Journal of Forecasting, J. Forecast.

36, 91–99 (2017)

Published online 25 April 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/for.2416

Yield Curve Forecasting with the Burg Model

PIERRE ROSTAN,1* RACHID BELHACHEMI2 AND FRANÇOIS-ERIC RACICOT3
1
Department of Management, American University in Cairo, Egypt
2
Department of Mathematical Science, Xi’an Jiaotong-Liverpool University, Jiangsu, China
3
Telfer School of Management, University of Ottawa, Ontario, Canada

ABSTRACT
We introduce a versatile and robust model that may help policymakers, bond portfolio managers and financial
institutions to gain insight into the future shape of the yield curve. The Burg model forecasts a 20-day yield curve,
which fits a pth-order autoregressive (AR) model to the input signal by minimizing (least squares) the forward and
backward prediction errors while constraining the autoregressive parameters to satisfy the Levinson–Durbin recursion.
Then, it uses an infinite impulse response prediction error filter. Results are striking when the Burg model is compared
to the Diebold and Li model: the model not only significantly improves accuracy, but also its forecast yield curves stick
to the shape of observed yield curves, whether normal, humped, flat or inverted. Copyright © 2016 John Wiley &
Sons, Ltd.

key words Burg model; forecasting; yield curve

INTRODUCTION

Yield curve forecasting may be useful for building economic forecasts, pricing interest rate and credit derivatives,
appraising the potential impact of interest rate changes on the default probabilities of financial instruments, managing
duration or computing value at risk of a bond portfolio. Various forecasting models such as autoregressive (Diebold
and Li, 2006), multivariate nonparametric (Audrino and Trojani, 2007) or factor-augmented vector autoregression
(Füss and Nikitina, 2011) have been proposed to capture the future shape of the yield curve but none of these applies
the signal processing theory to yield curve forecasting. Signal processing has applications in many fields such as
electrical signals, audio signal processing, wireless communication, waveform generation, demodulation, filtering,
equalization and seismology. We propose to extend its application fields to yield curve forecasting. In the methodol-
ogy section, we present the reasons behind the choice of signal processing in yield curve forecasting. The results
section emphasizes the rationale of selecting signal processing. The next section reviews the literature of yield curve
forecasting. The third section presents the methodology in three steps. In the fourth section we report our main results
before concluding in the fifth section.

LITERATURE REVIEW

The term structure of interest rates was initially estimated by spline-based methods pioneered by McCulloch (1971)
and extended by Vasicek and Fong (1982); Coleman et al. (1992); Adams and Van Deventer (1994) and Fisher et al.
(1995). McCulloch (1971) used quadratic splines estimated by linear regression. Later, McCulloch (1975) used cubic
splines. Shea (1984, 1985) identified two limitations of McCulloch’s (1975) model since forward rates can become
negative and the resulting yield function tends to bend sharply towards the end of the maturity range observed in the
sample. The second class of estimation methods of the term structure is the parsimonious parametrization of the yield
curve initiated by Nelson and Siegel (1987). They developed a three-factor function to fit the yield curve where fac-
tors are interpreted as the long-run levels of interest rates, the short-term component the medium-term component, the
last two factors inputting a decay time factor. Svensson (1995) estimated forward rates mainly using the original
Nelson and Siegel (1987) model, in some cases offering an extended version. Dolan (1999) captured the curvature
parameter of the yield curve, estimated with the Nelson and Siegel (1987) model and predicted with simple parsimo-
nious models. Litterman and Scheinkman (1991); Chen and Scott (1993); Andersen et al. (1997); Bliss (1997); Dai
and Singleton (2000); De Jong (2000); Hong (2001) and Duffee (2002) used new assumptions to explain the evolu-
tion of the term structure of interest rates driven by dynamics of several factors such as macroeconomic shocks or
related to the level of interest rates, slope of the term structure, curvature and volatility of the changes. They

*Correspondence to: Pierre Rostan, Department of Management, American University in Cairo, AUC Avenue, PO Box 74, New Cairo 11835,
Egypt. E-mail: [email protected]

Copyright © 2016 John Wiley & Sons, Ltd.

