## Macrofinance Model of the Czech Economy: Asset Allocation Perspective (_wp1278)

## Source details

**Canonical URL:** [Macrofinance Model of the Czech Economy: Asset Allocation Perspective (_wp1278)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1278.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1278.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1278.pdf.json)

---

### Abstract — key findings
- The paper develops a VAR macrofinance model of the Czech economy based on the Nelson-Siegel framework.
- Yield misalignments from the macro-implied yields partially determine subsequent yield changes over three to nine months.
- Misalignments tend to persist for a number of months. Persistence is explained by:
  - (a) the macro-economy influencing asset markets only at lower frequencies,
  - (b) liquidity effects particularly during times of capital inflows to Czech Republic,
  - (c) some misalignments not exceeding their historical one standard deviation.
- JEL Classification Numbers: C58, E43, G11
- Keywords: yield curve modeling, macrofinance modeling, vector auto-regression model

### Methodology and model structure
- Core specification:
  - Dynamic Nelson-Siegel representation of zero-coupon yields with time-varying latent factors β1 (level), β2 (slope), β3 (curvature).
  - Diebold-Li dynamic interpretation: treat β1t, β2t, β3t as latent factors following a VAR.
- Two-step empirical approach (Diebold, Li (2006) two-step):
  - Step 1: Yields-only Nelson-Siegel factors estimated by OLS across cross-section of yields; λ fixed at 0.56 (minimizing sum of squared residuals).
  - Step 2: Embed the latent factors in a VAR(1) with macro variables to obtain a macrofinance Nelson‑Siegel VAR.
- Rationale for statistical (Nelson-Siegel) vs no-arbitrage models:
  - Transitory arbitrage opportunities and low liquidity can make no-arbitrage restrictions misspecified.
  - No-arbitrage models require many more parameters and may increase estimation error.
  - Statistical Nelson-Siegel is parsimonious and practical for Czech context.

### Data, stylized facts, and PCA evidence
- Sample: monthly data from May 2000 to February 2010 on Czech government zero-coupon spot curve (maturities three months up to ten years), with short maturities up to six months from PRIBOR rates; Kladivko (2010) estimation procedures and exclusions applied.
- Additional monthly macro variables (May 2000 to February 2010): IPP_GAP, IPP_GAP_EUR, INFL, INFL_EUR, INFL_TARGET, EURCZK_GAP, EURIBOR_3M, GER_10Y.
- Stylized facts:
  - Average Czech yield curve is increasing and concave.
  - Persistence of yield dynamics is high.
  - Short end more volatile than long end; short rates more persistent than long yields.
- PCA on 3M–10Y yields (sample 2000M05–2010M02; 118 included observations; extracting 3 components):
  - Eigenvalues (Sum = 13, Average = 1):
    - Number 1: Value 12.02478; Proportion 0.9250; Cumulative Proportion 0.9250
    - Number 2: Value 0.916361; Proportion 0.0705; Cumulative Proportion 0.9955
    - Number 3: Value 0.056431; Proportion 0.0043; Cumulative Proportion 0.9998
  - Interpretation:
    - PC1 (~level) explains 92.5 percent of variation.
    - First three PCs explain 99.98 percent of variation.
    - PC2 (~slope) and PC3 (~curvature) loadings align with standard interpretations.

### VAR(1) Nelson-Siegel macrofinance model specification and estimation outputs
- Model vector Ft (11x1):
  - Ft = (β1t, β2t, β3t, IPP_GAPt, INFLt, EURCZK_GAPt, IPP_GAP_EURt, INFL_TARGETt, INFL_EURt, GER_10Yt, EURIBOR_3Mt)ᵗ
- VAR(1) form: Ft = c + ρ Ft-1 + εt; εt IID.
- Endogenous variables: β1, β2, β3, IPP_GAP, INFL, EURCZK_GAP.
- Exogenous variables: intercept, IPP_GAP_EUR, INFL_TARGET, INFL_EUR, GER_10Y, EURIBOR_3M.
- Selected autoregressive coefficients (lagged → current):
  - BETA1(-1) → BETA1: 0.702332
  - BETA2(-1) → BETA2: 0.898208
  - BETA3(-1) → BETA3: 0.579669
  - IPP_GAP(-1) → IPP_GAP: 0.525097
  - INFL(-1) → INFL: 0.759299
  - EURCZK_GAP(-1) → EURCZK_GAP: 0.859264
- Selected cross-lag and exogenous coefficients (examples):
  - BETA1(-1) → BETA3: -0.534034
  - BETA2(-1) → BETA1: -0.148055
  - IPP_GAP_EUR → IPP_GAP: 0.562117
  - INFL_TARGET → BETA1: 0.000959
  - GER_10Y → BETA1: 0.001857
  - EURIBOR_3M → IPP_GAP: 0.382345
- Adjusted R-squared by equation:
  - BETA1: 0.933425
  - BETA2: 0.917258
  - BETA3: 0.710352
  - IPP_GAP: 0.829275
  - INFL: 0.954243
  - EURCZK_GAP: 0.801448

### Estimation interpretation and diagnostics
- Persistence:
  - Level and slope highly persistent; all three yield factors highly persistent.
- Determinants:
  - Level: own lag, lagged slope and curvature, lagged IPP_GAP (sign opposite to expectation), lagged INFL (level rises when inflation rises), lagged EURCZK_GAP (level rises when currency depreciates), foreign long- and short-term rates.
  - Slope: lagged slope and curvature; lagged domestic IPP_GAP (short end increases when IPP_GAP increases); INFL_TARGET not significant.
  - Curvature: own lag, lagged level and slope; lagged INFL, lagged EURCZK_GAP, IPP_GAP_EUR, EURIBOR_3M significant.
  - IPP_GAP: own lag and IPP_GAP_EUR.
  - INFL: lagged level and curvature, lagged INFL persistence, EURCZK_GAP pass-through, INFL_EUR, GER_10Y.
- Diagnostic tests:
  - Lag order: 1.
  - Residual stationarity: stationary.
  - No residual serial correlation after lag 1 at 5% level.
  - Residual autocorrelations: none significant except β2 up to third lag.
  - Augmented Dickey-Fuller:
    - Stationary (I(0)): β3, IPP_GAP, EURCZK_GAP, INFL_EUR, IPP_GAP_EUR.
    - Non-stationary / I(1): β1, β2, INFL, EURIBOR_3M (exogenous), GER_10Y (exogenous), INFL_TARGET (exogenous).
  - Slight normality/kurtosis issues and mild heteroskedasticity (exchange rate gap equation).
  - VAR satisfied stability condition.

