Glossary

The 97 terms Time Series Forecasting defines, each in the book's own words, with the section that defines it.

From Edition 2026.11.11, the book's current edition. Where a definition depends on a package, it says the version this edition pins and runs its examples on.

A

AICc also Akaike information criterion corrected for small samples

The criterion AutoETS uses is AICc, the Akaike information criterion corrected for small samples.

Defined in Forecast the next day with one model, chapter 1. In the book since Edition 2026.11.11. Permanent link to AICc

alpha

The constant alpha, between 0 and 1, sets how fast the level follows the data.

Defined in Fit exponential smoothing, chapter 4. In the book since Edition 2026.11.11. Permanent link to alpha

ARIMA

An ARIMA model writes the next reading as a weighted sum of earlier readings and earlier errors, after differencing

ARIMA(p,d,q)(P,D,Q)[m]

Defined in Fit a fixed seasonal ARIMA, chapter 4. In the book since Edition 2026.11.11. Permanent link to ARIMA

AutoARIMA

AutoARIMA searches for the orders instead of fixing them.

True for statsforecast 2.1.1, as this edition pins it.

  • Implemented by: statsforecast 2.1.1

Defined in Why the book avoids automatic ARIMA at season 24, chapter 4. In the book since Edition 2026.11.11. Permanent link to AutoARIMA

AutoETS

statsforecast 2.1.1 describes AutoETS as selecting the best ETS (error, trend and seasonality) form with an information criterion (statsforecast/models.py, class AutoETS).

True for statsforecast 2.1.1, as this edition pins it.

  • Implemented by: statsforecast 2.1.1

Defined in Forecast the next day with one model, chapter 1. In the book since Edition 2026.11.11. Permanent link to AutoETS

average of weeks

Average of weeks takes the mean of the same hour on the same weekday over the four weeks before

forecast(T + k) = (y(T + k - 168) + y(T + k - 336) + y(T + k - 504) + y(T + k - 672)) / 4

Defined in Three baselines to beat, chapter 2. In the book since Edition 2026.11.11. Permanent link to average of weeks

B

backtest

a backtest is the score of many windows, each made as it would have been made on its day

Defined in Why a baseline comes first, chapter 2. In the book since Edition 2026.11.11. Permanent link to backtest

baseline

A baseline is a forecast that needs no fitting.

Defined in Why a baseline comes first, chapter 2. In the book since Edition 2026.11.11. Permanent link to baseline

bottom-up also bottom-up total

Bottom-up forecasts the bottom level, here the clients, and sums the results.

Defined in Why the total needs a check, chapter 7. In the book since Edition 2026.11.11. Permanent link to bottom-up

C

calendar feature also calendar features

Calendar features, unlike lags, are facts about the target hour itself: its hour of day, its day of the week and its month.

Defined in Add calendar and holiday features, chapter 5. In the book since Edition 2026.11.11. Permanent link to calendar feature

calibration window also calibration windows

A calibration window is a cutoff whose residuals size the bands.

Defined in Build intervals from residuals, chapter 8. In the book since Edition 2026.11.11. Permanent link to calibration window

Chronos-2

Chronos-2 is an encoder-only transformer that takes a context of past values and outputs quantiles for the next steps

Defined in Load a recorded foundation model's forecasts, chapter 9. In the book since Edition 2026.11.11. Permanent link to Chronos-2

client identifier

A client identifier puts the client in the table as a static categorical feature.

Defined in Identify each client, chapter 6. In the book since Edition 2026.11.11. Permanent link to client identifier

coherent also coherence

Forecasts that add up are called coherent.

Defined in Why the total needs a check, chapter 7. In the book since Edition 2026.11.11. Permanent link to coherent

conformal prediction

Conformal prediction builds a band from the residuals of forecasts made the same way on earlier days.

Defined in Build intervals from residuals, chapter 8. In the book since Edition 2026.11.11. Permanent link to conformal prediction

coverage

The share of readings the bounds contain is the coverage

Defined in Why one number is not enough, chapter 8. In the book since Edition 2026.11.11. Permanent link to coverage

coverage gap

Barber and co-authors (2023) call the loss in coverage when exchangeability fails a coverage gap.

Defined in Measure coverage on held-out days, chapter 8. In the book since Edition 2026.11.11. Permanent link to coverage gap

critical fractile

The level cu / (cu + co) is the critical fractile of the newsvendor model.

tau = cu / (cu + co)

Defined in Choose the level from the cost of error, chapter 8. In the book since Edition 2026.11.11. Permanent link to critical fractile

cross-learning

Cross-learning, which would let one series inform another, is off by default

Defined in Load a recorded foundation model's forecasts, chapter 9. In the book since Edition 2026.11.11. Permanent link to cross-learning

cutoff

The moment a forecast is made is its cutoff.

