Time Series User Guide

StochX is primarily designed to make practical time-series analysis feel familiar to EViews users while remaining a native Python library.

The time-series guide follows the workflow most users already know:

1. load or create a workfile;
2. inspect and describe the data;
3. generate transformations and lags;
4. decompose or smooth the series when appropriate;
5. test stationarity and identify deterministic terms;
6. inspect ACF/PACF and the correlogram;
7. estimate AR, MA, ARMA, ARIMA and SARIMA models;
8. validate residuals and model assumptions;
9. compare candidate models with transparent selection rules;
10. forecast and report prediction intervals.

The same sequence also leads naturally to the linear state-space and Kalman workflow.

Why StochX for an EViews user?

StochX deliberately uses familiar EViews concepts and names where the underlying operation is implemented: workfiles, samples, series expressions, lags, differences, C, @TREND, equations, LS estimation, ADF, AR/MA error terms, summary tables, interpretation helpers, roots reports, and forecast reports.

This means that an EViews workflow can be translated into Python without replacing the user's statistical methodology with a completely different framework.

The design goal is EViews-inspired workflow compatibility, not a hidden black-box wrapper. The implementation is Python-native, testable, scriptable, and reproducible. Where StochX includes published or frozen numerical benchmarks, those are treated as explicit regression contracts.

A complete workflow in one page

import numpy as np
from stochx.timeseries import (
    Workfile,
    adf,
    correlogram,
    identify_box_jenkins,
    estimate_box_jenkins_candidates,
    validate_box_jenkins_candidates,
    select_box_jenkins_model,
    forecast_box_jenkins,
)

wf = Workfile.from_dataframe(dataframe, frequency="M")
wf.set_sample(0, wf.nobs - 1)

series = wf["Y"]
print(series.describe())

unit_root = adf(series, regression="c", lags=1, autolag=None)
print(unit_root.summary())

corr = correlogram(series.values, nlags=24)
print(corr.table())

ident = identify_box_jenkins(series, d=1, nlags=24, max_p=3, max_q=3)
estimation = estimate_box_jenkins_candidates(series, ident.candidate_orders)
validation = validate_box_jenkins_candidates(estimation, lags=12, alpha=0.05)
selection = select_box_jenkins_model(validation)
forecast = forecast_box_jenkins(selection, steps=12, alpha=0.05)

The purpose of each step is explained in the pages below rather than hidden inside one convenience function.

Reading the guide

Each method page follows the same structure:

• What it does — practical goal and when to use it.
• Mathematical idea — model, statistic, or hypothesis being computed.
• Syntax — the StochX call and its important parameters.
• Worked example — a small runnable example.
• Reading the output — how to interpret tables and diagnostics.
• Failure modes — common mistakes and what StochX reports.
• API reference — exact public objects and related functions.

This organization intentionally follows the teaching style of mature scientific-Python user guides: concept first, then implementation details, examples, output interpretation and references rather than a list of signatures without context.

Time-series map

Design principles

StochX time-series functionality is built around five principles:

1. Course-faithful terminology. Definitions, notation, hypotheses and decision rules remain explicit.
2. EViews familiarity. Common commands and expressions are intentionally recognizable.
3. Report-oriented output. Results expose statistics, tables, diagnostics, coefficients and interpretations rather than returning an opaque model object alone.
4. Deterministic behavior. Frozen fixtures protect numerical conventions and regression stability.
5. Composable Python. Every step can be called independently, inspected, tested and embedded in a larger workflow.