### Impulse responses and variance decomposition — key quantitative summaries
- Timing and directions (selected summaries):
  - Shock to IPP_GAP: slightly increases INFL in first five periods; negligible effect on EURCZK_GAP.
  - Shock to INFL: negligible effect on IPP_GAP; slightly appreciates domestic currency.
  - Shock to EURCZK_GAP (depreciation): slightly raises IPP_GAP; significantly increases INFL (strong pass-through).
  - Shock to BETA1 (level ↑): slightly turns IPP_GAP positive; increases INFL.
  - Shock to BETA2 (less steep): initially turns IPP_GAP positive then slightly negative; decreases INFL and appreciates currency.
- Variance decomposition highlights (periods labeled 1–10):
  - BETA1 decomposition (columns: BETA1 BETA2 BETA3 IPP_GAP INFL EURCZK_GAP):
    - Period 1: 100.0000 0.000000 0.000000 0.000000 0.000000 0.000000
    - Period 5: 84.86376 2.248344 2.305724 4.145620 1.843713 4.592840
    - Period 10: 77.25710 2.937539 3.231278 5.069988 2.485937 9.018158
  - BETA2 decomposition:
    - Period 1: 65.07486 34.92514 0.000000 0.000000 0.000000 0.000000
    - Period 5: 58.94736 28.46636 5.603015 5.938474 0.489865 0.554923
    - Period 10: 58.57061 25.17115 7.628816 7.300449 0.818796 0.510178
  - BETA3 decomposition:
    - Period 1: 3.084574 7.528247 89.38718 0.000000 0.000000 0.000000
    - Period 5: 3.011119 16.54306 71.97728 0.000299 1.220865 7.247376
    - Period 10: 2.852483 17.53539 66.21879 0.000331 1.201227 12.19178
- Macro variables variance decomposition (selected):
  - IPP_GAP: Period 1: 97.84057 percent explained by itself; Period 10: 88.38979 percent by itself.
  - INFL:
    - Period 1: INFL own = 86.38093
    - Period 2: INFL own = 77.24799
    - Period 6: INFL own = 42.83474; EURCZK_GAP = 24.48237
    - Period 10: BETA1 = 20.91725; BETA2 = 4.934916; BETA3 = 5.843199 (combined yield-factor contribution = 31.695365)
  - EURCZK_GAP: Period 1: 91.49292 by itself; Period 10: 88.87560 by itself; INFL contribution at period 10 = 4.875614
- Key interpretation:
  - At one-month horizon, yield-curve variation driven entirely by yield-curve factors.
  - Macroeconomy begins materially influencing yield-curve variation after ~five to six months.
  - By ten periods, macro variables explain almost 40 percent of variation in yield curve factors (aggregate statement in source).

### Making the macrofinance model operational for asset-allocation
- Definitions:
  - Macro-implied betas: beta factors from the VAR macrofinance model.
  - Macro-implied zero-coupon yield curve: Nelson‑Siegel curve constructed with macro-implied betas (fair-value curve).
  - Misalignment: actual zero-coupon yield minus macro-implied fair-value yield.
- Practical use:
  - Misalignments used to detect over-valuation / under-valuation relative to macro fundamentals.
  - Predictive exercise: compare current misalignments to subsequent yield changes lagged 3, 6, 9 months.
  - One-month lag produces unfavorable forecasting results; preferred horizon is six months (consistent with variance decomposition).

### Forecasting performance and regression evidence
- Out-of-sample forecast comparison (January 2007–September 2007; nine-month horizon):
  - Macrofinance VAR(1) (latent factors estimated until December 2006) vs yields-only VAR(2):
    - RMSE: macrofinance VAR has lower RMSE than yields-only VAR → improved forecasting with macro variables.
    - Note: yields-only model used VAR(2) per lag selection; using VAR(1) would have strengthened macrofinance dominance.
- Multiple regression drivers of 3Y yield changes over 3 months (sample 2000M11–2007M12; 86 observations):
  - Regression highlights (Table 11):
    - MIS_3Y_BOND(-3): Coefficient 0.941697; Std. Error 0.077110; t-Statistic 12.21232; Prob. 0.0000
    - VOLUME: Coefficient -4.74E-08; Std. Error 9.28E-09; t-Statistic -5.106903; Prob. 0.0000
    - INFL: Coefficient 0.001883; Std. Error 0.000365; t-Statistic 5.157560; Prob. 0.0000
    - EURCZK_GAP(-3): Coefficient 0.000449; Std. Error 0.000157; t-Statistic 2.860761; Prob. 0.0054
    - EURIBOR_3M(-3): Coefficient -0.003890; Std. Error 0.000537; t-Statistic -7.244333; Prob. 0.0000
    - GER_10Y: Coefficient 0.004588; Std. Error 0.000779; t-Statistic 5.888624; Prob. 0.0000
    - DUMMY2003*VOLUME: Coefficient 3.10E-08; Std. Error 1.48E-08; t-Statistic 2.086248; Prob. 0.0403
    - C: Coefficient -0.011095; Std. Error 0.002564; t-Statistic -4.327860; Prob. 0.0000
  - Goodness of fit:
    - R-squared 0.792256; Adjusted R-squared 0.770673
    - S.E. of regression 0.002452; F-statistic 36.70616; Prob(F-statistic) 0.000000
    - Mean dependent var -0.000726; S.D. dependent var 0.005120
  - Interpretation:
    - MIS_3Y_BOND(-3) is the strongest predictor of subsequent 3Y yield changes.
    - VOLUME negative coefficient suggests liquidity/demand effects supporting persistence of misalignments.
    - All included variables statistically significant at ten-percent level.

### Empirical illustrations, signal filtering, and preferred horizons
- Empirical illustrations show actual zero-coupon yields track macro-implied yields but with persistent misalignments in periods:
  - 2001–2004 and 2005–2008 (linked to rapid capital inflows).
- Forecast horizons and signal filtering:
  - Misalignments predict yield changes best at six months (preferred) and up to nine months; one-month corrections not supported.
  - One-standard-deviation rule extracts relatively few signals (2000–2009), implying:
    - Macrofinance framework not for high-frequency trading.
    - Investors should reshuffle allocations only occasionally based on macro signals.

### Investment illustration and performance metrics
- Example: hypothetical trading of 10Y bond CZECH REPUBLIC 2003 3.7% 16/06/13 from July 2003 to February 2010 using 1-standard-deviation misalignment rule.
  - Signals identified: 4
  - Initial investment price: CZK 91.22
  - Total cash profit from trading decisions: CZK 16.21
  - Approximate coupon profit during holdings: CZK 10.18
  - Total profit from strategy: CZK 26.39
  - Years not invested: 3.92
  - Approximate coupon profit from buy-and-hold alternative: CZK 24.3
  - Relative outperformance: at least 8.6 percent vs buy-and-hold during the trading period
- Caveats:
  - Proper implementation should trade duration, not a single issue.
  - Example excludes transaction costs, taxation, sequential real‑time signal evaluation.
  - Exercise is ex post; signals were identified retrospectively.

