o 0Gf@s dZddlZddlmZddlZddlZddlm Z ddl m Z ddl m Z ddlmZddlmmZddlmZd d lmZmZd d lmZmZmZd d lmZd d lm Z m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&GdddeZ'GdddeZ(GdddeZ)e*e)e(dS)zs Vector Autoregressive Moving Average with eXogenous regressors model Author: Chad Fulton License: Simplified-BSD N)warn)Appender)Bunch)_is_using_pandas) var_model)EstimationWarning)INVERT_UNIVARIATESOLVE_LU)MLEModel MLEResultsMLEResultsWrapper)Initialization) is_invertibleconcat prepare_exog!constrain_stationary_multivariate#unconstrain_stationary_multivariateprepare_trend_specprepare_trend_datacseZdZdZ    dfd d Zd d d Zed dZeddZeddZ ddZ ddZ fddZ  d!ddZ ejddZeejj    d"fdd ZZS)#VARMAXuM Vector Autoregressive Moving Average with eXogenous regressors model Parameters ---------- endog : array_like The observed time-series process :math:`y`, , shaped nobs x k_endog. exog : array_like, optional Array of exogenous regressors, shaped nobs x k. order : iterable The (p,q) order of the model for the number of AR and MA parameters to use. trend : str{'n','c','t','ct'} or iterable, optional Parameter controlling the deterministic trend polynomial :math:`A(t)`. Can be specified as a string where 'c' indicates a constant (i.e. a degree zero component of the trend polynomial), 't' indicates a linear trend with time, and 'ct' is both. Can also be specified as an iterable defining the non-zero polynomial exponents to include, in increasing order. For example, `[1,1,0,1]` denotes :math:`a + bt + ct^3`. Default is a constant trend component. error_cov_type : {'diagonal', 'unstructured'}, optional The structure of the covariance matrix of the error term, where "unstructured" puts no restrictions on the matrix and "diagonal" requires it to be a diagonal matrix (uncorrelated errors). Default is "unstructured". measurement_error : bool, optional Whether or not to assume the endogenous observations `endog` were measured with error. Default is False. enforce_stationarity : bool, optional Whether or not to transform the AR parameters to enforce stationarity in the autoregressive component of the model. Default is True. enforce_invertibility : bool, optional Whether or not to transform the MA parameters to enforce invertibility in the moving average component of the model. Default is True. trend_offset : int, optional The offset at which to start time trend values. Default is 1, so that if `trend='t'` the trend is equal to 1, 2, ..., nobs. Typically is only set when the model created by extending a previous dataset. **kwargs Keyword arguments may be used to provide default values for state space matrices or for Kalman filtering options. See `Representation`, and `KalmanFilter` for more details. Attributes ---------- order : iterable The (p,q) order of the model for the number of AR and MA parameters to use. trend : str{'n','c','t','ct'} or iterable Parameter controlling the deterministic trend polynomial :math:`A(t)`. Can be