ETC3450 Chap.1 Time-Ordered Data and Stochastic Processes

Time-Ordered Data and Stochastic Processes

A time series records the same kind of quantity at ordered points. Frequency may be hourly, monthly, quarterly or irregular, but the ordering permits dependence, trend, seasonality and response delays. Shuffling preserves the marginal values while destroying the lag relationships that make dynamic modelling possible. A monthly average cannot identify hourly peaks, and an annual series cannot reveal within-year seasonality.

Aggregation can smooth volatility or change apparent dependence. State the observational clock, timestamp convention and missing-time rule before fitting a stochastic process. Pedestrian counts sampled hourly reveal a daily rhythm that vanishes in a daily total. Both datasets describe traffic, yet they support different forecasts and interventions because their indexes retain different temporal information.

Check for duplicate times, gaps, irregular spacing and changes in measurement before computing correlations. A lag means a number of index steps; its calendar duration changes when frequency changes. Econometric interpretation begins by fixing the stochastic object behind time index: its conditioning information, time index and maintained assumptions.

Derive the implication for frequency before inserting estimated values, because a numerical answer can conceal a wrong model. The observed path is a finite realisation. A stochastic process is a collection of random variables indexed by time and describes possible paths through its joint distribution. Data summaries describe what occurred; model statements encode assumptions about what could occur and how observations depend.

A smooth plot does not prove a deterministic trend, and a model equation does not guarantee fit. Econometric reasoning moves from pattern to candidate process, derives implications, then returns to data for diagnostics and comparison. One random-walk realisation may drift upward for most of the sample even though each innovation has zero mean.

Reading the path as evidence of a positive deterministic trend confuses realised accumulation with the process expectation. Write model implications before examining the matching statistic. This prevents an attractive plot from deciding the model. Record which discrepancies would reject or revise the candidate. A time-series calculation should be accompanied by a probability statement.

For realisation, separate a realised path from the process that could generate many paths. State which moments are stable and how stochastic process changes with the lag. When the result depends on implication, identify the parameter region and initial condition explicitly. The limiting case often reveals whether the claimed long-run interpretation is mathematically coherent. Turn realisation into a diagnostic sequence.

Plot the series and transformations, inspect dependence over several lags, estimate a defensible candidate and then examine the residuals for structure left behind. A forecast of the next observation is conditional on an information set available at the forecast origin. Including the target or later values creates look-ahead bias.

The notation should make timing explicit so the conditional mean and forecast error are defined against the same information. White noise has the same conditional and unconditional mean under its defining assumptions, so past values do not improve the point forecast. Autoregression changes that result because the latest observation enters the conditional expectation.

A backtest accidentally constructs a lag after sorting data in reverse order. The apparent predictor contains future values and produces unrealistically low error. Restoring chronological order removes the artificial performance. Label origin, horizon and target beside every forecast. Then verify each predictor's timestamp. A clean formula cannot rescue an invalid information set.

Forecast reasoning distinguishes an information set from future innovations. Write the projection for information set, set the conditional mean of the new disturbance to zero only when the assumptions justify it, and propagate forecast origin recursively. Uncertainty usually accumulates differently from the point forecast.

Use look-ahead bias to explain what the interval covers and why a narrow interval cannot rescue a misspecified dynamic model. Compare two specifications for information set using the same estimation sample and forecast origin. Ask whether adding forecast origin removes a residual pattern, changes uncertainty materially or merely improves in-sample fit.

A time plot can reveal trend, changing variance, breaks, seasonality, outliers and persistence. These features suggest transformations and model families. They do not uniquely identify a process, because different mechanisms can produce visually similar finite paths. A short window can hide mean reversion; a long window can compress local shocks. Logarithms, differences and aggregation answer different questions.

Keep the original scale available and explain what each transformation removes or preserves. A price index rises steadily in levels while its first differences fluctuate around a stable mean. The two plots support different modelling objects: a persistent level and a potentially stationary change. Pair visual diagnosis with a stated implication and a statistic able to challenge it.

Mark unusual periods instead of deleting them silently, since an intervention or measurement change may explain the break. Model comparison should not reduce to selecting the largest fit statistic. Ask whether time plot leaves serial structure in the residuals, whether transformation is parsimonious enough for the sample, and whether parameters remain stable over the forecast origin.

Interpret structural break against a substantive dynamic mechanism. If rival models perform similarly, report that uncertainty and prefer a specification whose assumptions can be checked transparently. Separate three uncertainty layers around time plot: future innovations, estimated parameters and uncertainty about the model form itself.

A standard conditional forecast may represent the first while treating transformation and the specification as fixed.

What this chapter covers

• 01

  Time order creates statistical structure

• 02

  Data land and model land answer different questions

• 03

  The information set controls a forecast

• 04

  Plots diagnose candidates rather than prove models

A worked test of Time order creates statistical structure

Key terms

Time-series dependence structure

Time order creates statistical structure — A time series records the same kind of quantity at ordered points. Frequency may be hourly, monthly, quarterly or irregular, but the ordering permits dependence, trend, seasonality and response delays. Shuffling preserves the marginal values while destroying the lag relationships that make dynamic modelling possible. Check for duplicate times, gaps, irregular spacing and changes in measurement before computing correlations. A lag means a number of index steps; its calendar duration changes when frequency changes.

Observations and stochastic models

Data land and model land answer different questions — The observed path is a finite realisation. A stochastic process is a collection of random variables indexed by time and describes possible paths through its joint distribution. Data summaries describe what occurred; model statements encode assumptions about what could occur and how observations depend. Write model implications before examining the matching statistic. This prevents an attractive plot from deciding the model. Record which discrepancies would reject or revise the candidate.

Forecast information set

The information set controls a forecast — A forecast of the next observation is conditional on an information set available at the forecast origin. Including the target or later values creates look-ahead bias. The notation should make timing explicit so the conditional mean and forecast error are defined against the same information. Label origin, horizon and target beside every forecast. Then verify each predictor's timestamp. A clean formula cannot rescue an invalid information set.

Time-Ordered Data and Stochastic Processes FAQ

Exam move

Keep a derivation and diagnostic sheet for Time-Ordered Data and Stochastic Processes. Write every process assumption before calculating a moment, response or forecast. Begin with time index and reconstruct the reasoning without looking at the worked response. Then change one condition in the example and decide whether frequency still explains the outcome.

Use the chapter questions to compare direct observation with inference, and write the strongest rival account in full. Before closing the chapter, return to structural break and state the precise boundary it places on transfer. Check that every conclusion names an observable consequence and that uncertainty is attached to the step it affects.

A final retrieval pass should be fast enough to reproduce the method from headings and diagrams while leaving the detailed prose for checking nuance.

ETC2410 Introductory Econometrics · ETC2420 Statistical Thinking · ETC3460 Financial Econometrics