Defined in The case and its data, chapter 1. In the book since Edition 2026.11.11. Permanent link to cutoff

D

daily MASE

The daily MASE takes each client's mean absolute error over the day's 24 hours, divides it by the client's scale, and averages the results over the thirty clients

daily MASE = mean over clients of (MAE over the day's 24 hours / scale)

Defined in Score the log as actuals arrive, chapter 10. In the book since Edition 2026.11.11. Permanent link to daily MASE

decision window also decision windows

Decision windows are the windows the feature set is chosen on, never the standard cutoffs, which are the windows the book reports.

Defined in Choose on decision windows, chapter 5. In the book since Edition 2026.11.11. Permanent link to decision window

differencing

differencing: subtracting a reading from the reading one season earlier

Defined in Fit a fixed seasonal ARIMA, chapter 4. In the book since Edition 2026.11.11. Permanent link to differencing

direct total

The direct total is one model fitted to the total series itself.

Defined in Forecast the total directly, chapter 7. In the book since Edition 2026.11.11. Permanent link to direct total

E

error

The error of an hour is the actual demand minus the forecast.

Defined in Score both on a held-out week, chapter 1. In the book since Edition 2026.11.11. Permanent link to error

ETS

ETS, for error, trend and season, extends the simplest form of exponential smoothing.

Defined in Fit exponential smoothing, chapter 4. In the book since Edition 2026.11.11. Permanent link to ETS

exponential smoothing

Exponential smoothing moves its level toward each new reading.

level_t = alpha * y_t + (1 - alpha) * level_(t-1)

  • Comes before: ETS

Defined in Fit exponential smoothing, chapter 4. In the book since Edition 2026.11.11. Permanent link to exponential smoothing

F

feature also features

the features: the readings and facts the model may use for the row's hour.

Defined in Turn history into features, chapter 5. In the book since Edition 2026.11.11. Permanent link to feature

feature table

A feature table has one row for each hour a model must predict.

Defined in Turn history into features, chapter 5. In the book since Edition 2026.11.11. Permanent link to feature table

fill

The fill is the mean of the hour before and the hour after, taken from the history before the cutoff only.

Defined in Mark clock-change hours as missing, chapter 3. In the book since Edition 2026.11.11. Permanent link to fill

finite-sample levels also finite-sample correction

Each new residual is one more among n + 1, so the finite-sample levels move outward a little: each quantile level moves to floor((n + 1) * a) / n for the lower bound and ceil((n + 1) * (1 - a)) / n for the upper, where n is the number of residuals and a the tail share.

Defined in Calibrate on earlier windows, chapter 8. In the book since Edition 2026.11.11. Permanent link to finite-sample levels

flag

A day is flagged when its daily MASE is above twice a reference.

Defined in Flag drift against the backtest, chapter 10. In the book since Edition 2026.11.11. Permanent link to flag

forecast

A forecast here is the expected demand of one client in each hour, made from readings recorded before the forecast was made.

Defined in The case and its data, chapter 1. In the book since Edition 2026.11.11. Permanent link to forecast

forecast log

A forecast log is a table with one row for each client and hour.

Defined in Log every forecast before its actuals arrive, chapter 10. In the book since Edition 2026.11.11. Permanent link to forecast log

forecast row also forecast rows

Forecast rows are the 24 hours after the cutoff, with features but no target.

Defined in Turn history into features, chapter 5. In the book since Edition 2026.11.11. Permanent link to forecast row

forward fill

Forward fill repeats the last reading that arrived.

Defined in Score the log as actuals arrive, chapter 10. In the book since Edition 2026.11.11. Permanent link to forward fill

foundation model

A foundation model is trained once on many series and then applied to a context, with no fitting on that series.

Defined in Load a recorded foundation model's forecasts, chapter 9. In the book since Edition 2026.11.11. Permanent link to foundation model

G

gap hour also clock_change_gap

clock_change_gap is 01:00 on a March clock-change day.

Defined in Load one client's hourly demand, chapter 1. In the book since Edition 2026.11.11. Permanent link to gap hour

global model also global models, global method, global methods

A global model fits one set of parameters to many series in a single pass, so every client contributes to what it learns.

Defined in One model for many clients, chapter 6. In the book since Edition 2026.11.11. Permanent link to global model

H

history

the history is every reading before the cutoff

Defined in Why a baseline comes first, chapter 2. In the book since Edition 2026.11.11. Permanent link to history

history half

The history half compares each feature on the rows of the eight weeks before each of the three evaluation cutoffs

Defined in Check a backtest for leakage, chapter 3. In the book since Edition 2026.11.11. Permanent link to history half

holiday table

Prophet takes holidays as a table of dates and effects

True for prophet 1.5.0, as this edition pins it.