specified as a string where 'c' indicates a constant (i.e. a degree zero component of the trend polynomial), 't' indicates a linear trend with time, and 'ct' is both. Can also be specified as an iterable defining the non-zero polynomial exponents to include, in increasing order. For example, `[1,1,0,1]` denotes :math:`a + bt + ct^3`. error_cov_type : {'diagonal', 'unstructured'}, optional The structure of the covariance matrix of the error term, where "unstructured" puts no restrictions on the matrix and "diagonal" requires it to be a diagonal matrix (uncorrelated errors). Default is "unstructured". measurement_error : bool, optional Whether or not to assume the endogenous observations `endog` were measured with error. Default is False. enforce_stationarity : bool, optional Whether or not to transform the AR parameters to enforce stationarity in the autoregressive component of the model. Default is True. enforce_invertibility : bool, optional Whether or not to transform the MA parameters to enforce invertibility in the moving average component of the model. Default is True. Notes ----- Generically, the VARMAX model is specified (see for example chapter 18 of [1]_): .. math:: y_t = A(t) + A_1 y_{t-1} + \dots + A_p y_{t-p} + B x_t + \epsilon_t + M_1 \epsilon_{t-1} + \dots M_q \epsilon_{t-q} where :math:`\epsilon_t \sim N(0, \Omega)`, and where :math:`y_t` is a `k_endog x 1` vector. Additionally, this model allows considering the case where the variables are measured with error. Note that in the full VARMA(p,q) case there is a fundamental identification problem in that the coefficient matrices :math:`\{A_i, M_j\}` are not generally unique, meaning that for a given time series process there may be multiple sets of matrices that equivalently represent it. See Chapter 12 of [1]_ for more information. Although this class can be used to estimate VARMA(p,q) models, a warning is issued to remind users that no steps have been taken to ensure identification in this case. References ---------- .. [1] Lütkepohl, Helmut. 2007. New Introduction to Multiple Time Series Analysis. Berlin: Springer. Nrrc unstructuredFTrc  s*|_|_|_|_|_t|d_t|d_|dvr%tdjdkr3jdkr3tdjdkrBjdkrBt dt |_ | _ t j \__jjdko]jddk_t|\_}jdk_t|dsvt|}tjd} | j_|jd} | } | j}| dd| d ttBtj|f||| d | jdksjdkrjsd j _!i_"j#jj"d <j#d jj"d<j#d jj"d<j#jj"d<jdkrj#j"d<njdkrtj#j#dd j"d<j#jj"d<t$j"%_&t'jjj(dj d}|dd_)|dd_*jdkr>jrDjdkrPt+j,j(fj d<t-j#}dj d|<jdkrt-jdj#}|dj#|df}dj d|<t-jdj#}|d| dj#|d| j#f}dj d|<t-j#}dj d|<|d| j#|df}jdkrdj d|<jrjdkrtj.dd| ddf_/njdksjdkrtj.dd| ddf_/jdkrtj.dd| ddf_0n tj.dd| | df_0jdkr/dt-j#_1n jdkr7_._slice)orderr error_cov_typemeasurement_errorenforce_stationarityenforce_invertibility)?r6r7r8r9r5intk_ark_ma ValueErrorrrr trend_offsetrpolynomial_trendk_trendsize_trend_is_constrk_exogmle_regressionrr,Z asanyarraymaxZ_k_ordershape setdefaultr r super__init__ssmZ_time_invariantr+k_endogsumvaluesk_paramsrnobs _trend_data _final_trendzerosrZ diag_indicesr-_idx_state_intercept_idx_transition_idx_state_covZ tril_indices_idx_lower_state_cov _idx_obs_cov _params_trend _params_ar _params_ma_params_regression_params_state_cov_params_obs_cov _final_exogZ _init_keyslistkeys)r1endogrr5r r6r7r8r9r>kwargsZ _min_k_arrKrrZ trend_dataidxr4r& __class__r0r3rIs                        zVARMAX.