### Conclusions and research suggestions
- Conclusions:
  - Macro-economy determines a fair value of bond yields; actual yields tend to move toward macro-implied fair values over months.
  - Misalignments persist but are eventually eliminated; not all misalignments are economically actionable.
  - Including macro variables improves yield forecasting relative to yields-only models.
  - Liquidity (traded volume) contributes to longer persistence of misalignments.
- Suggestions for further research:
  - Test macro-implied yield curves in real-life yield-curve investing strategies and compare to buy-and-hold.
  - Develop multi-maturity and cross-asset allocation strategies based on macro-implied signals.
  - Investigate filtering rules (e.g., one-standard-deviation cutoff) to separate signal from noise and determine trading frequency.

*Prepared by Miroslav Kollar; WP/12/78; March 2012 — Macrofinance Model of the Czech Economy: Asset Allocation Perspective (Sections 1–6).*

### Section 1

### Macrofinance Model of the Czech Economy: Asset Allocation Perspective

### Abstract — key findings
- The paper develops a VAR macrofinance model of the Czech economy.
- Yield misalignments from the yields implied by the macrofinance model partially determine subsequent yield changes over three to nine months.
- These yield misalignments tend to persist for a number of months. The persistence is explained by:
  - (a) the fact that the macro-economy influences asset markets only at lower frequencies,
  - (b) the liquidity effect particularly during the times of capital inflows to Czech Republic, and
  - (c) the fact that not all misalignments were greater than their historical one standard deviation.
- JEL Classification Numbers: C58, E43, G11
- Keywords: yield curve modeling, macrofinance modeling, vector auto-regression model

### Research objective and motivation
- Bridge finance and macroeconomics by estimating a macrofinance model of the Czech economy based on the Nelson-Siegel framework.
- Produce a dynamic term structure (yield curve) driven by macroeconomic variables and investigate whether the actual yield curve reverts to the macro-implied (fair-value) yield curve.
- Study usefulness of the approach for yield forecasting and investment purposes.

### Methodology and model structure
- Core model: a VAR macrofinance specification that embeds the Nelson-Siegel representation of the yield curve (Diebold, Li (2006) interpretation).
- Nelson-Siegel framework: a parsimonious exponential functional form that models the whole zero-coupon yield curve with a small number of parameters; factors interpreted as level, slope, and curvature.
- Diebold-Li dynamic interpretation: treat Nelson-Siegel factors as time-varying latent factors following a vector autoregression.
- Empirical approach chosen: statistical macrofinance model (Nelson-Siegel + VAR) rather than no-arbitrage models, motivated by several practical considerations in the Czech context.

### Rationale for preferring statistical (Nelson-Siegel) approach over no-arbitrage models
- Transitory arbitrage opportunities may persist in less developed or less liquid parts of the yield curve; imposing no-arbitrage restrictions can cause misspecification.
- No-arbitrage models rely on the expectations hypothesis of the term structure (after adjusting for risk premium).
- No-arbitrage models are less successful in describing yield curve dynamics over time compared to cross-sectional fitting at a point in time.
- No-arbitrage macrofinance models require estimating many more parameters, increasing estimation error and complexity of interpretation.
- Even with best no-arbitrage models, a wide diversity in estimates of no-arbitrage values exists among experts and practitioners.

### Empirical facts and prior evidence used
- Principal components evidence (Litterman, Scheinkman (1991)): most behavior in government bond yield curves can be explained by up to three factors (interpreted as level, slope, curvature).
- Existing Czech literature: only a few comprehensive attempts to apply macrofinance models to Czech data; multiple studies exist on Czech term structure or relationships between macro variables and parts of the yield curve, but few that model the whole yield curve linked to macro variables.

### Intended empirical analyses (as described)
- Estimate yields-only Nelson-Siegel model and then embed Nelson-Siegel factors in a VAR with macroeconomic variables (creating a VAR(1) Nelson-Siegel macrofinance model).
- Use principal components analysis of Czech zero-coupon yields to validate three-factor structure.
- Perform diagnostic tests, impulse response analysis, and variance decompositions.
- Make the macrofinance model operational for asset-allocation purposes by:
  - Identifying misalignments between actual yields and macro-implied fair values,
  - Testing timing of subsequent yield movements,
  - Evaluating forecasting performance versus yields-only models,
  - Examining examples and aggregate market statistics (e.g., traded volume on the secondary Czech government bond market).

### Practical implications and applications
- The macro-implied yield curve can serve as a fair-value benchmark for yields given macroeconomic state.
- Detected misalignments and their persistence can inform yield-forecasting and tactical asset-allocation decisions over horizons of three to nine months.
- The model can be used for central bank and market analyst purposes, including incorporating longer-term interest rates into macroeconomic models and for financial stability analyses (e.g., stress testing).

*Prepared by Miroslav Kollar; WP/12/78; March 2012 — Macrofinance Model of the Czech Economy: Asset Allocation Perspective (Section 1).*

### Section 2

### _wp1278 - Section 2

### Nelson-Siegel yield curve representation and interpretation
- Static Nelson-Siegel exponential representation for the spot rate r(τ) expresses a zero-coupon bond maturing in τ periods as a linear combination of three factors with factor loadings; the three parameters β1, β2, β3 are interpretable as level, slope and curvature and λ>0 is the rate of exponential decay.
- Economic interpretations:
  - β1 (level): changes result in parallel shifts; loading does not change with maturity; as τ → ∞, spot rate converges to β1.
  - β2 (slope): as τ → 0, spot rate converges to β1 + β2; β2 is the difference between the long rate and the short spot rate; increases raise short yields more than long yields; loading equals one at zero maturity and approaches zero as maturity increases.
  - β3 (curvature): converges to zero for shortest and longest maturities, creating a hump if β3>0 or a U-shape if β3<0; loading starts at zero at shortest maturities, increases at intermediate maturities, and falls back to zero at longest maturities.
- Parameter λ determines the location of the hump (or U-shape) and exponential decay speed; larger λ → faster decay of yield curve curvature.
- Desirable properties of Nelson-Siegel (Diebold, Li, 2006):
  - Forward rates are positive along the whole yield curve.
  - Discount factor approaches zero with increasing maturity.
  - Yield curve has a limited number of shapes.
- Dynamic interpretation:
  - Treat level (β1t), slope (β2t), and curvature (β3t) as time-varying latent factors so the entire yield curve at any time is described by these three factors (dynamic Nelson-Siegel formulation, equation (2) / (3) in source).
  - Diebold, Li (2006) find the dynamic interpretation well suited for interest rate forecasting and dynamic analysis.

