Split normal distribution

<h2 id="definition">Definition</h2>
The split normal distribution arises from merging two opposite halves of two <a href="/facts/Probability_density_function/zvfybna4">probability density functions</a> (PDFs) of <a href="/facts/Normal_distribution/UapjjPyQ">normal distributions</a> in their common <a href="/facts/Mode_(statistics)/vq5tGp0G">mode</a>.
The PDF of the split normal distribution is given by<a class="footnote-ref" id="fnref:7" href="#fn:7">7</a>

f
 (
 x
 ;
 μ
 ,
 
 σ
 
 1
 
 
 ,
 
 σ
 
 2
 
 
 )
 =
 
 
 {
 
 
 
 A
 exp
 ⁡
 
 (
 
 −
 
 
 
 
 (
 x
 −
 μ
 
 )
 
 2
 
 
 
 
 2
 
 σ
 
 1
 
 
 2
 
 
 
 
 
 
 
 )
 
 
 
 
 if 
 
 x
 <
 μ
 
 
 
 
 A
 exp
 ⁡
 
 (
 
 −
 
 
 
 
 (
 x
 −
 μ
 
 )
 
 2
 
 
 
 
 2
 
 σ
 
 2
 
 
 2
 
 
 
 
 
 
 
 )
 
 
 
 
 otherwise
 
 
 
 
 
 
 
 
 
 {\displaystyle f(x;\mu ,\sigma _{1},\sigma _{2})={\begin{cases}A\exp \left(-{\dfrac {(x-\mu )^{2}}{2\sigma _{1}^{2}}}\right)&{\text{if }}x<\mu \\[1ex]A\exp \left(-{\dfrac {(x-\mu )^{2}}{2\sigma _{2}^{2}}}\right)&{\text{otherwise}}\end{cases}}}

where

A
        =
        
          
            2
            
              /
            
            π
          
        
        (
        
          σ
          
            1
          
        
        +
        
          σ
          
            2
          
        
        
          )
          
            −
            1
          
        
        .
      
    
    {\displaystyle \quad A={\sqrt {2/\pi }}(\sigma _{1}+\sigma _{2})^{-1}.}

<h3>Discussion</h3>
The split normal distribution results from merging two halves of normal distributions. In a general case the 'parent' normal distributions can have different variances which implies that the joined PDF would not be <a href="/facts/Continuous_function/sKbl02pB">continuous</a>. To ensure that the resulting PDF <a href="/facts/Integral/V7lyV997">integrates</a> to 1, the <a href="/facts/Normalizing_constant/KKTfybqr">normalizing constant</a> A is used.
In a special case when 
 
 
 
 
 σ
 
 1
 
 
 2
 
 
 =
 
 σ
 
 2
 
 
 2
 
 
 =
 
 σ
 
 ∗
 
 
 2
 
 
 
 
 {\displaystyle \sigma _{1}^{2}=\sigma _{2}^{2}=\sigma _{*}^{2}}
 
 the split normal distribution reduces to <a href="/facts/Normal_distribution/UapjjPyQ">normal distribution</a> with variance 
 
 
 
 
 σ
 
 ∗
 
 
 2
 
 
 
 
 {\displaystyle \sigma _{*}^{2}}
 
.
When σ2≠σ1 the constant A is different from the constant of normal distribution. However, when 
 
 
 
 
 σ
 
 1
 
 
 2
 
 
 =
 
 σ
 
 2
 
 
 2
 
 
 =
 
 σ
 
 ∗
 
 
 2
 
 
 
 
 {\displaystyle \sigma _{1}^{2}=\sigma _{2}^{2}=\sigma _{*}^{2}}
 
 the constants are equal.
The sign of its third central moment is determined by the difference (σ2-σ1). If this difference is positive, the distribution is skewed to the right and if negative, then it is skewed to the left.
Other properties of the split normal density were discussed by Johnson et al.<a class="footnote-ref" id="fnref:8" href="#fn:8">8</a> and Julio.<a class="footnote-ref" id="fnref:9" href="#fn:9">9</a>

