![MathJax Logo](/templates/jsp/_style2/_tandf/pb2/images/math-jax.gif)
Abstract
Encompassing test has been well developed for fully parametric modeling. In this study, we are interested on encompassing test for parametric and nonparametric regression methods. We consider linear regression for parametric modeling and nearest neighbor regression for nonparametric methods. We establish asymptotic normality of encompassing statistic associated to the encompassing hypotheses for the linear parametric method and the nonparametric nearest neighbor regression estimate. We also obtain convergence rate depending only on the number of neighbors while it depends on the number of observation
and the bandwidth
for kernel method. We achieve the same convergence rate when
. Moreover, asymptotic variance of the encompassing statistic associated to kernel regression depends on the density, this is not the case for nearest neighbor regression estimate.
PUBLIC INTEREST STATEMENT
Regression techniques are used for quantitative analysis method in many fields such as in economic or financial modeling. They are a useful tool for identification of factors, which may explain the evolution of any variable of interest. In economic modeling for example, when we want to analyze the evolution of Gross Domestic Product or GDP, it might be affected by many variables like interest rate, inflation, exchange rate, sentiment indicators. Researchers or experts may face several admissible models from parametric or/and nonparametric regression methods. Encompassing test can be helpful for detection of redundant models among admissible models. The findings in this study contribute on encompassing test between linear and nearest neighbor regression estimates.
1. Introduction
Encompassing tests lie on model selection step. They are used for detection of redundant models among admissible models. In that case an encompassing model is intended to account for the results found by encompassed model. Theoretical development on encompassing test can be found in Mizon (Citation1984), Gouriéroux and Monfort (Citation1995) and Florens et al. (Citation1996). For their application, we refer readers to the general to specific computer based model selection procedure, Hendry and Doornik (Citation1994).
Recently, Bontemps et al. (Citation2008) have developed encompassing test for linear parametric against kernel nonparametric regression methods. They provide asymptotic normality of the associated encompassing statistics under the independent and identically distributed hypothesis (i.i.d.). As stated in Hendry et al. (Citation2008) that the work of Bontemps et al. (Citation2008) is the starting treatment of encompassing tests to functional parameter based on nonparametric methods.
We extend this result to nearest neighbor regression method, which has been claimed more flexible compared to kernel. Other motivation would be its interest in application like in Nowman and Saltoglu (Citation2003), Guégan and Huck (Citation2005), Ferrara et al. (Citation2010), Guégan and Rakotomarolahy (Citation2010), and Puspitasari and Rustam (Citation2018), among others.
In the next section, we provide an overview of the encompassing test. After, we establish asymptotic normality for various encompassing statistics associated to linear parametric and nearest neighbor regression methods. Last, we conclude.
2. Encompassing test for independent processes
This section introduces the encompassing test and then builds the corresponding encompassing hypothesis. So, given two regression models and
, we are interested in knowing if model
can account the result of model
. In fact, we want to know if
encompasses
or, in a short notation
. Testing such a hypothesis will be done using the notion of encompassing test.
Generally speaking, model encompasses model
, if the parameter
of the latter model can be expressed in function of the parameter
of the former model. In other words, let
be the pseudo-true value of
on
. In general, the pseudo-true value is defined as the plim of
on
. For more discussion on pseudo-true value associated with the KLIC,Footnote1 we refer to Sawa (Citation1978) and Govaerts et al. (Citation1994). The encompassing statistic is given by the difference between
and
scaled by a coefficient
.
Let be a zero mean random process with valued in
x
x
where
. For
and
, we consider the two models
and
defined as follows:
In addition, the general unrestricted model is given by . Following the encompassing test for functional parameter in Bontemps et al. (Citation2008), we have the null hypothesis:
This null states that is the owner model, and
will be served on validating this statement and is called the rival model. We test this hypothesis
through the following implicit encompassing hypothesis:
The following homoskedasticity condition will be assumed all along this work:
Moreover, a necessary condition for the encompassing test relies on the errors of both models where the intended encompassing model should have smaller standard error than the encompassed model
.
Given a sample, of size ,
for
as realization of the random process
. We suppose that
,
are
. Then, for given functional estimates
and
of the functions
and
, respectively, we have the following encompassing statistic:
where is an estimate of the pseudo-true value, associated with
on
, in the LHS of the hypothesis
. Bontemps et al. (Citation2008) has provided asymptotic normality of this encompassing statistic
by considering kernel regression estimate for nonparametric method. This result can be extended to nearest neighbor regression estimate but of course with different assumptions.