__init__cKs|j|fd|i|S)Nr)Z_clone_from_init_kwds)r1rarrbr2r2r3clone7sz VARMAX.clonecCs dttfiS)Nfit) VARMAXResultsVARMAXResultsWrapperr0r2r2r3 _res_classes:s zVARMAX._res_classescCstj|jtjd}t|j}|}tj | dd}d}|j dkr4|j dkr4tj |j|jf}n|j dkr=|j}n|j dkrE|j}tt|rdtjt|dd}||}|durd||}td}td}|j dksx|j dkrtj||}|t||8}|j dkr|d|j j}|jdkr||j dj}g}|jdkr|jnd} t|} | j| ddd} |jdkrt| jj}| j}|jdkr|jr| |j|j|jj |j|j|jj} t!dgt"| } | st#d |d9}g}|j$dkrKt|}|j|j$ddd}t|jj}|j%rK| |j|j$|jj |j|j|j$j} t!dgt"| }|sKt#d |d9}|jdkr|j dks[|j&r| |j|j|jj |j|j|jj} t'|jtj(| dd}|j dkrt||}|j&dkrt||}|j dkr|||j)<|jdkr|||j*<|j$dkr|||j+<|j&r|||j,<|j-d kr| j./||j0<n|j-d krtj1| j.}||j2||j0<|j3r|j$dkr|j./||j4<|S| j./||j4<|S) NdtypeW)Z requirementsrr)Zaxisn)ZmaxlagsZicr z\Non-stationary starting autoregressive parameters found. Using zeros as starting parameters.z\Non-stationary starting moving-average parameters found. Using zeros as starting parameters.rr)5r,rRrNZfloat64pdZ DataFrameracopyZ interpolateZrequireZbfillr@rCc_rPranyZisnanZlinalgZpinvdotTrKr;rVARrgarrayparamsravelZresidr8reshaperr_rr<r9rDeyerLrXrYrZr[r6Zsigma_urr\ZcholeskyrVr7r])r1rwrarmask trend_params exog_paramsZtrendexog_params ar_paramsr;Zmod_arZres_arcoefficient_matricesr ma_paramsZmod_maZres_maZ invertibletmpZ cov_factorr2r2r3 start_params>s                        zVARMAX.start_paramscsfg}jtjtsgjdkrLtjD]3}jdD])}|dkr1|d|g7}q!|dkr?|d|g7}q!|d||fg7}q!q|fddtjD7}|fddtjD7}|fd dtjD7}jd kr|fd dtjD7}njd kr|fd dtjD7}j r|fddtjD7}|S)Nrz intercept.%srzdrift.%sz trend.%d.%sc Fg|]}tjD]}tjD]}d|d||fqq qS)z L%d.%s.%sr)ranger;rK.0jik endog_namesr1r2r3  z&VARMAX.param_names..c r)z L%d.e(%s).%sr)rr<rKrrr2r3rrcs2g|]}tjD] }dj||q qS)z beta.{}.{})rrCformatZ exog_namesrrrrr2r3rsrcg|]}d|qS)z sigma2.%sr2rrrr2r3r rcsFg|]}t|dD]}||krd|n d||q qS)rz sqrt.var.%szsqrt.cov.{}.{})rrrrr2r3rs cr)zmeasurement_variance.%sr2rrr2r3rr) r isinstancer_r@rrKr?Znonzeror6r7)r1 param_namesrrr2rr3rsF        zVARMAX.param_namescCstj|dd}tj|j|jd}||j||j<|jdkrm|jrm|jdkr0t ||j d}n |jdkrPtj|j dj|jd}||j ||j <t ||j}||j|j|j|j}t||\}}|||j<n||j||j<|jdkr|jrtj|j|jd}||j|j|j|j}t||\}}|||j<n||j||j<||j||j<|jdkr||j d||j <n |jdkr||j ||j <|jr||jd||j<|S) a[ Transform unconstrained parameters used by the optimizer to constrained parameters used in likelihood evaluation Parameters ---------- unconstrained : array_like Array of unconstrained parameters used by the optimizer, to be transformed. Returns ------- constrained : array_like Array of constrained parameters which may be used in likelihood evaluation. Notes ----- Constrains the factor transition to be stationary and variances to be