<h2 id="alternative-formulations">Alternative formulations</h2>
The formulation discussed above originates from John.<a class="footnote-ref" id="fnref:10" href="#fn:10">10</a> The literature offers two mathematically equivalent alternative parameterizations . Britton, Fisher and Whitley<a class="footnote-ref" id="fnref:11" href="#fn:11">11</a> offer a parameterization if terms of mode, dispersion and normed skewness, denoted with 
 
 
 
 
 
 S
 N
 
 
 (
 μ
 ,
 
 
 σ
 
 2
 
 
 ,
 γ
 )
 
 
 {\displaystyle {\mathcal {SN}}(\mu ,\,\sigma ^{2},\gamma )}
 
. The parameter μ is the mode and has equivalent to the mode in John's formulation. The parameter σ 2>0 informs about the dispersion (scale) and should not be confused with variance. The third parameter, γ ∈ (-1,1), is the normalized skew.
The second alternative parameterization is used in the <a href="/facts/Bank_of_England/9ITfYV3q">Bank of England's</a> communication and is written in terms of mode, dispersion and unnormed skewness and is denoted with 
 
 
 
 
 
 S
 N
 
 
 (
 μ
 ,
 
 
 σ
 
 2
 
 
 ,
 ξ
 )
 
 
 {\displaystyle {\mathcal {SN}}(\mu ,\,\sigma ^{2},\xi )}
 
. In this formulation the parameter μ is the mode and is identical as in John's <a class="footnote-ref" id="fnref:12" href="#fn:12">12</a> and Britton, Fisher and Whitley's <a class="footnote-ref" id="fnref:13" href="#fn:13">13</a> formulation. The parameter σ 2 informs about the dispersion (scale) and is the same as in the Britton, Fisher and Whitley's formulation. The parameter ξ equals the difference between the distribution's mean and mode and can be viewed as unnormed measure of skewness.
The three parameterizations are mathematically equivalent, meaning that there is a strict relationship between the parameters and that it is possible to go from one parameterization to another. The following relationships hold:<a class="footnote-ref" id="fnref:14" href="#fn:14">14</a>

σ
                  
                    2
                  
                
              
              
                
                =
                
                  σ
                  
                    1
                  
                  
                    2
                  
                
                (
                1
                +
                γ
                )
                =
                
                  σ
                  
                    2
                  
                  
                    2
                  
                
                (
                1
                −
                γ
                )
              
            
            
              
                γ
              
              
                
                =
                
                  
                    
                      
                        σ
                        
                          2
                        
                        
                          2
                        
                      
                      −
                      
                        σ
                        
                          1
                        
                        
                          2
                        
                      
                    
                    
                      
                        σ
                        
                          2
                        
                        
                          2
                        
                      
                      +
                      
                        σ
                        
                          1
                        
                        
                          2
                        
                      
                    
                  
                
              
            
            
              
                ξ
              
              
                
                =
                
                  
                    2
                    
                      /
                    
                    π
                  
                
                (
                
                  σ
                  
                    2
                  
                
                −
                
                  σ
                  
                    1
                  
                
                )
              
            
            
              
                γ
              
              
                
                =
                sgn
                ⁡
                (
                ξ
                )
                
                  
                    1
                    −
                    
                      
                        (
                        
                          
                            
                              
                                
                                  1
                                  +
                                  2
                                  β
                                
                              
                              −
                              1
                            
                            β
                          
                        
                        )
                      
                      
                        2
                      
                    
                  
                
                ,
                
                
                  where
                
                
                β
                =
                
                  
                    
                      π
                      
                        ξ
                        
                          2
                        
                      
                    
                    
                      2
                      
                        σ
                        
                          2
                        
                      
                    
                  
                
                .
              