92 P. Rostan et al.

developed multifactor models. Fabozzi et al. (2005) tested for statistical significance in the predictive power of the
Nelson and Siegel (1987) model. Bernadell et al. (2005) and Diebold and Li (2003) revisited an early version of
Nelson and Siegel (1987) by adding a regime-switching expansion. Diebold and Li (2006) applied the Nelson and
Siegel (1987) model to forecasting the yield curve with an autoregressive model. Théoret et al. (2006) forced simu-
lated interest rates inside Bollinger bands. Audrino and Trojani (2007) proposed a multivariate nonparametric tech-
nique, based on a functional gradient descent (FGD) estimation of the conditional mean vector and covariance matrix
of a multivariate interest rate series. Vicente and Tabak (2008), after testing the predictive ability of a variety of
models, compared affine term-structure models with the Diebold and Li (2006) model and suggested that forecasts
made with the Diebold and Li model are superior, and appear to be more accurate at long horizons than other different
benchmark forecasts. Leite et al. (2010) proposed a statistical model using data from a market survey and the forward
rate risk premium. Füss and Nikitina (2011) dusted off yield curve dynamics in terms of the unobservable compo-
nents level, slope and curvature, and applied the factor-augmented vector autoregression (FAVAR) framework for
forecasting interest rates. Christensen et al. (2011) derived the class of affine arbitrage-free dynamic term-structure
models that approximate the Nelson–Siegel yield curve specification. Rostan and Rostan (2012) forced simulated in-
terest rates inside bands built from the shapes of the most recent yield curves. Rostan and Rostan (2014) proposed a
framework to find the optimal distribution of innovation terms to forecast the yield curve. Caldeira et al. (2014) com-
bined forecasts from univariate and multivariate autoregressive specifications including dynamic factor models,
equilibrium term-structure models and forward rate regression models. Nowadays, the modeling of the term structure
focuses on two classes of models: the Nelson–Siegel models (NSMs) and affine term-structure models (ATSMs).
Among proponents of ATSMs, we may mention Diebold et al. (2005); Van Deventer et al. (2005); Baz and Chacko
(2004) and Bolder (2001). The ATSMs depend on arbitrage-free opportunities and assume that unobservable factors
underlying the term structure follow stochastic processes. The first step of ATSMs specifies the instantaneous interest
rate as a linear combination of a set of state variables; the second step describes the evolution of the factor processes,
typically as stochastic differential equations. The NSM models (Gasha et al., 2010) presuppose a specific form of the
term structure and are arbitrage-free. These models integrate both unobservable factors and observable macroeco-
nomic factors. As proponents of NSMs, Haustsch and Yang (2012) developed and applied Bayesian inference for
an extended Nelson–Siegel term-structure model capturing interest rate risk. Exterkate et al. (2013) compared various
ways of extracting macroeconomic information from a data-rich environment for forecasting the yield curve using the
Nelson–Siegel model. Diebold and Rudebusch (2013) proposed two extensions of the Nelson and Siegel (1987)
model, the first being the dynamic Nelson–Siegel model (DNS) and the second the DNS that they make arbitrage-
free. In the following methodology section, we apply the Burg (1975) model to yield curve forecasting, an original
model neither NSM nor ATSM.