positive. rZndminrkrrr!rr$)r,rvrRrFrlrXr;r8r6diagr\rJrVrsrtrYryrKrrxr<r9rzrZr[r7r])r1 unconstrained constrainedr$state_cov_lower coefficientsrvariancer2r2r3transform_paramssR        zVARMAX.transform_paramscCstj|dd}tj|j|jd}||j||j<|jdkrk|jrk|jdkr.t ||j }n |jdkrNtj|j dj|jd}||j ||j <t ||j}||j|j|j|j}t||\}}|||j<n||j||j<|jdkr|jrtj|j|jd}||j|j|j|j}t||\}}|||j<n||j||j<||j||j<|jdkr||j d||j <n |jdkr||j ||j <|jr||jd||j<|S) a Transform constrained parameters used in likelihood evaluation to unconstrained parameters used by the optimizer. Parameters ---------- constrained : array_like Array of constrained parameters used in likelihood evaluation, to be transformed. Returns ------- unconstrained : array_like Array of unconstrained parameters used by the optimizer. rrrkrrrr$g?)r,rvrRrFrlrXr;r8r6rr\rJrVrsrtrYryrKrrxr<r9rzrZr[r7r])r1rrr$rrZunconstrained_matricesrr2r2r3untransform_paramsRsR        zVARMAX.untransform_paramscst|tt|jdd}ddt|j|D\}}}}}}|j rN|j dkrN|j dks8|j dkrN| |}t ||dk}|rN|sNtd|jrt|jdkrv|j s^|jdkrz| |}t ||dk}|rx|s|tddSdSdSdSdS)Nr'css|]}|VqdSr*)tolist)rZarrr2r2r3 s z2VARMAX._validate_can_fix_params..rrzCannot fix individual autoregressive parameters when `enforce_stationarity=True`. In this case, must either fix all autoregressive parameters or none.zCannot fix individual moving average parameters when `enforce_invertibility=True`. In this case, must either fix all moving average parameters or none.)rH_validate_can_fix_paramsr,Zcumsumr_r+rMZ array_splitrr8r;rK issupersetlen intersectionr=r9r<)r1rZix_Zar_namesZma_namesZfix_allZfix_anyrdr2r3rs6    zVARMAX._validate_can_fix_paramsc CsT|j|||d}|jr>||j|j|jj}t|j dd|}|j|j |j <|j dur>t|j ||j dd|jdf<|j dkr|jsWtjd|jd}||j dddf<||j|j|j j}|jri|}n t|jdd|}|j |j |j7<|jdur|j djdkr|j dd|jddft|j|j7<|jr|j durtjtj|jd} | |j dd|jdf<||j|j|j|j} ||j|j|j|j} tj| | f|j |j<|jdkr||j|j |j<n$|jdkrtj|j d j |jd} ||j| |j!<t| | j|j d <|j"r(||j#|j |j$<dSdS) N) transformedincludes_fixedrr(r'rrkrrr$)%Z handle_paramsrDr[ryrKrCrtr,rsrrJrSr^r@rvrlrXrBrPrQstopnanrYr;rZr<rqrTr6r\rUrRrFrVr7r]rW) r1rwrrZ complex_stepr} interceptZzeror|rr"r#rr2r2r3updatest         z VARMAX.updateccs|j}|jdkrE|durBt|}|jdkr|dd}zt|dd|jf}WntyAtdt|jft|jfw||_z dVW||_dS||_w)a8 Set the final state intercept value using out-of-sample `exog` / trend Parameters ---------- exog : ndarray Out-of-sample `exog` values, usually produced by `_validate_out_of_sample_exog` to ensure the correct shape (this method does not do any additional validation of its own). out_of_sample : int Number of out-of-sample periods. Notes ----- We need special handling for simulating or forecasting with `exog` or trend, because if we had these then the last predicted_state has been set to NaN since we did not have the appropriate `exog` to create it. Since we handle trend in the same way as `exog`, we still have this issue when only trend is used without `exog`. rNr!rzPProvided exogenous values are not of the appropriate shape. Required %s, got %s.) r^rCr,Z