            
          
        
      
    
    {\displaystyle {\begin{aligned}\sigma ^{2}&=\sigma _{1}^{2}(1+\gamma )=\sigma _{2}^{2}(1-\gamma )\\\gamma &={\frac {\sigma _{2}^{2}-\sigma _{1}^{2}}{\sigma _{2}^{2}+\sigma _{1}^{2}}}\\\xi &={\sqrt {2/\pi }}(\sigma _{2}-\sigma _{1})\\\gamma &=\operatorname {sgn} (\xi ){\sqrt {1-\left({\frac {{\sqrt {1+2\beta }}-1}{\beta }}\right)^{2}}},\quad {\text{where}}\quad \beta ={\frac {\pi \xi ^{2}}{2\sigma ^{2}}}.\end{aligned}}}

<h2 id="multivariate-extensions">Multivariate Extensions</h2>
The multivariate generalization of the split normal distribution was proposed by Villani and Larsson.<a class="footnote-ref" id="fnref:15" href="#fn:15">15</a> They assume that each of the <a href="/facts/Principal_component_analysis/pCbVTqdR">principal components</a> has univariate split normal distribution with a different set of parameters μ, σ2 and σ1.

<h2 id="estimation-of-parameters">Estimation of parameters</h2>
John<a class="footnote-ref" id="fnref:16" href="#fn:16">16</a> proposes to estimate the parameters using <a href="/facts/Maximum_likelihood/0Yq2dpQD">maximum likelihood</a> method. He shows that the likelihood function can be expressed in an intensive form, in which the scale parameters σ1 and σ2 are a function of the location parameter μ. The likelihood in its intensive form is:

L
 (
 μ
 )
 =
 −
 
 
 [
 
 
 ∑
 
 
 x
 
 i
 
 
 :
 
 x
 
 i
 
 
 <
 μ
 
 
 (
 
 x
 
 i
 
 
 −
 μ
 
 )
 
 2
 
 
 
 ]
 
 
 1
 
 /
 
 3
 
 
 −
 
 
 [
 
 
 ∑
 
 
 x
 
 i
 
 
 :
 
 x
 
 i
 
 
 >
 μ
 
 
 (
 
 x
 
 i
 
 
 −
 μ
 
 )
 
 2
 
 
 
 ]
 
 
 1
 
 /
 
 3
 
 
 
 
 {\displaystyle L(\mu )=-\left[\sum _{x_{i}:x_{i}<\mu }(x_{i}-\mu )^{2}\right]^{1/3}-\left[\sum _{x_{i}:x_{i}>\mu }(x_{i}-\mu )^{2}\right]^{1/3}}

and has to be maximized numerically with respect to a single parameter μ only.
Given the maximum likelihood estimator 
 
 
 
 
 
 
 μ
 ^
 
 
 
 
 
 {\displaystyle {\hat {\mu }}}
 
 the other parameters take values:

σ
 ^
 
 
 
 
 1
 
 
 2
 
 
 =
 
 
 
 −
 L
 (
 μ
 )
 
 N
 
 
 
 
 [
 
 
 ∑
 
 
 x
 
 i
 
 
 :
 
 x
 
 i
 
 
 <
 μ
 
 
 (
 
 x
 
 i
 
 
 −
 μ
 
 )
 
 2
 
 
 
 ]
 
 
 2
 
 /
 
 3
 
 
 ,
 
 
 {\displaystyle {\hat {\sigma }}_{1}^{2}={\frac {-L(\mu )}{N}}\left[\sum _{x_{i}:x_{i}<\mu }(x_{i}-\mu )^{2}\right]^{2/3},}

σ
                ^
              
            
          
          
            2
          
          
            2
          
        
        =
        
          
            
              −
              L
              (
              μ
              )
            
            N
          
        
        
          
            [
            
              
                ∑
                
                  
                    x
                    
                      i
                    
                  
                  :
                  
                    x
                    
                      i
                    
                  
                  >
                  μ
                
              
              (
              
                x
                
                  i
                
              
              −
              μ
              
                )
                
                  2
                
              
            
            ]
          
          
            2
            
              /
            
            3
          
        
        ,
      
    
    {\displaystyle {\hat {\sigma }}_{2}^{2}={\frac {-L(\mu )}{N}}\left[\sum _{x_{i}:x_{i}>\mu }(x_{i}-\mu )^{2}\right]^{2/3},}

where N is the number of observations.
Villani and Larsson<a class="footnote-ref" id="fnref:17" href="#fn:17">17</a> propose to use either <a href="/facts/Maximum_likelihood/0Yq2dpQD">maximum likelihood</a> method or <a href="/facts/Bayes_estimator/BpFRxYBg">bayesian estimation</a> and provide some analytical results for either univariate and multivariate case.