### Data used in the model
- Sample: monthly data from May 2000 to February 2010 on Czech government zero-coupon bond spot curve for maturities from three months up to ten years, as estimated by Kladivko (2010).
- Short maturities (up to six months): Czech interbank money market rates used (one-week, one-month and three-month PRIBOR rates). These interbank rates were transformed into zero-coupon bond prices for yield-curve estimation; zero-coupon yields up to one year are affected by both interbank money market and bond market.
- Maturities from six months up to ten years: Czech government bond yield data from Prague Stock Exchange (mid quotes for six months, nine months, and one- to ten-years maturities).
- Bonds excluded from estimation by Kladivko (2010):
  - All bonds with less than 180 days to maturity (low liquidity).
  - For the first tranche of each bond, all bonds before they reach 30 days after the issue date (odd behavior of price quotes of some new issues).
  - The 6.08%/2001 bond (issue number 27) excluded (constantly over-priced).
  - The 4.85%/2057 bond (issue number 54) excluded (not actively traded).
  - All floating interest rate bonds excluded.
- Reasons for starting estimation from year 2000:
  - Insufficient number of bonds available before 1999; Kladivko (2010) needed at least four bonds to estimate the yield curve.
  - Avoid structural breaks: inflation targeting monetary-policy regime implemented from 1998; from 1998 to 2000 CNB conducted disinflationary policy with high interest rates and high volatility. Bonds issued before 1998 excluded for differing taxation regimes.
- Additional monthly macroeconomic variables (May 2000 to February 2010):
  - Index of seasonally adjusted real industrial production in Czech Republic and Euro Area; real industrial production gap estimated via Hodrick-Prescott filter (IPP_GAP, IPP_GAP_EUR).
  - Year-on-year changes of domestic and Euro Area consumer price-inflation rate (INFL, INFL_EUR).
  - Inflation target set by the Czech National Bank (INFL_TARGET).
  - CZK/EUR exchange rate and exchange rate gap via Hodrick-Prescott filter (EURCZK_GAP).
  - 3M EURIBOR (EURIBOR_3M) and 10Y German government bond yield (GER_10Y).

### Stylized facts and descriptive properties of Czech yield data
- Stylized facts inferred:
  - The average Czech yield curve is increasing and concave.
  - The persistence of yield dynamics is high.
  - The short end of the yield curve is more volatile than the long end.
  - Short rates are relatively more persistent than long yields.
- Figures and tables described (average yield curve, descriptive statistics, autocorrelations) support the above stylized facts.

### Principal Components Analysis (PCA) of Czech zero-coupon yields
- PCA applied to 3-month to 10-year Czech government zero-coupon bond yields (sample 2000M05–2010M02).
- Extraction details (from source table):
  - Sample (adjusted): 2000M05 2010M02
  - Included observations: 118 after adjustments
  - Extracting 3 of 13 possible components
  - Eigenvalues (Sum = 13, Average = 1):
    - Number 1: Value 12.02478; Difference 11.10842; Proportion 0.9250; Cumulative Value 12.02478; Cumulative Proportion 0.9250
    - Number 2: Value 0.916361; Difference 0.859930; Proportion 0.0705; Cumulative Value 12.94114; Cumulative Proportion 0.9955
    - Number 3: Value 0.056431; Difference 0.054064; Proportion 0.0043; Cumulative Value 12.99757; Cumulative Proportion 0.9998
- Key PCA findings:
  - The first principal component explains as much as 92.5 percent of the variation in Czech government bond yields.
  - The first three components cumulatively explain up to 99.98 percent of the variation.
- Interpretation of principal components via factor loadings (Figure 3):
  - PC1: positive and uniform across maturities → approximates level of interest rates.
  - PC2: factor loading increasing with maturity → approximates slope of the yield curve.
  - PC3: U-shaped loading → approximates curvature of the yield curve.

### Empirical and macroeconomic proxies for principal components
- First principal component:
  - Common empirical approximation of level: an empirical proxy (formula shown in source).
  - Suggested macroeconomic approximation: inflation rate.
  - Observation: PC1 corresponds well to inflation rate and empirical level proxy.
- Second principal component:
  - Common empirical approximation of slope: empirical slope proxy (formula shown in source); slope is inverted for the exercise.
  - Suggested macroeconomic approximation: industrial-production gap (IPP_GAP).
  - Observation: PC2 and the slope proxy exhibit very similar behavior; co-movement between PC2 and IPP_GAP has increased in recent years of the sample.
- Third principal component:
  - Common empirical approximation of curvature: empirical curvature proxy (formula shown in source).
  - No straightforward macroeconomic approximation for curvature.
  - Observation: empirical curvature proxy tracks PC3 very well.

### From yields-only Nelson-Siegel to VAR macrofinance Nelson-Siegel model
- Modeling approach:
  - Two-step estimation following Diebold, Li (2006):
    - Step 1: Estimate the dynamic Nelson-Siegel specification using ordinary least squares regression on the cross-section of yields to obtain yields-only Nelson-Siegel β factors for level, slope, curvature for each period.
    - Step 2: (implied) introduce macroeconomic variables in a VAR framework to form a macrofinance Nelson-Siegel model (Diebold, Li two-step approach retained for parsimony vs. one-step state-space approach).
  - Lambda selection: λ fixed at 0.56 so as the sum of squared residuals of the Nelson-Siegel regression is minimized.
- Properties of yields-only Nelson-Siegel factor estimates (summary from source Table 3 discussion):
  - Both the average empirical slope and the estimated Nelson-Siegel slope factor are positive.
  - The Nelson-Siegel slope factor is more volatile that the empirical slope.
  - The level (both empirical and derived) is highly persistent.
  - The slope (both empirical and derived) is more persistent than the curvature.
- Notes:
  - Diebold, Rudebush, Aruoba (2005) used a one-step state-space approach, which requires estimation of many parameters; the present work follows Diebold, Li (2006) two-step approach.
  - The Nelson-Siegel representation’s slope is the negative of the traditional slope (i.e., as long yield minus short yield).

*Source: _wp1278 - Section 2*

### Section 3

### _wp1278 - Section 3

### Yields-only Nelson-Siegel factors and empirical/macro proxies
- The yields-only Nelson-Siegel level factor closely tracks the empirical proxy of the level of interest rates defined in the text.
- The level factor and its suggested macro-economic approximation of inflation rate are less closely related.
- The yields-only Nelson-Siegel slope factor follows very closely the empirical approximation of the slope factor defined in the text.
- The slope factor is also closely related to its macro-economic approximation in terms of industrial production gap, although the co-movement is weaker than with the empirical approximation.
- The yields-only Nelson-Siegel curvature factor closely follows its empirical approximation; there exists no meaningful macro-economic approximation of the curvature of the yield curve.

### VAR(1) Nelson-Siegel macrofinance model specification
- Model form (non-structural VAR(1)):
  - Ft = (β1t, β2t, β3t, IPP_GAPt, INFLt, EURCZK_GAPt, IPP_GAP_EURt, INFL_TARGETt, INFL_EURt, GER_10Yt, EURIBOR_3Mt)ᵗ — an (11x1) vector of model variables.
  - ρ is the (11x11) matrix of autoregressive coefficients.
  - c is a (11x1) vector of intercept terms.
  - εt is a (11x1) vector of IID error term.
- Endogenous variables in estimation: latent Nelson-Siegel factors (from yields-only model), industrial production gap, domestic inflation rate, EURCZK exchange rate gap.
- Exogenous variables: intercept, EUR industrial production gap, domestic inflation target, EUR inflation rate, German 10Y government bond yield, 3M EURIBOR.
- Rationale: exogenous treatment of abroad variables and domestic inflation target justified for a small open economy.