For nearest neighbor regression estimate, we consider the representation in Mack (Citation1981), that is the nearest neighbor (or
-NN) estimate
of
is given by:
where will be defined as distance, according to the Euclidean norm in
, from
to its
neighbors, and
is a bounded, non-negative weight function satisfying
To establish an asymptotic distribution of , we need some assumptions. The following assumptions will be used for insuring the asymptotic normality and are taken from Mack (Citation1981). Without loss of generality, the function
will be a marginal density or a conditional density or a joint density according to the variables on its arguments.
The first assumption relies on the density function of the couple .
Assumption 1. The function is bounded and continuous at
for
and continuously differentiable in a neighborhood of
for
.
The following assumption concerns conditions on the moments up to order three of the variable of interest.
Assumption 2. ,
and
.
The last assumption states conditions on the relationship between the number of neighbors and the sample size
.
Assumption 3. with
.
When assumptions 1–3 hold and the relation (3) is satisfied, then Mack (Citation1981) has established the asymptotic normality of the centered -NN regression of
. Moreover, under assumption 3, the bias of such
-NN regression estimate vanishes to zero.
Without loss of generality, we proceed as previously when model will be estimated by
-NN regression method. In the rest of the paper,
denotes the normal distribution with mean
and variance
. We now present the asymptotic normality of the encompassing statistic.
3. Asymptotic normality of the encompassing statistic
In general, or
can be estimated using nonparametric or parametric regression methods. We can encounter the following four situations:
and
are both estimated parametrically,
and
are both estimated nonparametrically,
is estimated nonparametrically and
parametrically and
is estimated parametrically and
nonparametrically.
For development on the asymptotic behavior of the encompassing statistic for fully parametric case, i.e the two models and
have parametric specification, we refer readers to Gouriéroux et al. (Citation1983) and Mizon and Richard (Citation1986) among others. For recent discussion on this encompassing test for fully parametric case, Bontemps et al. (Citation2008) is a good reference.
Next, we will study the completely nonparametric case.
3.1. Nonparametric specification for ![](//:0)
and![](//:0)
![](//:0)
We consider the case where the two models and
defined in (1) are estimated using the nonparametric nearest neighbor regression method. To test the hypothesis”
encompasses
”, we establish asymptotic normality of the associated encompassing statistic.
Theorem 3.1. Assume that assumptions 1–3 and relations (2) and (3) hold. Then under , we have:
where for
are the residuals from model
and
is the volume of unit ball in
with
the gamma function.
Proof of Theorem 3.1
The proof will be based on the decomposition of the encompassing statistic into two parts as an expression of nearest neighbor regression and a kind of bias. Before all, let’s denote by:
We write down our encompassing statistic by replacing our estimates and
at a given point
, and we have:
where A is the first expression in RHS of the equality. This involves a -NN regression of
on
scaled by the coefficient
seeing as convergence speed rate when
goes to infinity. Using Mack (Citation1981), under assumptions 1–3 and when relation (3) holds, we have:
Next, for the second expression , we can bound by taking its supremum with respect to
and then we get:
When using the expression of the bias, Theorem 1 in [2], becomes:
where is a function depending only on
and its expression can be found in Mack (Citation1981). Then from Assumption 3,
vanishes to zero when
. It remains on showing that
goes to zero also. This can be achieved using result of Mukerjee (Citation1993) extension of Cheng’s work (Cheng, Citation1984). Therefore, we remark that when the number of neighbors
increases more the weights given to neighbors decrease, then rewriting
and we have the following equivalence:
where is a given weight function which satisfies condition (3),
is a bounded weight equal to zero when
is larger than the number of neighbors and
is the distance between
and its
neighbor. When we denote by
, then from Theorem 2.1 in Mukerjee (Citation1993), we have:
with and
a positive sequence which tends to zero as
. So we get
converges to zero in probability as
. This completes the proof of theorem.
Next, we will consider the mixed situation where the owner model has parametric specification and the rival is from nonparametric method.
3.2. Parametric modelling for ![](//:0)
vs nonparametric specification for![](//:0)
![](//:0)
In this section, we consider the case that model is a linear parametric model and
is estimated by nearest neighbor regression technique. Therefore, the hypothesis
will have linear parametric specification. The encompassing statistic associated to the null
can be rewritten as follows:
where is an estimate of the pseudo-true value
associated with
on
, and is defined as
.
We estimate the rival model using
-NN regression method where the owner model
is still with linear parametric specification. The following theorem provides the asymptotic normality of the encompassing statistic introduced in relation (5).
Theorem 3.2. Assume that assumptions 1–3, relations (2) and (3) hold.
Then under , we get:
where with
is the volume of unit ball in
.
Proof of Theorem 3.2.