Defined in Where Prophet fits, chapter 9. In the book since Edition 2026.11.11. Permanent link to holiday table

horizon

A forecast uses only the readings before its cutoff, and it reaches forward a fixed number of hours, its horizon.

Defined in The case and its data, chapter 1. In the book since Edition 2026.11.11. Permanent link to horizon

I

interval score

The interval score adds a band's width to a penalty for each reading that misses the band, the penalty growing with the distance of the miss and scaled by 2 over the miss rate

Defined in Calibrate on earlier windows, chapter 8. In the book since Edition 2026.11.11. Permanent link to interval score

L

lag also lags, lag feature

The lag of k hours for the hour t is the reading at t minus k.

Defined in Add lags the forecast can see, chapter 5. In the book since Edition 2026.11.11. Permanent link to lag

late starter

A late starter is a client whose first reading falls inside the history its model reads.

Defined in Meet the late starters, chapter 4. In the book since Edition 2026.11.11. Permanent link to late starter

leakage also leak

A backtest leaks when a window's model sees a reading from the hours it is asked to forecast.

Defined in Keep the future out of the history, chapter 3. In the book since Edition 2026.11.11. Permanent link to leakage

level

The level of a series is its typical reading

Defined in Why a statistical model comes next, chapter 4. In the book since Edition 2026.11.11. Permanent link to level

level

A prediction interval's level, such as 80 or 95, is the share of readings the bounds should contain.

Defined in Why one number is not enough, chapter 8. In the book since Edition 2026.11.11. Permanent link to level

local method also local methods

local methods, which fit each series on its own

Defined in One model for many clients, chapter 6. In the book since Edition 2026.11.11. Permanent link to local method

LocalStandardScaler

mlforecast's LocalStandardScaler (mlforecast/target_transforms.py), which subtracts the client's mean and divides by its standard deviation.

True for mlforecast 1.1.0, as this edition pins it.

  • Implemented by: mlforecast 1.1.0

Defined in Measure what each feature adds, chapter 5. In the book since Edition 2026.11.11. Permanent link to LocalStandardScaler

long format

The data is held in long format: one row per client and hour, with three columns.

Defined in One model for many clients, chapter 6. In the book since Edition 2026.11.11. Permanent link to long format

M

masked reading

A masked reading is neither zero nor a copy; it is missing.

Defined in Take the MASE scale from history, chapter 2. In the book since Edition 2026.11.11. Permanent link to masked reading

maximum a posteriori estimate

By default Prophet's fit is a maximum a posteriori estimate, an optimisation rather than a set of sampled draws.

True for prophet 1.5.0, as this edition pins it.

Defined in Where Prophet fits, chapter 9. In the book since Edition 2026.11.11. Permanent link to maximum a posteriori estimate

mean absolute error also MAE

The mean absolute error, MAE, is the average size of the miss

MAE = mean(abs(actual - forecast))

  • A kind of: score
  • Computed from: error
  • Implemented by: utilsforecast 0.2.17

Defined in Score both on a held-out week, chapter 1. In the book since Edition 2026.11.11. Permanent link to mean absolute error

mean absolute scaled error also MASE

MASE, the mean absolute scaled error, is the mean of the scaled errors.

MASE = mean of |actual(t) - forecast(t)| / scale

  • Computed from: scaled error
  • Implemented by: utilsforecast 0.2.17

Defined in Scale errors with MASE, chapter 2. In the book since Edition 2026.11.11. Permanent link to mean absolute scaled error

mean per-client MASE

The mean per-client MASE is MASE for each client, averaged over the eight clients.

Defined in Measure what each feature adds, chapter 5. In the book since Edition 2026.11.11. Permanent link to mean per-client MASE

MinT

MinT chooses the coherent forecasts that minimise the mean squared error of the coherent forecasts, under an unbiasedness assumption

Defined in Why the total needs a check, chapter 7. In the book since Edition 2026.11.11. Permanent link to MinT

model string

The model string sets which forms the model may take, with one letter each for error, trend and season.

True for statsforecast 2.1.1, as this edition pins it.

Defined in Forecast the next day with one model, chapter 1. In the book since Edition 2026.11.11. Permanent link to model string

monitoring system

A monitoring system covers alerts, dashboards, data quality across many feeds and the people who respond.

