*! version 1.1.0 20nov2025 I I Bolotov program define arimaauto, rclass byable(recall) version 15.1 /* Finds the best [S]ARIMA[X] model with the help of the Hyndman-Khandakar algorithm through stepwise traversing of the model space or a bulk estimation. The user can choose between LLF, AIC, and SIC and pass arguments to arima (estimation), hegy, dfgls, and kpss (unit root tests) commands. The output is consistent with (SSC) arimasel. Author: Ilya Bolotov, MBA, Ph.D. Date: 15 January 2022 */ tempname limits tests models vmaxLLF vminAIC vminSIC // check for third-party packages from SSC cap which hegy if _rc { di as err "hegy not found; click to install {net sj 16-3 st0453:hegy}" exit 499 } cap which kpss if _rc { di as err "kpss not found; install from SSC" exit 499 } // replay last result if replay() { if _by() { error 190 } cap confirm mat r(models) if _rc { di as err "results of arimaauto not found" exit 301 } /* copy return values */ loc best `=r(best)' loc ictests `=r(ictests)' loc icarima `=r(icarima)' mat `limits' = r(limits) mat `tests' = r(tests) mat `models' = r(models) sca `vmaxLLF' = r(maxllf) sca `vminAIC' = r(minaic) sca `vminSIC' = r(minsic) /* print output */ cap confirm mat `tests' if ! _rc { loc title = "Unit root tests:" loc rspec = "& - `= "& " * rowsof(`tests')'" loc cspec = "& %12s | %10.0f | %5.0f | %9.6f & %9.6f & " + /// "%9.6f & %9.6f &" matlist `tests', title(`title') rspec(`rspec') cspec(`cspec') } cap confirm mat `models' if ! _rc { loc title = "Model space:" loc rspec = "& - `= "& " * rowsof(`models')'" loc cspec = "& %12s | %4.0f & %4.0f & %4.0f & %4.0f & " + /// "%5.0f | %9.4f & %9.4f & %9.4f &" matlist `models', title(`title') rspec(`rspec') cspec(`cspec') } di as res _n "Max LLF: Model `=`vmaxLLF''" di as res "Min AIC: Model `=`vminAIC''" di as res "Min SIC: Model `=`vminSIC''" di as res _n "Best selected based on `=ustrupper("`icarima'")':" arima /* return output */ cap ret loc best `best' cap ret loc ictests `ictests' cap ret loc icarima `icarima' cap ret hidden mat limits = `limits' cap ret mat tests = `tests' cap ret mat models = `models' cap ret sca maxllf = `vmaxLLF' cap ret sca minaic = `vminAIC' cap ret sca minsic = `vminSIC' cap ret sca N = e(N) cap ret sca np = e(df_m) + 1 exit 0 } // syntax syntax /// [varlist(ts fv)] [if] [in] [iw] [, /// ARIMA(numlist int min=3 max=3 >-1) /// SARIMA(numlist int min=4 max=4 >-1) /// MAX(numlist int min=2 max=2 >-1) /// MMAX(numlist int min=2 max=2 >-1) /// HEGY(string asis) DFGLS(string asis) KPSS(string asis) /// MAXLag(numlist integer >=0 max=1) /// Level(cilevel) Mode(string) IC(string) /// STATionary noSEASonal /// noSTEPwise /// MAXModels(numlist integer >=1 max=1) /// INVRoot(real `=1/1.001') /// ITERate(int 100) TRACE(int 0) * /// ] // adjust and preprocess options loc iw = cond(`"`weight'`exp'"' == "", "", `"[`weight'`exp']"' ) loc maxlag = cond(`"`maxlag'"' == "", ".", `"`maxlag'"' ) loc maxmodels = cond(`"`maxmodels'"' == "", ".", `"`maxmodels'"' ) // examine data qui tsset, noq if "`r(panelvar)'" != "" { di as err "command may not be used with panel data" exit 459 } if trim(`"`seasonal'"') == "" & ! inlist(r(unit1), ".", "q", "m") { di as err "hegy must be used with monthly or quarterly data" /// _n as txt "please check " as res "tsset" as txt " or " as res "xtset" exit 459 } if trim(`"`seasonal'"') == "" & _N <= cond(r(unit1) == "q", 4, 12) { di as err "observation numbers out of range for hegy" /// _n as txt "must be greater than " as res cond(r(unit1) == "q", 4, 12) exit 459 } // pass arguments to ARIMAAuto mata: AA = ARIMAAuto() mata: AA.put("varlist","`varlist'" ) mata: AA.put("ifin", `"`if' `in'"' ) mata: AA.put("iw", `"`iw'"' ) mata: AA.put("level", `level' ) mata: AA.put("mode", `"`mode'"' ) mata: AA.put("ic", `"`ic'"' ) mata: AA.put("o_hegy", `"`hegy'"' ) mata: AA.put("o_dfgls",`"`dfgls'"' ) mata: AA.put("o_kpss", `"`kpss'"' ) mata: AA.put("o_arima",`"`options'"' ) mata: AA.put("f_s", `"`seasonal'"' == "" ? 