METHODOLOGY

The US yield curve of Treasury bills, notes and bonds is divided into 10 constant maturities: 1-month, 3-month, 6-
month, 1-year, 2-year, 3-year, 5-year, 7-year, 10-year and 20-year. We forecast US Treasury yield curves 20 business
days ahead with the Burg model, which relies on signal processing theory. We compare our forecasts to yield curves
observed between 27 September 2002 and 21 May 2015, i.e. 3165 daily forecasts. Our choice of selecting a 20-
business day forecast horizon, equivalent to a 1-month calendar day horizon, is based on the following reasons: first,
our assumption is that the shorter the forecasting horizon, the more accurate the forecast. Second, if an economist
looks at predicting recessions by tracking the level of the 10-year to 3-month Treasury spread, it is optimal to interpret
the signal on a short-term horizon when the trend starts to build up. Third, in terms of monetary policy forecasts,
economists regularly based their forecasts on the futures price level, which incorporates the market expectations of
change in bank rate before monetary policy committee meetings. For this purpose, they use in the USA the
30-Day Federal Funds Futures or in Canada the 30-Day Overnight Repo Rate Futures. Instead, they may use the Burg
model promoted in this paper. Fourth, bond portfolio managers will shorten or lengthen their portfolio duration on a
short-term basis depending on their short-term expectations of interest rate changes. We benchmark our model to the
Diebold and Li (2006) model presented at step 3. We explain our intuition of applying signal processing theory to
yield curve forecasting by the fact that interest rates move through time in waveforms. Figure 1 illustrates the US
yield curve split in times series of interest rates of 10 different maturities from 1-month to 20-year. We observe that
the times series behave like signals, specifically waveforms. As acknowledged in the literature, interest rates follow a
mean-reverting process modeled, for example, by Vasicek (1977) or Cox et al. (1985). In communication systems,
signal processing and electrical engineering, a signal refers to ‘a function that conveys information about the behavior
or attributes of some phenomenon’ (Priemer, 1991). In electrical engineering, the embodiment of a signal in electrical
form is made by a transducer that converts the signal from its original form to a waveform expressed as a current or a
voltage, or an electromagnetic waveform: for example, an optical signal or radio transmission. Figure 1 illustrates
how interest-rate time series propagate in waveforms. In signal processing, wave propagation is how waves travel.

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
 Yield Curve Forecasting with the Burg Model 93

0.06

0.05

0.04

0.03
20-year
10-year
7-year
0.02 5-year
3-year
2-year
0.01 1-year
6-month
3-month
0 1-month
7/31/2001 4/26/2004 1/21/2007 10/17/2009 7/13/2012 4/9/2015

Figure 1. The 10 daily yields of US government securities (constant maturity from 1-month to 20-year maturity) constituting the
US yield curve from 31 July 2001 to 21 May 2015. Source: H.15 page, Federal Reserve website

Our model, applied previously to population projections (Rostan et al., 2015), is able to capture the amplitude,
frequency, and trend of increasing or decreasing amplitude of the waves represented by times series during their prop-
agation through time. Our research assumption is therefore that interest-rate time series propagate through time like
signals through space.
We explain below the methodology in three steps.

Step 1: fitting the time series with the Burg model

For each of the 10 maturities, we gather 269 historical daily yields from 31 July 31 2001 to 28 August 2002.
Given x the times series of 269 observations per maturity, we generate a vector a of all-pole filter coefficients that
model an input data sequence using the Levinson–Durbin algorithm (Levinson, 1946; Durbin, 1960). We use the
Burg method (Burg, 1975) to fit a pth-order autoregressive (AR) model to the input signal, x, by minimizing (least
squares) the forward and backward prediction errors while constraining the AR parameters to satisfy the
Levinson–Durbin recursion. x is assumed to be the output of an AR system driven by white noise. Vector a contains
the normalized estimate of the AR system parameters, A(z), in descending powers of z:
pffiffiffi pffiffiffi
e e
H ðzÞ ¼ ¼ (1)
AðzÞ 1 þ a2 z1 þ … þ aðpþ1Þ zp

where H(z) is the optimal linear predictor.

