, and premia
Model performance:
- All state parameters statistically significant; MCMC chain convergence supported by numerical standard errors and CD diagnostics.
- Standard deviation of yield fitting errors: ߪ
ఌ
ଶ
= 8.41.
- Measurement errors rarely exceed 30 basis points in absolute value across fitted maturities, but the model underfits the 20-year maturity (possibly due to institutional investor local demand).
- Measurement errors large during 2008–2009 crisis due to deflation and illiquidity in inflation-protected securities.

Factor dynamics and interpretation:
- Estimated factor correlation: ρ = 0.24 (positive correlation between short-rate factor and demand pressures).
  - Interpretation: higher excess supply (lower demand pressures) associated with higher policy rates.
- Mean half-lives:
  - First factor (short-rate): around 3 years.
  - Second factor (demand): around 5 years.
- Arbitrageurs’ price-of-risk parameters are significant; product aα = 42.56, indicating time-varying real bond term premia.

Factor loadings (Figure 7 implications):
- Short-rate factor loadings decline quickly with maturity and are negligible beyond 15 years.
- Demand factor loadings:
  - Approximately zero at very short maturities.
  - Peak around 10 years.
  - Approximately unity for longer maturities (>10 years).
- Implication: short-term rates mainly driven by short-rate factor; long-term rates mostly driven by demand factor (preferred-habitat). Short-lived liquidity shocks (e.g., fall 2008 spike) are largely captured by increases in the short-rate factor.

Term-premia decomposition (Figure 8):
- Estimated term premia:
  - Averaged above 100 basis points up to 2004.
  - Fell below zero in 2005, around zero until the crisis, then fell dramatically into negative territory after the crisis.
- The conundrum (2004–2005): long-term real rates fell despite rising expected short rates because rising demand drove down term premia; thus the model suggests the conundrum reflects demand-driven term-premia changes rather than puzzling expectations of policy rates.
- Post-2009 decreases in long-term rates attributed mainly to lower expected policy rates; from 2011 onwards decreases mainly due to decreasing term premia while policy rate expectations stabilized.
- Correlation between the ten-year term premium and the demand factor: around 70 percent.

### Demand factor analysis — linking filtered factor to observables
- Main hypothesis: both size and maturity structure of official holdings explain the filtered excess supply factor (β_t). Regression specification: β_t = γ_0 + γ_1 FOdp_t + γ_2 FEDdp_t + γ_3 PubAvgMat_t + γ_4 FEDAvgMat_t + ε_t (Table 3).
- Estimated signs and economic effects:
  - Increases in foreign official and Fed shares of Treasury holdings decrease excess supply (i.e., increase preferred-habitat demand).
  - Increase in average maturity of Fed holdings decreases excess supply (ߛො₄ = -0.028).
  - Increase in average maturity of public holdings increases excess supply (ߛො₃ = 0.044).
- The fitted demand factor is constructed as ߚ
௧
௢௕௦
 = ܦො′
௧ା∆
 using estimated ߛො parameters (equation (15)); re-estimating the model with observed demand produces comparable pricing errors.

Impact quantification — foreign officials (Table 4):
- Model-implied cumulative impact of foreign official purchases on long-term real yields:
  - By August 2008 (2001–08 period), foreign purchases had a cumulative negative impact of around 81 basis points on long-term US Treasury yields.
  - During the LSAP1 and LSAP2 periods the effect of foreign official demand was negligible (explained by coincident increases in Treasury supply).
- During 2004–05 the estimated drop due to foreign demand was more contained (consistent with Greenspan’s “less than 50 basis points” view), and estimates are broadly consistent with Warnock and Warnock (2009).

Impact quantification — Federal Reserve demand and maturity effects (Table 5):
- Method: decompose Fed demand pressure into size (Panel A) and maturity profile (Panel B); Panel C reports total Fed impact (Panel A + Panel B).
- LSAP1 (Mar-09 to Nov-09) estimated impact on 10-year real yield:
  - Panel A (size alone): -63 basis points.
  - Panel C (size + maturity): impact decreases to -41 basis points when maturity effects included.
  - Comparison: results comparable to D'Amico and King (2013) (-45 bps), Bomfim and Meyer (2010) (-60 bps), Gagnon et al. (2011) (-58 to -91 bps).
- LSAP2 impacts on 10-year yields estimated to be larger (authors estimate yields would have been at least 111 basis points higher in absence of LSAP2); other studies report smaller magnitudes (e.g., -33 to -55 bps).
  - Note: this paper focuses on real rates; other studies focus on nominal rates (and thus include impacts via inflation expectations and inflation risk premia).
- Maturity Extension Program (MEP):
  - The reduction in yields during MEP attributed mainly to changes in Fed portfolio maturity relative to public’s portfolio rather than changes in portfolio size.
  - Estimated impact of MEP on 10-year real yield: around -47 basis points.
  - Comparison: Swansson (2011) suggested -15 basis points; Meaning and Zhu (2012) estimated -80 basis points.