When the owner model is the linear regression parametric and the rival model
is the
-NN regression, we write the encompassing statistic as follows:
where is the weight associated to the nearest neighbor regression of
on
and
is the distance from
to its
neighbor.
We remark that and
as fitted values of
would have the same
nearest to
. We then have
. Otherwise, this can happen asymptotically, that is
and
as fitted values of
have the same
nearest to
when
and
tend to infinity. Thus,
is asymptotically equivalent to
.
For the first expression , with
. Under assumptions in Theorem 3.2, then using result in Mack (Citation1981), we have:
where .
For , under assumptions in Theorem 3.2, we know that the estimate
converges in distribution to a normal law
with mean zero. The remaining part of
has the following expression
which converges in distribution to zero. Thus, from Slutsky’s theorem,
tends to zero in distribution.
We will consider the last case where the owner model is a nonparametric method and the rival model
is a linear parametric model.
3.3. Nonparametric specification for ![](//:0)
vs parametric modelling for ![](//:0)
![](//:0)
We now consider the owner model to be estimated using a
-NN nonparametric regression and the rival model
to be a linear parametric method. Therefore, the encompassing statistic associated to the null
is given by:
where is an estimate of the pseudo-true value
associated with
on
, which is defined by
. We estimate the unknown conditional mean
associated to the model
using
-NN regression estimate. We state in the following theorem the asymptotic normality of the encompassing statistic in relation (7). For precision, we use the assumptions introduced in previous section for
-NN regression estimate
instead of
Theorem 3.3. Assume that relations (2) and (3), assumptions 1–3, and the regularity conditions in linear regression are satisfied.
Then under , we get:
where .
Proof of Theorem 3.3.
When the functional parameters is from
-NN regression estimate, we rewrite the associated encompassing statistic as follows:
where corresponds to the first expression in the RHS of the equality (8). It coincides to the linear regression of the error
(with
) on
.
Under i.i.d. assumption in Theorem 3.3, converges in distribution to
where
is normally distributed with mean zero and variance
. For the second expression
, we bound it by taking the maximum with respect to
and then we get
where
and
. We remark that
is asymptotically equivalent to the bound of
in EquationEquation 4
(4)
(4) which converges to zero in probability. Thus, the product vanishes to zero also from Slutsky’s theorem. This completes the proof.
4. Illustration
In this section, we illustrate our theoretical results on real data. We focus on socio-economic factor determinants of Life expectancy. As explanatory variables for Life expectancy at birth, we consider the Gross National Income per capita in US , the Gross Domestic Product per capita in US
and the government health expenditure per capita in US
. Impact of these variables on Life expectancy at birth has been analyzed a long way in the literature, for regression analysis we may look at Hussain (Citation2002) and Ali and Ahmad (Citation2014). We use cross sectional data for 169 countries in 2017, which have been collected from the United Nation and the World Health Organization websites. To start our empirical study, we compute some basic statistics.
The highest life expectancy hits 84 years and belongs to Japan. While, the lowest is around 52 years belonging to Central African Republic. The best life expectancy of 84 years would be remarkable. Besides, life expectancy mean 72 years seems interesting. Moreover, the median value 73.69 indicates that around 84 countries have life expectancy above 73 years, largely beyond the retirement ages.
For socio-economic variables; Luxembourg, Switzerland and USA have the highest GDP, Income and health expenditure per capita, respectively. Burundi has the lowest GDP and Income per capita. Congo Democratic Republic registers the lowest government spending on health care. These variables exhibit some common behaviors such as the median of each variable is around fifteen times of its minimum and one over fifteen times of its maximum. They also have high dispersion. We now proceed on analysis of their relationship with life expectancy.
Let compute the correlation coefficients between life expectancy and the predictor variables.
From Table , Life expectancy has positive and high correlation with each explanatory variables. Such correlations indicate that higher GDP, income or expenditure on health will link with longer life expectancy. This preliminary analysis could motivate us on exploring other statistic and econometric analysis of the relationship between life expectancy and the three socio-economic variables. We will use the linear and the nearest neighbor regression methods. In sequel, we will work on demeaned and scaled (by a factor ) variables.
Table 1. Summary statistics
Table 2. Correlation between life expectancy and the socio-economic variables
For the linear regression, we explain life expectancy at birth by health expenditure per capita
, gross income per capita
and GDP per capita
. Considering several combination of these explanatory variables, following we summarize regression coefficient estimates with their standard errors in parenthesis.
where an estimate of the error term of model
,
.
Coefficients of models ,
and
are all significant. In contrast, models
and
nest to model
due to non-significance of
and
coefficient estimates. Besides,
nests to
as
’s coefficient estimate is not significant. We then focus our analysis on models
,
and
and proceed on their diagnostics. Results are reported in Table .