Since the method characterizes the input data using an all-pole model, the correct choice of the model order p is
important. In order to determine the optimal pth-order AR coefficient for the database, we apply the Burg method to
fit a pth-order AR model to the input signal, which is represented by 249 observations from 31 July 2001 to 31 July
2002. We then forecast 20 interest rates from 1 August 2002 to 28 August 2002 for the 10 maturities. We choose the
model order p that minimizes the root mean square error (RMSE) defined in equation 2 between the observed values
yi and the values forecast by the model ŷi for the forecast window of 20 observations times 10 maturities:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 n
RMSE ¼ ∑ ðy  ^y i Þ2 (2)
n i¼1 i

Figure 2 illustrates the model order p versus the RMSE. When p = 2, the RMSE value is minimized. We use
Matlab to implement and test the Burg model.

Step 2: forecasting the time series of US yield curve from 27 September 2002 to 21 May 2015
At step 1, we have seen that, in order to evaluate the model order p, we need to forecast a 20-day sample.
Minimization of the forecasting error helps determine p. At step 2, we explain the forecasting method. We run

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
94 P. Rostan et al.

Figure 2. Model order p of the Burg model versus RMSE of 20 forecasting values between 1 August 2002 and 28 August 2002
for 10 maturities based on 249 past observations

Figure 3. Prediction error filter used for forecasting the US yield curve from 27 September 2002 to 21 May 2015. H(z) and A(z)
are defined in equation 1

the initial interpolated time series x from 31 July 2001 to 29 August 2002 (270 data per maturity) through the filter
to obtain the filter state. In Figure 3 the prediction error, e(n), can be viewed as the output of the prediction error filter
A(z), where H(z) defined by equation 1 is the optimal linear predictor, x(n) is the input signal and ^x (n) is the
predicted signal.
Finally, we use the infinite impulse response (IIR) filter to extrapolate the 20-day interest rate forecast per maturity
from 27 September 2002 until 21 May 2015, i.e. 3165 daily forecasts. IIR filters are digital filters with infinite impulse
response. Unlike finite impulse response (FIR) filters, IIR filters have a feedback (a recursive part of a filter) and are
therefore known as recursive digital filters.
We benchmark the Burg model to the Diebold and Li (2006) model presented in step 3.

Step 3: Diebold and Li (2006) model

Diebold and Li (2006) used the Nelson and Siegel (1987) three-factor model to fit the yield curve yt (τ):
    
yt ðτ Þ ¼ β1t þ β2t 1 –eλ t τ =ðλt τ Þ þ β3t 1 –eλ t τ =ðλt τ Þ –eλ t τ (3)

The inputs are the daily US Treasury yield curve. The value of λt = 0.91 for all t is obtained by minimizing the
RMSE over 249 days between 31 July 2001 and 31 July 2002. The minimization method involves equations 3, 4
and 5.

0.00135

0.0013

0.00125

0.0012
RMSE

0.00115

0.0011
Average RMSE reaches a
0.00105 minimum when λ = 0.91
0.001
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
Lambda

Figure 4. Setting λ = 0.91 in the Nelson and Siegel (1987) model by minimizing the average RMSE between 31 July 2001 and 31
July 2002

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
 Yield Curve Forecasting with the Burg Model 95

Table I. Types of US yield curve (normal, humped, flat, inverted) between 31 July 2001 and 21 May 2015
Type of curve Normal Humped Flat Inverted Total
Number of observations 3,047 104 168 135 3454
Percentage of observations 88 3 5 4 100

Table II. Criteria used for the classification of US yield curves into four types: inverted, flat, humped and normal
Type of curve Inverted Flat Humped Normal
Criteria 1-month rate is Else: All rates remain in a Else: 6-month rate is higher Else:
higher than 20-year range of 50 basis points than 5-year rate
rate

Figure 4 illustrates the optimal choice of λ that minimizes the average RMSE over 249 days. Nelson and Siegel
applied a nonlinear least squares method to daily observed yield curves and obtained a time series of estimates of
{β1t, β2t, β3t}. Because the yield curve depends only on the three estimated factors {β1t, β2t, β3t}, forecasting the yield
curve is equivalent to forecasting {β1t, β2t, β3t}. Diebold and Li (2006) forecast the Nelson and Siegel factors as a
univariate AR(1) process. Diebold and Li produced yield forecasts based on an underlying univariate AR(1) specifi-
cation, as
    
ytþh=t ðτ Þ ¼ β1;tþh=t þ β2;tþh=t 1 –eλτ =ðλτ Þ þ β3;tþh=t 1 –eλτ =ðλτ Þ –eλτ (4)

where
βi;tþh=t ¼ C i þ ωi βit (5)

Ci and ωi are obtained by regressing βit on an intercept and βi,th.