Counterfactuals and cumulative effects (Figure 9):
- Counterfactual experiments fix foreign official demand pressure or Fed demand pressure at Jan-2001 levels and compute model-implied rates.
- Results:
  - Fed asset purchase programs materially contributed to lower five- and ten-year real yields from 2010 onwards; without Fed purchases real yields would have been significantly higher since 2010.
  - Foreign officials’ cumulative impact concentrated during the conundrum period (mid-2000s) and was negligible during the crisis due to offsetting supply increases.

Other potential determinants of preferred-habitat demand (explaining residual between filtered and fitted demand)
- Residual between filtered β_t and fitted observed demand ߚ
ݐ
ݏܾ݋
 remains; possible contributors:
  - Financial intermediaries’ hedging and portfolio rebalancing (e.g., MBS hedging causing Treasury purchases when outstanding MBS duration drops).
  - Market sentiment and macroeconomic uncertainty (captured by US variance risk premium).
  - Institutional investors (pension funds, life insurers) with inelastic demand and liability-driven motivations; combined with mutual funds, pension funds and insurers represent around 11 percent of US government debt holdings as of 2012 (third largest group).
  - Commercial bank hoarding of Treasuries during crisis periods (notable 2008–2010 activity).
- Reduced-form regressions (equation (19), Table 6) relating unexplained filtered demand to: variance risk premium (VRP), MBS duration, commercial bank holdings:
  - Findings: growing market fear (higher VRP), lower MBS duration, and higher commercial bank Treasury holdings are all qualitatively consistent with additional demand pressures.
  - In univariate regressions, commercial banks' activity delivers the highest R-square.
  - All three variables remain significant when included jointly.

### Main quantitative findings and magnitudes
- Two principal factors explain >99.76percent of real yields (PC analysis).
- Estimated correlation between factor shocks: ρ = 0.24.
- Product of arbitrageur risk aversion and demand elasticity: aα = 42.56.
- Mean half-lives: ~3 years (first factor), ~5 years (second factor).
- Standard deviation of yield fitting errors: ߪ
ఌ
ଶ
 = 8.41.
- Term premia behavior:
  - Average term premia above 100 basis points up to 2004; fell below zero in 2005; large negative term premia post-crisis.
- Cumulative impacts:
  - Foreign official purchases: around -81 basis points on long-term yields over 2001–08.
  - Fed asset purchases (no-Fed counterfactual): 10-year real yields would have been up to 140 basis points higher in absence of Fed purchases (concluding remark).
  - LSAP1 impact on 10-year real yield: Panel A estimate -63 basis points (Panel C total -41 basis points when maturity included).
  - LSAP2 estimated to have raised 10-year yields by at least 111 basis points in absence of LSAP2 (authors’ estimate).
  - MEP impact on 10-year real yield: around -47 basis points.

### Conclusions and policy-relevant implications
- Structural two-factor no-arbitrage preferred-habitat model estimated on US real rates and official holdings:
  - Finds that official demand pressures (foreign officials and Fed) materially affected the term structure of US real interest rates, mainly through term-premium channels.
  - Fed asset purchase programs were effective; absence of Fed purchases would have left 10-year real yields up to 140 basis points higher.
  - Foreign official investors contributed about 80 basis points downward pressure on long-term rates (2001–08).
- Interpretation of historical episodes:
  - The “conundrum” (2004–05) can be rationalized by rising demand lowering term premia while expected short rates increased—no puzzle once demand effects are accounted for.
- Model limitations and research extensions:
  - Two-factor representation is parsimonious; richer models could introduce additional large players (pension funds, regulated commercial banks) or model multiple individual demand factors.
  - To assess impacts on nominal yields, a joint nominal-real term-structure model is required, as nominal effects include changes in inflation expectations and inflation risk premia.
  - The framework can be applied to subsequent policy events (e.g., Fed tapering in May 2013) to assess interplay of official demand changes and forward guidance.

*Italicized source: IMF Working Paper (content unit: _wp1466 - introduction of a third factor, would come at the cost of solving five additional equations that would seriously compromise the possibility to take the model to the data).*