### VAR(1) estimation output — key statistics (Table 4 highlights)
- Number of model variables in Ft: 11.
- Selected autoregressive coefficients (lagged variables → current equations):
  - BETA1(-1) → BETA1: 0.702332
  - BETA2(-1) → BETA2: 0.898208
  - BETA3(-1) → BETA3: 0.579669
  - IPP_GAP(-1) → IPP_GAP: 0.525097
  - INFL(-1) → INFL: 0.759299
  - EURCZK_GAP(-1) → EURCZK_GAP: 0.859264
- Selected cross-lag coefficients (examples):
  - BETA1(-1) → BETA3: -0.534034
  - BETA2(-1) → BETA1: -0.148055
  - BETA3(-1) → BETA1: -0.061912
  - IPP_GAP(-1) → BETA1: -0.000328
  - INFL(-1) → BETA1: 0.000728
  - EURCZK_GAP(-1) → BETA1: 0.000262
- Selected exogenous coefficients (examples):
  - IPP_GAP_EUR → IPP_GAP: 0.562117
  - INFL_TARGET → BETA1: 0.000959
  - INFL_EUR → INFL: 0.560488
  - GER_10Y → BETA1: 0.001857
  - EURIBOR_3M → IPP_GAP: 0.382345
- Adjusted R-squared by equation:
  - BETA1: 0.933425
  - BETA2: 0.917258
  - BETA3: 0.710352
  - IPP_GAP: 0.829275
  - INFL: 0.954243
  - EURCZK_GAP: 0.801448

### Estimation results and economic interpretation (main findings)
- Persistence:
  - The level and slope parameters are highly persistent.
  - All three yield curve factors are highly persistent.
- Determinants of the level factor:
  - Explained by its own lag, lagged slope factor, lagged curvature factor, lagged domestic industrial production gap (sign opposite to expectation), lagged domestic inflation rate (level of interest rates rises when inflation rises), lagged EURCZK exchange rate gap (level rises when domestic currency depreciates), and foreign long-term and short-term interest rates.
- Determinants of the slope factor:
  - Statistically significant determinants: lag in the slope factor and lagged curvature factor; followed by lagged domestic industrial production gap (short end increases when IPP_GAP increases, suggesting central bank stabilization role).
  - Domestic inflation target not statistically significant for the slope.
- Determinants of the curvature factor:
  - Determined by its own lag, lagged level and slope factors; statistically significant are lagged domestic inflation rate, lagged EURCZK exchange rate gap, foreign industrial production gap, and foreign short-term interest rates.
- Domestic industrial production gap:
  - Explained by its own lag and by foreign industrial production gap — evidence of openness and dependence on Euro Area industrial developments.
  - Lagged slope of interest rates not a statistically significant determinant.
- Domestic inflation rate:
  - Primarily explained by lagged level factor and curvature factor, followed by lag in domestic inflation (high persistence), lagged EURCZK exchange rate gap (exchange rate pass-through), foreign inflation rate (role of Euro Area), and foreign long-term interest rates.
- EURCZK exchange rate gap: highly persistent.

### Diagnostic tests
- Lag order: VAR lag order selection criteria indicated lag of order one appropriate.
- Residual stationarity: all residuals stationary.
- Serial correlation: VAR Residual Serial Correlation LM test showed no residual serial correlation after lag of order one at five-percent significance level.
- Residual autocorrelations:
  - Not statistically significant at any lag for: beta1, beta3, domestic industrial production gap, domestic inflation rate, EURCZK exchange rate gap.
  - Statistically significant up to third lag for: beta2.
- Augmented Dickey-Fuller unit-root tests:
  - Stationary series: beta3, domestic industrial production gap, EURCZK exchange rate gap, foreign inflation rate, foreign industrial production gap.
  - Non-stationary / integrated of order one: beta1, beta2, domestic inflation rate, EURIBOR 3M rate (exogenous), German 10Y government bond yield (exogenous), domestic inflation target (exogenous).
- Rationale for mixing I(0) and I(1) variables:
  - Following Sims et al. (1990), consistent VAR coefficient estimates can be obtained even when unit roots are present; differencing may lose information; model run with both stationary and non-stationary data as theoretically reasonable.
- Other diagnostics:
  - Slight problem with normality of residuals in terms of kurtosis (possibly due to financial market data).
  - White Heteroskedasticity test indicates possible slight heteroskedasticity, particularly in exchange rate gap equation.
  - VAR model satisfied stability condition.

### Impulse responses and variance decomposition — summarized responses
- Responses of macro variables to macro shocks:
  - Shock to domestic IPP_GAP: slightly increases inflation in first five periods; negligible effect on exchange rate gap.
  - Shock to inflation: negligible effect on IPP_GAP; slightly appreciates domestic currency throughout periods (consistent with FX expectations of monetary tightening).
  - Shock to exchange rate gap (more depreciation): slightly shifts IPP_GAP into positive territory (consistent with export orientation); exchange rate shock significantly increases domestic inflation (strong pass-through).
- Responses of macro variables to yield curve shocks:
  - Shock to level factor (increase in level of interest rates): slightly turns IPP_GAP positive, negligible effect on exchange rate gap, increases inflation rate (consistent with level ~ long-term inflation expectations).
  - Shock to slope factor (less steep yield curve): initially turns IPP_GAP positive but eventually slightly negative; decreases inflation rate and appreciates domestic currency (reflecting tighter monetary policy).
  - Shock to curvature factor: no straightforward economic interpretation.
- Responses of yield curve to macro shocks:
  - Positive shock to IPP_GAP: slightly decreases level of interest rates over ten periods; makes yield curve less steep (shorter rates decrease by less than longer rates) — consistent with central bank stabilization role or output-gap role in inflation-targeting regimes.
  - Shock to inflation: increases level of interest rates over horizon (link between inflation expectations and level factor); negligible reaction of slope factor.
  - Shock to exchange rate gap (more depreciation): increases level of interest rates over horizon; negligible effect on slope factor — consistent with market pricing of potential interest rate hikes given strong exchange rate pass-through.
- Responses of yield curve to yield curve shocks:
  - Shock to level factor: increases slope of yield curve (longer end increases more after level increase) — consistent with level as long-term inflation expectations and potential monetary tightening.
  - Shock to slope factor (decrease in slope): decreases level of interest rates.

*Source: _wp1278 - Section 3 (source PDF: _wp1278 - Section 3).*

### Section 4

### _wp1278 - Section 4

### Impulse Responses (Figure 11) — qualitative note
- Figure 11: Impulse Responses from the Macrofinance Model. Response to Cholesky One S.D. Innovations ± 2 S.E.
- Impulse-response panels present responses of BETA1, BETA2, BETA3, IPP_GAP, INFL, and EURCZK_GAP to shocks in each variable over horizons labeled 2 4 6 8 10 (plots show numerical axis ranges specific to each panel as in the figure).