Our forecasting horizon is h = 20 days. We regress the first set of β1, β2, β3 from 1 to 250 days with the set of 250
{β1t, β2t, β3t} obtained between 20 and 270 days, then moving forward 1 day at a time. The resulting {β1t, β2t, β3t} are
then plugged into equation 5 to compute the 20-day forecast interest rate for each maturity of the term structure.

Database
The database includes market yields of US Treasury securities (bills and notes) with 1-, 3- and 6-month, and 1-, 2-, 3-,
5-, 7-, 10- and 20-year maturity, quoted on investment basis yields on actively traded non-inflation-indexed issues
adjusted to constant maturities. The US yield curves of 3454 days, extending from 31 July 2001 to 21 May 2015,
are obtained from the Federal Reserve website, page H15. We discarded the 30-year maturity since it was
discontinued on 18 February 2002 and reintroduced only on 9 February 2006. We divide the database into four sub-
samples: normal, humped, flat and inverted yield curves. We present the statistics regarding the four types of yield
curves in Table 1, using the criteria presented in Table 2.

4.5

3.5

2.5

1.5

0.5
7/31/2001 4/26/2004 1/21/2007 10/17/2009 7/13/2012 4/9/2015

Figure 5. Types of US yield curves between 31 July 2001 and 21 May 2015: humped, flat and inverted curves

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
96 P. Rostan et al.

Table III. RMSE per type of US yield curve with the Burg model versus the Diebold and Li model between 27 September 2002
and 21 May 2015 (3165 days)
Type of US yield curve/frequency Normal 88% Humped 3% Flat 5% Inverted 4% All types 100%
RMSE with Burg model 0.0019187 0.0023998 0.0017206 0.001606 0.0019107
RMSE with Diebold and Li model 0.0029756 0.0030833 0.0019893 0.0019395 0.0028826
Improvement in % with Burg model 36% 22% 14% 17% 34%

0.03

0.025

0.02

0.015

0.01 Normal observed

0.005 Normal forecasted with the Burg model

Normal forecasted with the Diebold and Li model

0
0 2 4 6 8 10 12 14 16 18 20
-0.005

Figure 6. Observed US Normal yield curve on 9 April 2012 versus 20-day forecasts with the Burg and the Diebold and Li models

0.054
0.053
0.052
0.051
0.05
0.049
0.048 Humped observed
0.047 Humped forecasted with the Burg model
0.046
Humped forecasted with the Diebold and Li model
0.045
0 2 4 6 8 10 12 14 16 18 20

Figure 7. Observed US Humped yield curve on 23 May 2006 versus 20-day forecasts with the Burg and the Diebold and Li
models

Figure 5 illustrates the repartition of the four types of yield curve—normal, humped, flat and inverted—over the
period from 31 July 2001 to 21 May 2015.

RESULTS

We compute 20-day forecasts of the US yield curve with the Burg and the Diebold and Li models, which we compare
to the observed yield curve between 27 September 2002 and 21 May 2015 (3165 days). We identify four types of
yield curve: normal, humped, flat and inverted. Based on the RMSE criteria (equation 2), the Burg model outperforms
the Diebold and Li method. Whether by type of yield curve or for the whole sample of 3165 days, the Burg model
improves the forecasting ability of the Diebold and Li model by 34% overall, 36% for normal yield curves, 22%
for humped, 14% for flat and 17% for inverted yield curves, as presented in Table 3.
We provide illustrations of yield curve forecasted with the Burg and the Diebold and Li models. They demonstrate
the ability of the Burg model to mimic the shape of forecast yield curves. In comparison, the Diebold and Li model
roughly captures the trend of the observed yield curve without sticking to it.
A first example is the observed US Normal yield curve on 9 April 2012 versus the 20-day forecasts with the Burg
and the Diebold and Li models (Figure 6).
A second example is the observed US Humped yield curve on 23 May 2006 versus the 20-day forecasts with the
Burg and the Diebold and Li models (Figure 7).