1 : 0 ) mata: AA.put("f_i", `"`stationary'"' == "" ? 1 : 0 ) mata: AA.put("f_sw", `"`stepwise'"' == "" ? 1 : 0 ) mata: AA.put("f_t", `trace' ) mata: AA.put("L", ("`max'","`mmax'","`invroot'","`maxlag'", /// "`maxmodels'","`iterate'") ) mata: AA.put("MS", ("`arima'","`sarima'") ) // run ARIMAAuto mata: AA.start() /* get information criteria */ mata: st_local("ictests", AA.get("mode") ) mata: st_local("icarima", AA.get("ic") ) /* get limits */ mata: st_matrix("`limits'", AA.get("L")' ) mata: if (length(AA.get("L"))) st_matrixrowstripe( /// "`limits'", (J(8,1,""),("AR","MA","MAR","MMA","invroot","lags", /// "models","iterations")') /// );; mata: if (length(AA.get("L"))) st_matrixcolstripe( /// "`limits'", (J(1,1,""),("value")) /// );; /* get tests */ mata: st_matrix("`tests'", AA.get("T") ) mata: if (length(AA.get("T"))) st_matrixrowstripe( /// "`tests'", (((""\(rows(AA.get("T")) == 1 ? J(0,1,"") : /// (subinstr(((mod(rows(AA.get("T")), 2) ? "S" + /// strofreal(AA.get("MS")[1,7]) + "." : "") + "D") :+ /// strofreal((0::floor(rows(AA.get("T"))/2)-1)# /// J(2,1,1)) :+ ".", "D0.", "")))) :+ /// "`: word 1 of `varlist''" /// ),("HEGY", /// tokens("DFGLS KPSS " * floor(rows(AA.get("T"))/2)))')[ /// (mod(rows(AA.get("T")), 2)?.:2::rows(AA.get("T"))+1),.] /// );; mata: if (length(AA.get("T"))) st_matrixcolstripe( /// "`tests'", (J(6,1,""),("unit root","lags","Stat", /// ("1%","5%","10%"):+" crit")') /// );; /* get models */ mata: st_matrix("`models'", AA.get("MS")[,(1,3,4,6,8,10..12)] ) mata: if (length(AA.get("MS"))) st_matrixrowstripe( /// "`models'", (J(rows(AA.get("MS")),1,""), /// ("Model" :+ strofreal(1::rows(AA.get("MS"))))) /// );; mata: if (length(AA.get("MS"))) st_matrixcolstripe( /// "`models'", (J(8,1,""),("AR","MA","MAR","MMA","const","LLF", /// "AIC","SIC")') /// );; /* get the best [S]ARIMA[X] model based on LLF, AIC, SIC */ mata: bm = AA.get("MS")[selectindex(AA.get("MS")[,(j=(AA.get("ic") :== /// ("llf","aic","sic")) * (10::12))] :== (*((j==10)? &max() : /// &min()))(AA.get("MS")[,j]))[1],]; /// st_local("best", "ARIMA" + /// "(" + invtokens(strofreal(bm[1..3]),",") + ")" + /// (AA.get("f_s") ? "(" + invtokens(strofreal(bm[4..7]),",") + ")" : /// "") + (bm[8] ? " with constant" : " without constant")) mata: st_numscalar( /// "`vmaxLLF'", /// selectindex(AA.get("MS")[,10] :== max(AA.get("MS")[,10]))[1] /// ) mata: st_numscalar( /// "`vminAIC'", /// selectindex(AA.get("MS")[,11] :== min(AA.get("MS")[,11]))[1] /// ) mata: st_numscalar( /// "`vminSIC'", /// selectindex(AA.get("MS")[,12] :== min(AA.get("MS")[,12]))[1] /// ) // print output cap confirm mat `tests' if ! _rc { di as res _n "Unit root tests:" loc rspec = "& - `= "& " * rowsof(`tests')'" loc cspec = "& %12s | %10.0f | %5.0f | %9.6f & %9.6f & " + /// "%9.6f & %9.6f &" matlist `tests', title(`title') rspec(`rspec') cspec(`cspec') } cap confirm mat `models' if ! _rc { di as res _n "Model space:" loc rspec = "& - `= "& " * rowsof(`models')'" loc cspec = "& %12s | %4.0f & %4.0f & %4.0f & %4.0f & " + /// "%5.0f | %9.4f & %9.4f & %9.4f &" matlist `models', title(`title') rspec(`rspec') cspec(`cspec') } di as res _n "Max LLF: Model `=`vmaxLLF''" di as res "Min AIC: Model `=`vminAIC''" di as res "Min SIC: Model `=`vminSIC''" di as res _n "Best model based on `=ustrupper("`icarima'")':" arima // return output cap ret loc best `best' cap ret loc ictests `ictests' cap ret loc icarima `icarima' cap ret hidden mat limits = `limits' cap ret mat tests = `tests' cap ret mat models = `models' cap ret sca maxllf = `vmaxLLF' cap ret sca minaic = `vminAIC' cap ret sca minsic = `vminSIC' cap ret sca N = e(N) cap ret sca np = e(df_m) + 1 // clear memory mata: mata drop AA bm end