### Variance decomposition — main summary
- The variance decomposition output from the VAR macrofinance model provides insight into dynamics between the yield curve and the macro-economy.
- At a one-month horizon, the variation in the yield curve is entirely driven by the yield curve itself, unrelated to macroeconomic variables.
- The macro-economy begins to add to the variation in the yield curve only in later periods; by the ten-period horizon, macroeconomic variables explain almost 40 percent of the variation in the yield curve factors.

### Variance Decomposition — Yield Curve Factors (Table 5)
- Variance Decomposition of BETA1 (columns: BETA1 BETA2 BETA3 IPP_GAP INFL EURCZK_GAP):
  - Period 1: 100.0000  0.000000  0.000000  0.000000  0.000000  0.000000
  - Period 2: 96.17831  0.644781  0.565258  1.275302  0.442138  0.894213
  - Period 3: 91.65966  1.343245  1.267111  2.583572  1.000655  2.145760
  - Period 4: 87.84332  1.876654  1.860399  3.528308  1.480847  3.410477
  - Period 5: 84.86376  2.248344  2.305724  4.145620  1.843713  4.592840
  - Period 6: 82.56427  2.501566  2.625788  4.534586  2.100288  5.673499
  - Period 7: 80.77008  2.674580  2.853671  4.775991  2.273055  6.652623
  - Period 8: 79.34580  2.794239  3.017717  4.924554  2.383740  7.533949
  - Period 9: 78.19643  2.878124  3.138804  5.015243  2.450147  8.321248
  - Period 10: 77.25710  2.937539  3.231278  5.069988  2.485937  9.018158

- Variance Decomposition of BETA2:
  - Period 1: 65.07486  34.92514  0.000000  0.000000  0.000000  0.000000
  - Period 2: 62.86656  33.65239  1.286973  1.829643  0.113794  0.250636
  - Period 3: 60.92349  31.72071  2.964828  3.689515  0.256902  0.444548
  - Period 4: 59.65992  29.92485  4.454481  5.043561  0.384522  0.532673
  - Period 5: 58.94736  28.46636  5.603015  5.938474  0.489865  0.554923
  - Period 6: 58.59741  27.35090  6.418665  6.507439  0.576613  0.548970
  - Period 7: 58.46395  26.52480  6.963084  6.862600  0.649727  0.535837
  - Period 8: 58.45044  25.92323  7.307112  7.082250  0.713029  0.523944
  - Period 9: 58.49759  25.48766  7.512991  7.217457  0.768941  0.515363
  - Period 10: 58.57061  25.17115  7.628816  7.300449  0.818796  0.510178

- Variance Decomposition of BETA3:
  - Period 1: 3.084574  7.528247  89.38718  0.000000  0.000000  0.000000
  - Period 2: 3.202990  11.13543  84.07904  0.000350  0.437091  1.145102
  - Period 3: 3.174594  13.78921  79.07070  0.000335  0.872790  3.092369
  - Period 4: 3.093995  15.51434  75.01836  0.000309  1.125443  5.247553
  - Period 5: 3.011119  16.54306  71.97728  0.000299  1.220865  7.247376
  - Period 6: 2.944768  17.10945  69.79587  0.000288  1.230290  8.919339
  - Period 7: 2.898358  17.39172  68.28536  0.000294  1.210187  10.21408
  - Period 8: 2.869643  17.51086  67.27465  0.000317  1.192147  11.15238
  - Period 9: 2.855291  17.54349  66.62282  0.000332  1.188701  11.78937
  - Period 10: 2.852483  17.53539  66.21879  0.000331  1.201227  12.19178

- Key numeric findings for yield curve factors:
  - The level factor (BETA1) variation is driven primarily by its own variation; by period 5 the combined effect of EURCZK_GAP and IPP_GAP accounts for 10 to 15 percent of BETA1 variation (text summary).
  - The slope factor (BETA2) is driven by its own variation only to approximately 35 to 25 percent (text); the level factor explains around 60 percent of the variation in the slope factor.
  - The macro-economy (IPP_GAP) begins to influence the slope factor at the four-month horizon; IPP_GAP accounts for 5 to 7 percent of variation in the slope factor.
  - The curvature factor (BETA3) is explained by its own variation to as much as 89 percent at short horizons, with the slope factor contributing up to 17 percent at longer horizons. EURCZK_GAP accounts for 5 to 12 percent of curvature variation within the ten periods-horizon.

### Variance Decomposition — Macroeconomic Variables (Table 6)
- Variance Decomposition of IPP_GAP:
  - Period 1: 0.010616  1.685220  0.463589  97.84057  0.000000  0.000000
  - Period 2: 0.476626  1.391647  0.378490  96.77831  0.463567  0.511358
  - Period 3: 1.100097  1.357173  0.393284  94.72476  1.042928  1.381759
  - Period 4: 1.620380  1.452991  0.533606  92.67506  1.522918  2.195049
  - Period 5: 1.956378  1.549039  0.804156  91.07901  1.872851  2.738561
  - Period 6: 2.135114  1.599476  1.140518  89.99224  2.121512  3.011144
  - Period 7: 2.211820  1.611923  1.469511  89.30427  2.299621  3.102860
  - Period 8: 2.234059  1.607587  1.744711  88.87622  2.427885  3.109540
  - Period 9: 2.233557  1.602502  1.949569  88.59592  2.519217  3.099238
  - Period 10: 2.228181  1.603914  2.087693  88.38979  2.582333  3.108092

- Variance Decomposition of INFL:
  - Period 1: 1.991394  0.267732  2.301584  9.058361  86.38093  0.000000
  - Period 2: 5.513484  0.196782  7.322385  7.547012  77.24799  2.172349
  - Period 3: 9.412245  0.652740  10.11514  5.890891  66.91840  7.010585
  - Period 4: 12.84048  1.508685  10.62588  4.586378  57.31738  13.12120
  - Period 5: 15.50122  2.463342  9.901441  3.701785  49.24990  19.18231
  - Period 6: 17.43721  3.309270  8.788686  3.147722  42.83474  24.48237
  - Period 7: 18.80607  3.966915  7.725268  2.816890  37.87936  28.80549
  - Period 8: 19.76589  4.435031  6.872454  2.626796  34.10733  32.19251
  - Period 9: 20.44029  4.744881  6.254930  2.522551  31.25812  34.77922
  - Period 10: 20.91725  4.934916  5.843199  2.469671  29.11705  36.71791

- Variance Decomposition of EURCZK_GAP:
  - Period 1: 0.017198  3.499733  2.867031  0.979076  1.144046  91.49292
  - Period 2: 0.012421  3.434781  1.821388  0.661659  1.361928  92.70782
  - Period 3: 0.009781  3.292294  1.422503  0.516990  1.710241  93.04819
  - Period 4: 0.015046  3.127266  1.368242  0.446705  2.152848  92.88989
  - Period 5: 0.037475  2.971524  1.495758  0.410707  2.651719  92.43282
  - Period 6: 0.087458  2.843503  1.708045  0.392152  3.169439  91.79940
  - Period 7: 0.173723  2.752609  1.943606  0.383498  3.672834  91.07373
  - Period 8: 0.301034  2.701341  2.163891  0.381120  4.135580  90.31703
  - Period 9: 0.469007  2.686866  2.347324  0.383156  4.539654  89.57399
  - Period 10: 0.672236  2.702699  2.485284  0.388562  4.875614  88.87560