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
 Yield Curve Forecasting with the Burg Model 97

0.053
Flat observed
0.052
Flat forecasted with the Burg model

0.051 Flat forecasted with the Diebold and Li model

0.05

0.049

0.048

0.047

0.046
0 2 4 6 8 10 12 14 16 18 20

Figure 8. Observed US Flat yield curve on 9 October 2006 versus 20-day forecasts with the Burg and the Diebold and Li models

0.054
Inverted observed
0.053
Inverted forecasted with the Burg model
0.052

0.051 Inverted forecasted with the Diebold and Li model

0.05

0.049

0.048

0.047

0.046

0.045

0.044
0 2 4 6 8 10 12 14 16 18 20

Figure 9. Observed US Inverted yield curve on 29 November 2006 versus 20-day forecasts with the Burg and the Diebold and Li
models

A third example is the observed US Flat yield curve on 9 October 2006 versus the 20-day forecasts with the Burg
and the Diebold and Li models. We define by ‘flat’ a yield curve whose yields remain in a range of 50 basis
points (Figure 8).
A fourth example is the observed US Inverted yield curve on 29 November 2006 versus the 20-day forecasts with
the Burg and the Diebold and Li models (Figure 9).

CONCLUSION

One key issue in economics and finance is to forecast the yield curve with accuracy. This paper presents a versatile
and robust model that may help policymakers, bond portfolio managers and financial institutions to gain insight into
the future shape of the yield curve. The model forecasts a 20-day yield curve with the Burg model that fits a pth-order
AR model to the input signal by minimizing (least squares) the forward and backward prediction errors while
constraining the autoregressive parameters to satisfy the Levinson–Durbin recursion; it then uses an infinite impulse
response prediction error filter. Results are striking when the Burg model is benchmarked to the Diebold and Li
(2006) model: the model not only significantly improves accuracy, but its forecast yield curves also stick to the shape
of observed yield curves, whether normal, humped, flat or inverted.
Further research may extend the forecasting horizon or use sophisticated and more recent models as benchmarks.

REFERENCES

Adams KJ, Van Deventer DR. 1994. Fitting yield curves and forward rate curves with maximum smoothness. Journal of Fixed Income
4: 52–62.
Andersen A, Lund J, Torben G. 1997. Estimating continuous-time stochastic volatility models of the short-term interest rate. Journal
of Econometrics 77: 343–377.
Audrino F, Trojani F. 2007. Accurate short-term yield curve forecasting using functional gradient descent. Journal of Financial
Econometrics 5: 591–623.
Baz J, Chacko G. 2004. Financial Derivatives: Pricing, Applications and Mathematics. Cambridge University Press: Cambridge, UK.

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
98 P. Rostan et al.