<h2 id="applications">Applications</h2>
The split normal distribution has been used mainly in econometrics and time series. A remarkable area of application is the construction of the <a href="/facts/Fan_chart_(time_series)/kW9yQA6X">fan chart</a>, a representation of the <a href="/facts/Inflation/YvKtKrVI">inflation</a> forecast distribution reported by <a href="/facts/Inflation_targeting/QP54rsfE">inflation targeting</a> central banks around the globe.<a class="footnote-ref" id="fnref:18" href="#fn:18">18</a><a class="footnote-ref" id="fnref:19" href="#fn:19">19</a>

<h2 id="references">References</h2>

<ol>
<li id="fn:1">Johnson, N.L., Kotz, S. and Balakrishnan, N. (1994). Continuous Univariate Distributions, Volume 1. John Wiley & Sons. p. 173. ISBN 978-0-471-58495-7.{{cite book}}: CS1 maint: multiple names: authors list (link) <a href="978-0-471-58495-7" target="_blank">978-0-471-58495-7</a> <a href="#fnref:1" class="footnote-back-ref">↩</a></li>
<li id="fn:2">Gibbons, J.F.; Mylroie, S. (1973). "Estimation of impurity profiles in ion-implanted amorphous targets using joined half-Gaussian distributions". Applied Physics Letters. 22 (11): 568–569. Bibcode:1973ApPhL..22..568G. doi:10.1063/1.1654511. <a href="/wiki/Bibcode_(identifier)" target="_blank">/wiki/Bibcode_(identifier)</a> <a href="#fnref:2" class="footnote-back-ref">↩</a></li>
<li id="fn:3">John, S. (1982). "The three-parameter two-piece normal family of distributions and its fitting". Communications in Statistics - Theory and Methods. 11 (8): 879–885. doi:10.1080/03610928208828279. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:3" class="footnote-back-ref">↩</a></li>
<li id="fn:4">Fechner, G.T. (ed. Lipps, G.F.) (1897). Kollectivmasslehre. Engelmann, Leipzig. <a href="#fnref:4" class="footnote-back-ref">↩</a></li>
<li id="fn:5">Wallis, K.F. (2014). The two-piece normal, binormal, or double Gaussian distribution: its origin and rediscoveries. Statistical Science, vol. 29, no. 1, pp.106-112. doi:10.1214/13-STS417. <a href="#fnref:5" class="footnote-back-ref">↩</a></li>
<li id="fn:6">de Roon, F. and Karehnke, P. (2016). A simple skewed distribution with asset pricing applications. Review of Finance, 2016, 1-29. <a href="/wiki/European_Finance_Association" target="_blank">/wiki/European_Finance_Association</a> <a href="#fnref:6" class="footnote-back-ref">↩</a></li>
<li id="fn:7">Johnson, N.L., Kotz, S. and Balakrishnan, N. (1994). Continuous Univariate Distributions, Volume 1. John Wiley & Sons. p. 173. ISBN 978-0-471-58495-7.{{cite book}}: CS1 maint: multiple names: authors list (link) <a href="978-0-471-58495-7" target="_blank">978-0-471-58495-7</a> <a href="#fnref:7" class="footnote-back-ref">↩</a></li>
<li id="fn:8">Johnson, N.L., Kotz, S. and Balakrishnan, N. (1994). Continuous Univariate Distributions, Volume 1. John Wiley & Sons. p. 173. ISBN 978-0-471-58495-7.{{cite book}}: CS1 maint: multiple names: authors list (link) <a href="978-0-471-58495-7" target="_blank">978-0-471-58495-7</a> <a href="#fnref:8" class="footnote-back-ref">↩</a></li>
<li id="fn:9">Juan Manuel Julio (2007). The Fan Chart: The Technical Details Of The New Implementation. Banco de la República. Retrieved 2010-09-11, direct link {{cite conference}}: External link in |postscript= (help)CS1 maint: postscript (link) <a href="http://ideas.repec.org/p/col/000094/004294.html" target="_blank">http://ideas.repec.org/p/col/000094/004294.html</a> <a href="#fnref:9" class="footnote-back-ref">↩</a></li>