- Key numeric findings for macro variables:
  - IPP_GAP variation is mainly determined by IPP_GAP itself (e.g., 97.84057 percent at period 1; 88.38979 percent at period 10). Above the six-month horizon, combined effect of INFL and EURCZK_GAP is around 5 percent and yield curve factors combined around 5 percent.
  - INFL variation is determined by its own variation in the first two periods (86 to 77 percent in text; table: 86.38093 at period 1, 77.24799 at period 2), but this falls quickly; at period 6 INFL's own share is 42.83474 and EURCZK_GAP accounts for 24.48237. By period 10 the combined contribution of yield curve factors to INFL is approximately 20.91725 + 4.934916 + 5.843199 = 31.695365 percent (table lists each component separately; text states approximately 30 percent at ten-month horizon).
  - EURCZK_GAP variation is largely its own: 91.49292 at period 1 and 88.87560 at period 10. INFL contributes 4.875614 at period 10, while slope and curvature factors contribute 2 to 3 percent each; combined yield-curve factors contribute approximately 6 percent at period 10 (text).

### Interpretation and conclusions from variance decompositions
- The macro-economy influences the variation in the yield curve significantly only after approximately five months.
- At the ten-period horizon, the variation in the macroeconomic variables explains almost 40 percent of the variation in the yield curve factors.
- For inflation rate specifically, there is a statistically significant interaction between the yield curve factors and the macro-economy; the macro-economy enters this interaction only after some time (i.e., not immediately).
- Note: variation of yield curve factors is significantly explained by their own autocorrelations; these autocorrelations may themselves include information about the macro-economy.

### Making the Macrofinance Model Operational for Asset-Allocation Purposes (Section IV) — definitions and approach
- Definitions:
  - The beta factors extracted from the VAR macrofinance model are called the macro-implied betas.
  - Inserting the macro-implied betas back into the Nelson-Siegel equation (2) yields the macro-implied zero-coupon yield curve (the fair-value zero-coupon yield curve given by macroeconomic fundamentals).
  - The difference between the fair-value yield curve and the actual zero-coupon yield curve is called the misalignment.
- Purpose:
  - Use the macro-implied zero-coupon yield curve and the actual zero-coupon yield curve to evaluate over-valuation or under-valuation of the actual curve relative to macroeconomic fundamentals.
  - Evaluate the extent to which yield-curve misalignments predict subsequent yield movements (measure predictive power of misalignments in explaining changes in bond yields).

### Empirical illustrations — Czech yields and spreads (figures 12–13) — qualitative summary
- The selected actual Czech zero-coupon yields track their macro-implied counterparts rather well, but misalignments exist and persist for some time.
- Persistence of bond overvaluations is particularly large for longer-term bond yields in periods between 2001 and 2004, and between 2005 and 2008 (text links these periods to rapid capital inflows into the Czech Republic).
- Yield-curve spreads and butterfly spreads show misalignments that persist for longer periods, coinciding with the same periods of rapid capital inflows.

### Identifying misalignments and timing subsequent yield movements — methodology and key practical points
- Constructed misalignments for three benchmark bonds: three-year, five-year, and ten-year bonds.
- To capture forecasting ability, subsequent changes in yields for these bonds were lagged by three, six and nine months and compared with current misalignments.
  - Postulate: current misalignments predict yield changes over three, six, or nine months.
  - The one-month lag did not produce favorable results, implying misalignments are not corrected within very short periods.
- This finding is consistent with VAR results showing macroeconomic variables started to affect yield curve variation after no less than five months.

*Source: own calculations*

### Section 5

### _wp1278 - Section 5

### Forecasting performance of macro-implied misalignments
- Current misalignments in the Czech yield curve produce good “forecasts” of yield changes over six to nine months, but forecasts are asymmetric over nine months and tilted towards positive yield changes.
- Preferred forecast horizon: six months (consistent with variance decomposition results from the macrofinance VAR model showing that the macro-economy starts influencing the yield curve after approximately six months).
- Highest correlation coefficients generally obtained for the 3Y bond and the 5Y bond, in contrast to the 10Y bond.
- Lags examined: 3 months, 6 months, and 9 months. Mean implied changes are the same by definition across these lags but start on different dates due to lagging.

### Summary of figures and interpretation
- Figures 14–16 (implied vs. realized yield change over 3, 6, and 9 months) show:
  - Mean implied positive and negative yield changes (3Y, 5Y, 10Y).
  - Mean realized positive and negative yield changes (lagged 3, 6, 9 months).
  - Over nine months, correlations are highest overall but the realized changes are asymmetric.
- Figure 17: bond yield misalignments and subsequent bond yield changes over six months lagged six months show lagged changes closely track current misalignments; accumulated misalignment over a short period contributes strongly to subsequent yield changes.

### Out-of-sample forecasting comparison (macrofinance VAR vs yields-only)
- Experimental setup:
  - Macrofinance VAR(1) of latent factors estimated until December 2006; forecasts produced January 2007–September 2007; latent factors inserted into Nelson‑Siegel equation to obtain macro-implied zero‑coupon yield forecasts.
  - Yields-only model: VAR(2) of latent factors estimated until December 2006; forecasts produced January 2007–September 2007; latent factors inserted into Nelson‑Siegel equation to obtain yields-only zero‑coupon yield forecasts.
  - Nine-month forecast horizon chosen.
  - Forecast period deliberately excludes the 2008 global financial crisis.
- Result summary:
  - RMSE comparison (Table 10) supports the macrofinance VAR model: its RMSE is lower than that of the yields-only model, indicating improved forecasting ability when macroeconomic variables are included.
  - Note: VAR(2) chosen for yields-only model due to lag-length selection criteria; had VAR(1) been used for yields-only, dominance of macrofinance model would be even stronger.