Bernadell C, Coche J, Nyholm K. 2005. Yield curve prediction for the strategic investor. Working paper No. 4722005, European Cen-
tral Bank.
Bliss R. 1997. Movements in the term structure of interest rates. Economic Review, Federal Reserve Bank of Atlanta 4: 16–33.
Bolder DJ. 2001. Affine term-structure models: theory and implementation. Working paper 2001-15, Bank of Canada, Ottawa.
Burg JP. 1975. Maximum entropy spectral analysis. Available: http://sepwww.stanford.edu/data/media/public/oldreports/sep06/
[accessed on January 2016].
Caldeira JF, Moura GV, Santos AAP. 2014. Predicting the yield curve using forecast combinations. Computational Statistics and Data
Analysis in press.
Chen R, Scott L. 1993. Maximum likelihood estimation for a multifactor equilibrium model of the term structure of interest rates. Jour-
nal of Fixed Income 3: 14–31.
Christensen JHE, Diebold FX, Rudebusch GD. 2011. The affine arbitrage-free class of Nelson–Siegel term structure models. Journal
of Econometrics 164: 4–20.
Coleman TS, Fisher L, Ibbotson R. 1992. Estimating the term structure of interest rates from data that include the prices of coupon
bonds. Journal of Fixed Income 2: 85–116.
Cox JC, Ingersoll JE, Ross SA. 1985. The theory of the term structure of interest rates. Econometrica 539: 363–384.
Dai Q, Singleton K. 2000. Specification analysis of affine term structure models. Journal of Finance 55: 1943–1978.
De Jong F. 2000. Time series and cross-section information in affine term-structure models. Journal of Business and Economic Sta-
tistics 18: 300–314.
Diebold F, Li C. 2003. Forecasting the term structure of government bond yields. Working paper No. 10048, National Bureau of Eco-
nomic Research.
Diebold F, Li C. 2006. Forecasting the term structure of government bond yields. Journal of Econometrics 130: 337–364.
Diebold F, Rudebusch GD. 2013. The Dynamic Nelson–Siegel Approach to Yield Curve Modeling and Forecasting. Princeton Uni-
versity Press: Princeton, NJ.
Diebold FX, Piazzesi M, Rudebusch GD. 2005. Modeling bond yields in finance and macroeconomics. American Economic Review:
Papers and Proceedings 95: 415–420.
Dolan C. 1999. Forecasting the yield curve shape: evidence from global markets. Journal of Fixed Income 10: 40–53.
Duffee GR. 2002. Term premia and interest rate forecasts in affine models. Journal of Finance 57: 405–443.
Durbin J. 1960. The fitting of time series models. Revue de l’Institut International de Statistique 28: 233–244.
Exterkate P, Van Dijk D, Heij C, Groenen PJF. 2013. Forecasting the yield curve in a data-rich environment using the factor-
augmented Nelson–Siegel model. Journal of Forecasting 32: 193–214.
Fabozzi J, Martellini L, Priaulet P. 2005. Predictability in the shape of the term structure of interest rates. Journal of Fixed Income 15:
40–53.
Fisher M, Nychka D, Zervos D. 1995. Fitting the term structure of interest rates with smoothing splines. Federal Reserve System
Working Paper 95-1.
Füss R, Nikitina O. 2011. Explaining yield curve dynamics. Journal of Fixed Income 21: 68–87.
Gasha G, He Y, Medeiros C, Rodriguez M, Salvati J, Yi J. 2010. On the estimation of term structure models and an application to the
United States. IMF Working Paper WP/10/258. Available: https://www.imf.org/external/pubs/ft/wp/2010/wp10258.pdf [accessed
on January 2016].
Haustsch N, Yang F. 2012. Bayesian inference in a stochastic volatility Nelson-Siegel model. Computational Statistics and Data Anal-
ysis 56: 3774–3792.
Hong Y. 2001. Dynamic models of the term structure. Financial Analysts Journal 57: 60–76.
Leite AL, Filho RBPG, Vicente JVM. 2010. Forecasting the yield curve: a statistical model with market survey data. International
Review of Financial Analysis 19: 108–112.
Levinson N. 1946. The Wiener RMS error criterion in filter design and prediction. Journal of Mathematical Physics 25: 261–78.
Litterman R, Scheinkman J. 1991. Common factors affecting bond returns. Journal of Fixed Income 1: 54–61.
McCulloch JH. 1971. Measuring the term structure of interest rates. Journal of Business 44: 19–31.
McCulloch JH. 1975. The tax adjusted yield curve. Journal of Finance 30: 811–830.
Nelson CR, Siegel AF. 1987. Parsimonious modeling of yield curves. Journal of Business 60: 473–489.
Priemer R. 1991. Introductory Signal Processing. World Scientific: Singapore.
Rostan P, Rostan A. 2012. Forecasting the yield curve with the stock dog technique. International Journal of Business and Manage-
ment 7: 31–43.
Rostan P, Rostan A. 2014. Fitting the Pareto-Lévy distribution on the yield curve: an application to forecasting. IEB International
Journal of Finance 8: 38–66.
Rostan P, Belhachemi R, Rostan A. 2015. Appraising the financial sustainability of the pension system with signal processing. In Sus-
tainability and Sufficiency of the Pension Systems: PAYG vs. Capitalization 33-3 Estudios De Economía Aplicada. ASEPELT:
Zaragoza, Spain; 801–816.
Shea G. 1984. Pitfalls in smoothing interest rate term structure data: equilibrium models and spline approximations. Journal of Finan-
cial and Quantitative Analysis 19: 253–269.
Shea G. 1985. Interest rate term structure estimation with exponential splines: a note. Journal of Finance 11: 319–325.
Svensson LEO. 1995. Estimating forward interest rates with the extended Nelson and Siegel method. Sveriges Riskbank Quarterly Re-
view 3: 13–26.
Théoret R, Rostan P, El Moussadek A. 2006. An essay on stochastic volatility and the yield curve. In Advances in Risk Management,
Gregoriou G (ed). Palgrave-Macmillan: London; 86–106.
Van Deventer DR, Imai K, Mesler M. 2005. Advanced Financial Risk Management. Wiley: Singapore.
Vasicek OA. 1977. An equilibrium characterization of the term structure. Journal of Financial Economics 5: 177–188.
Vasicek OA, Fong HC. 1982. Term structure modeling using exponential splines. Journal of Finance 37: 339–348.
Vicente J, Tabak B. 2008. Forecasting bond yields in the Brazilian fixed income market. International Journal of Forecasting 24:
490–497.