<li id="fn:10">John, S. (1982). "The three-parameter two-piece normal family of distributions and its fitting". Communications in Statistics - Theory and Methods. 11 (8): 879–885. doi:10.1080/03610928208828279. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:10" class="footnote-back-ref">↩</a></li>
<li id="fn:11">Britton, E.; P. Fisher; Whitley, J. (1998). "The inflation report projections: understanding the fan chart". Quarterly Bulletin. February 1998: 30–37. <a href="#fnref:11" class="footnote-back-ref">↩</a></li>
<li id="fn:12">John, S. (1982). "The three-parameter two-piece normal family of distributions and its fitting". Communications in Statistics - Theory and Methods. 11 (8): 879–885. doi:10.1080/03610928208828279. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:12" class="footnote-back-ref">↩</a></li>
<li id="fn:13">Britton, E.; P. Fisher; Whitley, J. (1998). "The inflation report projections: understanding the fan chart". Quarterly Bulletin. February 1998: 30–37. <a href="#fnref:13" class="footnote-back-ref">↩</a></li>
<li id="fn:14">Banerjee, N.; A. Das (2011). Fan Chart: Methodology and its Application to Inflation Forecasting in India. Reserve Bank of India Working Paper Series. <a href="#fnref:14" class="footnote-back-ref">↩</a></li>
<li id="fn:15">Villani, Mattias; Rolf Larsson (2006). "The Multivariate Split Normal Distribution and Asymmetric Principal Components Analysis". Communications in Statistics - Theory and Methods. 35 (6): 1123–1140. CiteSeerX 10.1.1.533.4095. doi:10.1080/03610920600672252. ISSN 0361-0926. S2CID 124959166. <a href="/wiki/CiteSeerX_(identifier)" target="_blank">/wiki/CiteSeerX_(identifier)</a> <a href="#fnref:15" class="footnote-back-ref">↩</a></li>
<li id="fn:16">John, S. (1982). "The three-parameter two-piece normal family of distributions and its fitting". Communications in Statistics - Theory and Methods. 11 (8): 879–885. doi:10.1080/03610928208828279. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:16" class="footnote-back-ref">↩</a></li>
<li id="fn:17">Villani, Mattias; Rolf Larsson (2006). "The Multivariate Split Normal Distribution and Asymmetric Principal Components Analysis". Communications in Statistics - Theory and Methods. 35 (6): 1123–1140. CiteSeerX 10.1.1.533.4095. doi:10.1080/03610920600672252. ISSN 0361-0926. S2CID 124959166. <a href="/wiki/CiteSeerX_(identifier)" target="_blank">/wiki/CiteSeerX_(identifier)</a> <a href="#fnref:17" class="footnote-back-ref">↩</a></li>
<li id="fn:18">Juan Manuel Julio (2007). The Fan Chart: The Technical Details Of The New Implementation. Banco de la República. Retrieved 2010-09-11, direct link {{cite conference}}: External link in |postscript= (help)CS1 maint: postscript (link) <a href="http://ideas.repec.org/p/col/000094/004294.html" target="_blank">http://ideas.repec.org/p/col/000094/004294.html</a> <a href="#fnref:18" class="footnote-back-ref">↩</a></li>
<li id="fn:19">Bank of England, Inflation Report Archived 2010-08-13 at the Wayback Machine <a href="http://www.bankofengland.co.uk/publications/inflationreport/irfanch.htm" target="_blank">http://www.bankofengland.co.uk/publications/inflationreport/irfanch.htm</a> <a href="#fnref:19" class="footnote-back-ref">↩</a></li>
</ol>

Split normal distribution open-in-new

Split normal distribution