atleast_1dndimryr=strrF)r1r cache_valuer2r2r3_set_final_exogs(      zVARMAX._set_final_exogc  s\||tj||f||||||| | | | d | }Wd|S1s'wY|S)N) measurement_shocks state_shocks initial_stateanchor repetitionsr extend_model extend_kwargsrr)rrHsimulate)r1rw nsimulationsrrrrrrrrrrrboutrdr2r3r)s  zVARMAX.simulate)NrrrFTTrr*)TFF) NNNNNNNNTF)__name__ __module__ __qualname____doc__rIrfpropertyrjrrrrrr contextlibcontextmanagerrrr r __classcell__r2r2rdr3rs<d 4   ?S M C  (rcseZdZdZ  dfdd ZdddZejddZejd d Z e e j j dfd d Z e e j j    dfdd Z dddZe e jjdfdd ZZS)rha Class to hold results from fitting an VARMAX model. Parameters ---------- model : VARMAX instance The fitted model instance Attributes ---------- specification : dictionary Dictionary including all attributes from the VARMAX model instance. coefficient_matrices_var : ndarray Array containing autoregressive lag polynomial coefficient matrices, ordered from lowest degree to highest. coefficient_matrices_vma : ndarray Array containing moving average lag polynomial coefficients, ordered from lowest degree to highest. See Also -------- statsmodels.tsa.statespace.kalman_filter.FilterResults statsmodels.tsa.statespace.mlemodel.MLEResults Nc  s$tj|||||fi|tdi|jj|jj|jj|jj|jj|jj |jj |jj |jj |jj |jjd |_d|_d|_|jj dkrft|j|jj}|jj}|jj } ||| |j||| j|_|jj dkrt|j|jj} |jj}|jj } | || |j||| j|_dSdS)N) r6r7r8r9r>r5r;r<r r@rCrr2)rHrIrmodelr6r7r8r9r>r5r;r<r r@rC specificationZcoefficient_matrices_varZcoefficient_matrices_vmar,rvrwrYrKryrtrZ) r1rrwfilter_resultsZcov_typeZcov_kwdsrbr~rKr;rr<rdr2r3rISsF      zVARMAXResults.__init__c Ks|dur |j|j|j|ddd}|j}|jd}|jd}n |jd}|jd}|d|j|jj|jj|fd|i|}t |j d||d|j _ |j durY||j} | S||j} | S) Nr)r).r).r'r>rZknownconstantstationary_cov)get_predictionrOZprediction_resultspredicted_statepredicted_state_covrGrr>rfrrrJrsmoother_resultsZsmoothrwfilter) r1rarrbZfcastZ fcast_resultsrZinitial_state_covmodresr2r2r3extends$       zVARMAXResults.extendc cs|j}||G|jjdddf}||j|dd|jdf|jjd|jdf<zdVW||jjdddf<n ||jjdddf<wWddS1sSwYdS)az Set the final state intercept value using out-of-sample `exog` / trend Parameters ---------- exog : ndarray Out-of-sample `exog` values, usually produced by `_validate_out_of_sample_exog` to ensure the correct shape (this method does not do any additional validation of its own). out_of_sample : int Number of out-of-sample periods. Notes ----- This context manager calls the model-level context manager and additionally updates the last element of filter_results.state_intercept appropriately. Nr'r()rrrr(rrwrK)r1rrrr2r2r3rs  ."zVARMAXResults._set_final_exogc csN|o|jjdk}|rt|jjddtd|jjfg}|jjdkr5t|jjdd|ddg}nd}|jj|j d}|jj |||d}|j j dddf}|j j dddddf} |jj|| d|j|jdddd } | j dddf|j j dddf<zdVW|rtj|j j dddf<dSdS|rtj|j j dddf<w) a Set the final predicted state value using out-of-sample `exog` / trend Parameters ---------- exog : ndarray Out-of-sample `exog` values, usually produced by `_validate_out_of_sample_exog` to ensure the correct shape (this method does not do any additional validation of its own). out_of_sample : int Number of out-of-sample periods. Notes ----- We need special handling for forecasting with `exog`, because if we had these then the last predicted_state has been set to NaN since we did not have the appropriate `exog` to create it. rr'Nr)rr>rT)rrZ return_ssm)rrCrrar,rRrKrr>rOrfrrrrJZinitialize_knownrrwr) r1r out_of_sampleflagZ tmp_endogZtmp_exogZtmp_trend_offsetZtmp_modrrZtmp_resr2r2r3_set_final_predicted_states: "  z(VARMAXResults._set_final_predicted_stateF predictedc s|durd}|jj|||dd\}} } } |j|| }i} |jjdkr,|jj|j| d<||8||| tj d||||||| d|} Wdn1sTwYWd| SWd| S1slwY| S)NrT)Zsilentr>)startenddynamicinformation_setindexrrr2) rZ_get_prediction_index_validate_out_of_sample_exogr@r>rOrrrHr)r1rrrrrrrbZ_startZ_endrrrrrdr2r3rs4      zVARMAXResults.get_predictionc  s|dus|dkr d} n|dkr|j} n |j|\} } } | dkr%|j| } | |jkr.tdt| ||jd} |j|| }||| tj|f|||||||| d| }Wd|S1sdwY|S)Nrrrz4Cannot anchor simulation after the estimated sample.)rrrrrrrr) rOrZ_get_index_locr=rErrrHr)r1rrrrrrrrrrbZilocrrrrdr2r3rs4   zVARMAXResults.simulatec Csd}|j|j}|jjdkr|dkr|jj| d}t'} | |j|| ||||j j |j ||||d} Wd| S1sIwY| S)Nr)rrrevisions_details_start state_index) rOrrCrr ExitStack enter_contextrrrZnews) r1ZpreviousrrZperiodsrrrrstackrr2r2r3_news_previous_results(s$     z$VARMAXResults._news_previous_results皙?Tc s^ddlm|j}|jdkr|jdkrd}d|j|j}n|jdkr+d}d|j}nd}d|j}|jdkr;|d7}||g}|jdkrJ|d |j rR|d t j ||| d }|rt t|j}dfd d } |jj} |jj} |jj} |jj} |jj}g}t| D]}g}d}| dkr|t ||| | | || | 7}| dkr|| | }|d| | }||t |||| | d7}| dkr|| | }|d| | }||t |||| | d7}|dkr||t |||d||| |7}|jj r#|t j|jj|dddt |}|||jj}t|ts:|g}d||}| |||}|j|qt t|j|jj}| ||ddd}|j|g}||gfD]}t |}t|dkr||qnt |}t tt|t|}t|dkr| ||ddd}|j||S)Nr)summary_paramsZVARMAz({},{})ruz(%s)ZVMAXrzmeasurement error)alphar model_nameZdisplay_paramsTcs||j||j||j||j|||f}g}t|jj| D]!}|r7d | ddd}n|}||j vrBd|}| |q&|d|d|dS)N.r'z %s (fixed)F)ZynameZxnamerZuse_ttitle)rwZbseZzvaluesZpvaluesZconf_intr,rvdatarrjoinsplitZ fixed_paramsappend)r1r{r strip_endrrnameZ param_namerrr2r3 make_tablefs   z)VARMAXResults.summary..make_tablerr!rzResults for equation %szError covariance matrixF)rzOther parameters)T)Zstatsmodels.iolib.summaryrrr;r<rrCr@rr7rHsummaryr,ZarangerrwrrKrrvrNZ concatenaterrr_Ztablesr\Zflattenset difference)r1rrZseparate_paramsspecrr5rindicesrrKr;r<r@rCZ endog_masksrZmasksr&rr{rrtableZstate_cov_maskmZ inverse_maskrdrr3rEs                         zVARMAXResults.summary)NNr*)NNFrNN)NNNNNNNN)FN)rNT)rrrrrIrrrrrrr rrrrrr2r2rdr3rh:s2 ,   0 ! rhc@s0eZdZiZeejeZiZeej eZ dS)riN) rrr_attrswrapZ union_dictsr Z _wrap_attrsZ_methodsZ _wrap_methodsr2r2r2r3risri)+rrwarningsrZpandasroZnumpyr,Zstatsmodels.compat.pandasrZstatsmodels.tools.toolsrZstatsmodels.tools.datarZstatsmodels.tsa.vector_arrZstatsmodels.base.wrapperbasewrapperrZstatsmodels.tools.sm_exceptionsrZ kalman_filterr r Zmlemodelr r r rrZtoolsrrrrrrrrrhriZpopulate_wrapperr2r2r2r3s8       $!