### REFERENCES

### _wp1466 - REFERENCES

### APPENDIX I. MODEL DETAILS
- Model conjecture:
  - Equilibrium spot rates are affine in the risk factors: the short rate (r_t) and the demand factor (β_t).
  - Equilibrium bond price takes the exponential form (A1) with three functions A(τ), B(τ), C(τ) that depend on maturity τ.
- Key derivations:
  - Applying Ito's Lemma to (A1) and using dynamics (1) of r_t and (5) of β_t yields the instantaneous return (A2) and the instantaneous expected return μ^{a,τ} defined in (A3).
  - Substituting (A2) into the arbitrageurs' budget constraint (7) allows solving the arbitrageurs' optimization problem and deriving bond risk premia (excess returns) as in eq. (9); point-wise maximization of (A4) yields (9).
  - Market clearing (x^{a,τ} = y^{a,τ}) together with eq. (2), (A1) and definition of 𝒴_{t,τ} produces (A5).
  - Substituting expressions from (A3), (10), (11), and (A5) into (9) yields affine equations in r_t and β_t; setting linear terms to zero leads to the matrix system (A6a) and (A6b) involving matrix M.
  - The solution to (A6a)-(A6b) is given by (A7) and (A8). Identification of terms yields four scalar non-linear equations (A13)-(A16) for unknowns (ν₁, ν₂, ζ_a, ζ_b).
- Functional-form assumptions and parameter normalization:
  - Adopt α(τ) ≡ α exp{-δτ} and θ(τ) = 1 (i.e., demand factor affects all maturities equally in absence of arbitrageurs).
  - Set α = 1 (without loss of generality because α matters only through the product α a).
- Determination of C(τ):
  - Equations (A17)-(A19) relate the pricing terms ρ_{a,t}, ρ_{b,t} and C(τ); the solution for C(τ) is given explicitly (equation after (A19)).
  - Definitions for auxiliary quantities ϕ_a and ϕ_b are given in (A20) and (A21).
  - Substituting C(τ) into (A20)-(A21) yields ϕ(τ) = (ϕ_a, ϕ_b) as the solution to a linear system.

### APPENDIX II. MCMC ALGORITHM
- Estimation approach:
  - Term structure model estimated using Bayesian methods and a block-wise Metropolis-Hastings (MH) within Gibbs sampler.
  - Joint posterior decomposed into conditional densities for parameters, state variables, pricing error variance, demand coefficients, and measurement error variance.
  - Metropolis step used for parameters whose conditionals are not in closed form; Gibbs steps used when conditionals are known in closed form.
- Sampling of factors and parameters:
  - Factor sampling uses the Carter and Kohn (1994) forward-filtering backward-sampling (simulation smoother) because the Vasicek model is linear and Gaussian.
  - Drift parameters (which also enter yield pricing) are drawn using a Random-Walk Metropolis (RWM) algorithm with proposal θ^(g+1) = θ^(g) + v_g ε, where ε ~ N(0,1) and v_g tuned to target acceptance probabilities between 10-50percent.
  - RWM performed for each individual drift parameter (k_r; k_β; r̄; β̄).
- Variance-covariance matrix Q:
  - Posterior of Q uses an inverse Wishart prior; draws from inverse Wishart proposal and accept/reject probability simplifies as described.
  - Perform accept/reject step for each individual candidate draw (e.g., ρ, σ_r, σ_β) to maintain reasonable acceptance rates.
- Identification and special parameters:
  - Arbitrageurs' risk aversion a and excess demand elasticity α are not separately identified; estimate their product aα.
  - Draws for aα use RWM; acceptance probability simplifies because aα does not enter transition equations.
- Measurement-error variances and demand loadings:
  - Gibbs sampler used to draw variance of bond pricing errors (inverse Gamma prior) pooling n vectors of residuals into a single series.
  - Demand factor loadings γ drawn from π(γ | X^T, D^T), a Gaussian with mean (X'^T X'^T)^{-1} (X'^T D'^T); conditional draws for demand measurement-error variance follow.
- Priors and constraints:
  - Priors are proper but weakly informative: inverse Wishart for transition covariance, inverse Gamma for measurement variances, normal or truncated normal for other parameters.
  - Impose positivity constraints on arbitrageurs' risk aversion and mean reversion parameters to ensure factors' stationarity; draws outside constraints are discarded, with a maximum number of proposal attempts after which the old draw is retained.
- Implementation details and tuning:
  - Perform 80,000 replications, of which the first 40,000 are burned to ensure convergence to the ergodic distribution.
  - Save 1 every 20 draws of the last 40,000 replications to limit autocorrelation.
  - Target RWM acceptance level around 20-40percent (Johannes and Polson (2004)).
  - Variance tuning procedure (first half of burn-in): check acceptance ratio every 100 draws; if accepted more than 50 draws over the last 100, double the standard deviation; if accepted less than 10 percent of draws, half the standard deviation.
- Convergence diagnostics:
  - Use numerical standard error (NSE) and Geweke convergence diagnostic (CD) (Geweke (1992)).
  - NSE accounts for correlation in draws; CD compares estimates from first half and last half of draws to detect lack of convergence or insufficient replications.
  - Implementation uses codes of James P. LeSage for computing NSE and CD.

*Italic: Source: _wp1466 - REFERENCES (PDF chapter/section).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp1466.pdf_