Table 3. Regression diagnostics
From Table , we accept the homoscedasticity property of residuals and their non-correlation with predictors. In addition, residuals of the three models have zero mean. Thus, our three models meet standard assumptions on linear regression.
,
and
are non-nested models. Thus, the decision on choosing one model will be based on encompassing test. A necessary condition is that the encompassing model should fit better than encompassed model. Therefore, encompassing model is expected to have smaller error variance than its rival. The standard errors of models
,
are
,
and
, respectively. Then, among the three models,
has the worst fit and
has the best fit. We report in Table various encompassing tests associated to models
,
and
.
Table 4. Encompassing tests for models ,
and
From Table , we accept the null and
that is,
encompasses
and
encompasses
. In contrast, we reject
and
, there are no mutual encompassing. Thus, we retain model
as it also has the smallest standard error. We will re-examine the link between life expectancy and explanatory variables using nearest neighbor regression.
For -NN regression of life expectancy, we need the specification of the weighting function
and the estimation of the parameter
. Two weighting functions have been mostly used in the literature: the exponential function
with
the
nearest to
, and the uniform function
. We also consider these two weighting functions.
Assumption 3 states that the number should satisfy
, for
observations and
explanatory variables. Then, as
, we have maximum values for
which are 60, 30, and 18 for
,
and
respectively. We estimate this parameter
by minimizing the root mean squared error (RMSE). Results are summarized in Table where we keep the following notation already used in linear regression:
for health expenditure per capita,
for gross income per capita and
for GDP per capita.
Table 5. Specification of -NN regression estimates
For models to
, model
has the lowest standard error. We also remark that standard errors of models
and
are very close. We will check if model
can account results of other models and if there are mutual encompassing between
and
. We now compute the following standardized encompassing statistics using result developed in Theorem 3.1:
where is the
-NN regression of the residuals
of owner model on explanatory variables
of rival model. Results are reported in Table .
Table 6. Encompassing tests for models to
Values in Table are all less than 1.96 in absolute value, except for . We accept null hypotheses
,
,
and
. In other word,
can account information content in other models. As
does not encompass
, there is no mutual encompassing. Thus, we can retain model
from all
-NN regression models.
Next illustration concerns encompassing test on nonparametric and parametric regression techniques in Theorem 3.3, having as null hypothesis: the nearest neighbor regression encompasses the linear regression
. Under this null, we have the following statistic from Theorem 3.3:
where is an estimate of the asymptotic variance
,
residuals of model
and
is a
-NN regression estimate of the conditional variance
.
Absolute value of the standardized encompassing statistic is less than
. Therefore, we accept the null hypothesis at a risk level
i.e the nearest neighbor regression
encompasses the linear regression
. We conclude that we may retain
-NN regression of life expectancy on health expenditure and income.
5. Conclusion
We know that different approaches of encompassing tests present in the literature provide different results. We have considered encompassing test in asymptotic way which is inline with the encompassing principle announced in the introduction. The work has been conducted for parametric and nonparametric regression techniques.
As stated in Hendry et al. (Citation2008) that the work of Bontemps et al. (Citation2008) is the starting treatment of encompassing tests to functional parameter based on nonparametric methods. We have extended that work to nearest neighbor functional parameter estimate under the i.i.d. assumption. When using linear and nearest neighbor regressions as estimators for conditional expectations, we have established asymptotic normality of the associated encompassing statistics for independent processes.
Comparing the convergence rate of the asymptotic encompassing statistic of -NN regression estimate to kernel regression obtained by Bontemps et al. (Citation2008), it depends only on the number of neighbors
for
-NN while for kernel ones depends on the number of observation
and the bandwidth
. We have the same convergence rate when
.
Moreover, Bontemps et al. (Citation2008) obtained asymptotic variance of the encompassing statistic associated to kernel regression depending on the density, which is not the case for nearest neighbor regression estimate.
Development of encompassing test to nonparametric methods opens new research direction in theory as well as in practice.
Acknowledgments
The author thanks the anonymous referees and the Editor Professor Hiroshi Shiraishi.
Additional information
Funding
Notes on contributors
Patrick Rakotomarolahy
Patrick Rakotomarolahy is the assistant professor in the department of mathematics and their applications at Fianarantsoa University. He has completed his Bsc and Msc in applied mathematics. He received his doctorate at the Panthéon-Sorbonne Paris 1 University. His current researches are in statistical model selection and in modeling macroeconomic and financial variables. He focuses especially on issues about model selection between parametric and nonparametric techniques. This study is in-line with this direction as the findings on asymptotic behavior of encompassing tests allow us to detect redundant models.