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
 Yield Curve Forecasting with the Burg Model 99

Authors’ biographies:
Pierre Rostan worked for more than 12 years in industry, having developed expertise in derivatives and risk management. He has
been New Products Manager in the R&D Department of the Montreal Exchange, Canada; Risk Consultant at APT Consulting Paris
and Analyst at Clearstream Luxembourg. He holds a PhD in Administration from the University of Quebec, Canada. His major areas
of research are numerical methods applied to derivatives and yield curve forecasting. As Associate Professor at the American Univer-
sity in Cairo, Egypt, he teaches courses in the area of financial markets.

Rachid Belhachemi holds a PhD in Mathematics from Oklahoma State University Stillwater. He is teaching in the Master program of
Financial Mathematics at Xi’an Jiaotong-Liverpool University, China. His teaching experience spans a broad spectrum of courses in
mathematical sciences, which include courses in financial mathematics, actuarial mathematics and statistics. His research interests and
recent publications are in statistics, pension modeling, asset pricing and risk management.

François-Eric PhD, is Associate Professor of Finance at the Telfer School of Management, University of Ottawa. His research inter-
ests focus on the problems of measurement errors, specification errors and endogeneity in financial models of returns. He is also in-
terested in developing new methods for forecasting financial time series, especially with regard to hedge fund risk. He has
published several books and many articles on quantitative finance and financial econometrics.

Authors’ addresses:
Pierre Rostan, Department of Management, American University in Cairo, Egypt.

Rachid Belhachemi, Department of Mathematical Science, Xi’an Jiaotong-Liverpool University, Jiangsu, China.

François-Eric Racicot, Telfer School of Management, University of Ottawa, Ontario, Canada.

Copyright © 2016 John Wiley & Sons, Ltd. J. Forecast. 36, 91–99 (2017)
Copyright of Journal of Forecasting is the property of John Wiley & Sons, Inc. and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.