### Multiple regression analysis: drivers of 3Y yield changes over 3 months
- Regression setup:
  - Dependent variable: change in three-year Czech government bond yield over three months.
  - Sample (adjusted): 2000M11–2007M12, included observations: 86 after adjustments.
  - Independent variables: MIS_3Y_BOND(-3), VOLUME, INFL, IPP_GAP(-1), EURCZK_GAP(-3), EURIBOR_3M(-3), GER_10Y, DUMMY2003*VOLUME.
  - Regression period chosen to avoid post-2008 crisis disturbance.
- Key regression results (Table 11):
  - C: Coefficient -0.011095; Std. Error 0.002564; t-Statistic -4.327860; Prob. 0.0000
  - MIS_3Y_BOND(-3): Coefficient 0.941697; Std. Error 0.077110; t-Statistic 12.21232; Prob. 0.0000
  - VOLUME: Coefficient -4.74E-08; Std. Error 9.28E-09; t-Statistic -5.106903; Prob. 0.0000
  - INFL: Coefficient 0.001883; Std. Error 0.000365; t-Statistic 5.157560; Prob. 0.0000
  - IPP_GAP(-1): Coefficient 0.000237; Std. Error 0.000139; t-Statistic 1.696624; Prob. 0.0938
  - EURCZK_GAP(-3): Coefficient 0.000449; Std. Error 0.000157; t-Statistic 2.860761; Prob. 0.0054
  - EURIBOR_3M(-3): Coefficient -0.003890; Std. Error 0.000537; t-Statistic -7.244333; Prob. 0.0000
  - GER_10Y: Coefficient 0.004588; Std. Error 0.000779; t-Statistic 5.888624; Prob. 0.0000
  - DUMMY2003*VOLUME: Coefficient 3.10E-08; Std. Error 1.48E-08; t-Statistic 2.086248; Prob. 0.0403
  - R-squared 0.792256; Adjusted R-squared 0.770673
  - Mean dependent var -0.000726; S.D. dependent var 0.005120
  - S.E. of regression 0.002452; Sum squared resid 0.000463; Log likelihood 399.6551
  - Akaike info criterion -9.085002; Schwarz criterion -8.828151
  - F-statistic 36.70616; Prob(F-statistic) 0.000000
  - Durbin-Watson stat 1.194986
- Interpretation:
  - Regression has relatively high explanatory power (adjusted R-squared 0.770673).
  - All included variables are statistically significant at the ten-percent level.
  - MIS_3Y_BOND(-3) has the largest coefficient in absolute terms and the expected positive sign, indicating misalignment strongly predicts subsequent yield change.
  - VOLUME has a negative coefficient (small magnitude), consistent with greater demand lowering yields.
  - DUMMY2003*VOLUME significant, reflecting the impact of substantial traded volume in 2002–2003 on yield dynamics and persistence of misalignments.

### Illustrative example and practical application
- Example (May 31, 2010): actual Czech zero-coupon yield curve was undervalued relative to macro-implied fair value; after nine months, the undervaluation was almost entirely corrected and the actual curve approached the macro-implied fair value from nine months earlier.
- Table 9 and Figure 18 illustrate a randomly chosen day: misalignment for yields, spreads, and butterflies and subsequent nine-month changes show the yield curve moving toward prior macro-implied signals.
- One-standard-deviation rule on misalignments (2000–2009) extracts relatively few signals, implying:
  - Macrofinance framework is not intended for high-frequency trading.
  - Investors should reshuffle allocations only occasionally based on macroeconomic signals.

### Conclusions and suggestions for further research
- Main conclusions:
  - The macro-economy determines a fair value of bond yields toward which actual yields tend to move; misalignments persist for months but are eventually eliminated.
  - Not all misalignments are statistically significant; small misalignments may not justify trading.
  - The macrofinance modeling framework proved relevant even when assessed across crisis and non-crisis periods (Nelson‑Siegel framework noted as robust during stressed markets).
- Suggestions for further research:
  - Test macro-implied yield curves in real-life yield-curve investing strategies; evaluate strategies using market data and compare to buy-and-hold approaches.
  - Develop strategies for investing across the yield curve (buying/selling bonds of various maturities based on macro-implied signals) and for asset allocation across asset classes (e.g., bonds versus equities).
  - Investigate filtering rules (e.g., one-standard-deviation cutoff) to separate signal from noise and determine appropriate trading frequency based on macro-implied misalignments.

*Source: _wp1278 - Section 5, own calculations*

### Section 6

### _wp1278 - Section 6

### Illustration of a macrofinance-based investment strategy
- Hypothetical trading evaluated: 10Y Czech Government bond issue CZECH REPUBLIC 2003 3.7% 16/06/13 S.40 from July 2003 until February 2010.
- Trading rule: trade when the 10Y bond yield misalignment from its macro-implied value exceeds one historical standard deviation; “buy” when undervalued, “sell” when overvalued.
- Signals identified during July 2003 to February 2010: 4 signals representing statistically significant bond yield misalignments.
- Investment performance details:
  - Initial investment: CZK 91.22 (the market price of the bond at the first purchase).
  - Total cash profit from the four buy and sell decisions: CZK 16.21.
  - Approximate coupon profit during the periods the bond was held: CZK 10.18.
  - Total profit from the investment strategy: CZK 26.39.
  - Years not invested with the bond during the trading period: 3.92 years.
  - Approximate cash coupon profit from an alternative buy-and-hold strategy for the same bond during the same period: 24.3 CZK.
  - Relative performance: the macrofinance-based investment strategy outperformed the buy-and-hold strategy by at least 8.6 percent during the trading period.
- Caveats and limitations noted:
  - Proper implementation should trade ten-year duration, not a single ten-year bond.
  - Example does not account for transaction costs, taxation, or trading until full maturity.
  - The exercise is ex post: the model was run through the whole time span and signals identified retrospectively rather than evaluated sequentially at each observation.
  - Some misalignments were within one historical standard deviation and may represent random noise not warranting trading.

### Main findings on macrofinance modeling and yield dynamics
- Model framework:
  - Macrofinance approach implemented via the dynamic version of the Nelson-Siegel model (parsimonious, single exponential functional form, up to three factors).
  - Resulting curve named the macro-implied (or fair-value) yield curve.
- Key empirical findings:
  - Macroeconomic factors tend to have a significant influence on the yield curve after approximately five to six months.
  - Current misalignments of actual yields from the macro-implied yields partially determine subsequent yield changes over three to nine months periods.
  - Incorporating the macro-economy into the yield curve model improved forecasting ability compared to a yields-only VAR model.
  - Multiple regression results: current misalignments played an important part in explaining subsequent yield changes; traded volume was also significant, suggesting a liquidity effect may contribute to longer persistence of misalignments.

### Interpretation of misalignment persistence and drivers
- Misalignments from the macro-implied fair value tend to persist for a number of months but are eventually eliminated; the macro-economy dominates bond yield behavior in the long run.
- Potential explanations for persistence:
  - The macro-economy influences asset markets more strongly at lower frequencies.
  - Liquidity effects, particularly during times of capital inflows that were substantially absorbed by the government bond market.
  - Some misalignments were not greater than their historical one standard deviation and thus may represent random noise not prompting trading by active participants.

### Practical implications and guidance
- Macro-implied misalignments can be used as trading signals when they exceed historical one standard deviation, but:
  - Implementations should consider using duration-based trades rather than single-issue trades.
  - Real-world application must account for transaction costs, taxes, sequential signal evaluation, and liquidity considerations.
  - Idle cash during uninvested periods could be deployed to other assets, potentially improving realized performance beyond the simple example.

*Source: _wp1278 - Section 6*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1278.pdf_