Notes
1. Kullback-Leibler Information Criterion
References
- Ali, A., & Ahmad, K. (2014). The impact of socio-economic factors on life expectancy in sultanate of Oman: An empirical analysis. Middle-East Journal of Scientific Research, 22(2), 218–12. https://www.idosi.org/mejsr/mejsr22(2)14/8.pdf
- Bontemps, C., Florens, J. P., & Richard, J. F. (2008). Parametric and non-parametric encompassing procedures. Oxford Bulletin of Economics and Statistics, 70(1), 751–780. https://doi.org/10.1111/j.1468-0084.2008.00529.x
- Cheng, P. E. (1984). Strong consistency of nearest neighbor regression function estimators. Journal of Multivariate Analysis, 15(1), 63–72. https://doi.org/10.1016/0047-259X(84)90067-8
- Ferrara, L., Guégan, D., & Rakotomarolahy, P. (2010). GDP nowcasting with ragged-edge data: A semi-parametric modeling. Journal of Forecasting, 29(1–2), 186–199. https://doi.org/10.1002/for.1159
- Florens, J. P., Hendry, D. F., & Richard, J. F. (1996). Encompassing and specificity. Econometric Theory, 12(4), 620–656. https://doi.org/10.1017/S0266466600006964
- Gouriéroux, C., & Monfort, A. (1995). Testing, encompassing, and simulating dynamic econometric models. Econometric Theory, 11(2), 195–228. https://doi.org/10.1017/S0266466600009142
- Gouriéroux, C., Monfort, A., & Trognon, A. (1983). Testing nested or non-nested hypotheses. Journal of Econometrics, 21(1), 83–115. https://doi.org/10.1016/0304-4076(83)90121-5
- Govaerts, B., Hendry, D. F., & Richard, J. F. (1994). Encompassing in stationary linear dynamic models. Journal of Econometrics, 63(1), 245–270. https://doi.org/10.1016/0304-4076(93)01567-6
- Guégan, D., & Huck, N. (2005). On the use of nearest neighbors in finance. Revue De Finance, 26(2), 67–86. https://www.cairn.info/revue-finance-2005-2-page-67.htm#
- Guégan, D., & Rakotomarolahy, P. (2010). A short note on the nowcasting and the forecasting of Euro-area GDP using non-parametric techniques. Economics Bulletin, 30(1), 508–518. http://www.accessecon.com/Pubs/EB/2010/Volume30/EB-10-V30-I1-P46.pdf
- Hendry, D. F., & Doornik, J. A. (1994). Modelling linear dynamic econometric systems. Scottish Journal of Political Economy, 41(1), 1–33. https://doi.org/10.1111/j.1467-9485.1994.tb01107.x
- Hendry, D. F., Marcellino, M., & Mizon, G. E. (2008). Encompassing. Oxford Bulletin of Economics and Statistics, Guest Editor Introduction. Special Issue
- Hussain, A. R. (2002). Life expectancy in developing countries: A cross-section analysis. The Bangladesh Development Studies, 28(1/2), 161–178.
- Mack, Y. P. (1981). Local Properties of k-NN Regression Estimates. SIAM Journal on Algebraic and Discrete Methods, 2(3), 311–323. https://doi.org/10.1137/0602035
- Mizon, G. E. (1984). The encompassing approach in econometrics. In D. F. Hendry & K. F. Wallis (Eds.), Econometrics and quantitative economics (pp. 135–172). Blackwell Publishers.
- Mizon, G. E., & Richard, J. F. (1986). The encompassing principle and its application to non-nested hypothesis tests. Econometrica, 54(3), 657–678. https://doi.org/10.2307/1911313
- Mukerjee, H. (1993). Nearest neighbor regression with heavy-tailed errors. Annals of Statistics, 21(2), 681–693. https://doi.org/10.1214/aos/1176349144
- Nowman, B., & Saltoglu, B. (2003). Continuous time and nonparametric modelling of U.S. interest rate models. International Review of Financial Analysis, 12(1), 25–34. https://doi.org/10.1016/S1057-5219(02)00123-0
- Puspitasari, D. A., & Rustam, Z. (2018). Application of SVM-KNN using SVR as feature selection on stock analysis for indonesia stock exchange. AIP Conference Proceedings 2023, Bali, Indonesia, 020207; https://doi.org/10.1063/1.5064204.
- Sawa, T. (1978). Information criteria for discriminating among alternative regression models. Econometrica, 46(6), 1273–1292. https://doi.org/10.2307/1913828