Defined in Know when to retire a model, chapter 10. In the book since Edition 2026.11.11. Permanent link to monitoring system

N

naive

Naive repeats the last reading before the cutoff

forecast(T + k) = y(T - 1)

  • A kind of: baseline
  • Implemented by: statsforecast 2.1.1

Defined in Three baselines to beat, chapter 2. In the book since Edition 2026.11.11. Permanent link to naive

NHITS

NHITS is a neural forecaster from the neuralforecast library, which packages neural forecasters for panels of series.

Defined in Why the neural model is recorded, chapter 9. In the book since Edition 2026.11.11. Permanent link to NHITS

O

one-step naive error also one-step scale

The 2006 paper scales by the one-step naive error: the mean absolute difference between each reading and the one before it.

Defined in Scale errors with MASE, chapter 2. In the book since Edition 2026.11.11. Permanent link to one-step naive error

oracle check

A check this chapter calls the oracle check applies the shares to the actual total, which no forecast can know

Defined in Split a total by share, chapter 7. In the book since Edition 2026.11.11. Permanent link to oracle check

P

per-client model

A per-client model is one fit for each client, on that client's history alone, so the cost is one fit per client per window.

Defined in Why a statistical model comes next, chapter 4. In the book since Edition 2026.11.11. Permanent link to per-client model

poisoning check

The poisoning check writes wrong values over every reading after a cutoff, then confirms that the refit and the forecast at that cutoff do not move.

Defined in What goes wrong in production, chapter 10. In the book since Edition 2026.11.11. Permanent link to poisoning check

pooled MAE

The pooled MAE is the MAE over every client-hour in the windows, the single number for the whole supplier.

Defined in Measure what each feature adds, chapter 5. In the book since Edition 2026.11.11. Permanent link to pooled MAE

prediction interval also band

A prediction interval gives a lower and an upper bound for each hour.

Defined in Why one number is not enough, chapter 8. In the book since Edition 2026.11.11. Permanent link to prediction interval

Prophet

Prophet writes a series as a trend, plus Fourier seasonal terms, plus holiday effects, plus an error term.

True for prophet 1.5.0, as this edition pins it.

  • Implemented by: prophet 1.5.0

Defined in Where Prophet fits, chapter 9. In the book since Edition 2026.11.11. Permanent link to Prophet

Q

quantile loss also pinball loss

The quantile loss is tau * (y - q) when the reading y is at or above the purchase q, and (1 - tau) * (q - y) when it is below.

Defined in Choose the level from the cost of error, chapter 8. In the book since Edition 2026.11.11. Permanent link to quantile loss

R

rebuilt log

A rebuilt log is leakage in disguise, because it knows what the forecast should have been.

Defined in Log every forecast before its actuals arrive, chapter 10. In the book since Edition 2026.11.11. Permanent link to rebuilt log

recursive prediction

In recursive prediction, for a 24-hour horizon, MLForecast forecasts one hour, writes that forecast back as the reading for the next hour, and repeats (_predict_recursive in mlforecast/core.py).

True for mlforecast 1.1.0, as this edition pins it.

  • Implemented by: mlforecast 1.1.0

Defined in Catch a lag that leaks the future, chapter 5. In the book since Edition 2026.11.11. Permanent link to recursive prediction

refit

A refit re-estimates the model's parameters on the history known at a cutoff.

Defined in Refit on a schedule, chapter 10. In the book since Edition 2026.11.11. Permanent link to refit

repeat hour also clock_change_repeat

clock_change_repeat is 01:00 on an October clock-change day.

Defined in Load one client's hourly demand, chapter 1. In the book since Edition 2026.11.11. Permanent link to repeat hour

residual

A residual is the error of an earlier forecast: the reading minus the forecast, e = y - f.

e = y - f

Defined in Build intervals from residuals, chapter 8. In the book since Edition 2026.11.11. Permanent link to residual

retiring a model also retire a model

Retiring a model means stopping its use and replacing it.

Defined in Know when to retire a model, chapter 10. In the book since Edition 2026.11.11. Permanent link to retiring a model

root mean squared error also RMSE

The root mean squared error, RMSE, is RMSE = sqrt(mean((actual - forecast)^2)).

RMSE = sqrt(mean((actual - forecast)^2))

Defined in Score both on a held-out week, chapter 1. In the book since Edition 2026.11.11. Permanent link to root mean squared error

S

scaled error

A scaled error divides each absolute error by a scale, the typical error of a simple method on the same client's own history.

Defined in Scale errors with MASE, chapter 2. In the book since Edition 2026.11.11. Permanent link to scaled error

score

A score turns the errors of a set of forecasts into one number.

Defined in Score both on a held-out week, chapter 1. In the book since Edition 2026.11.11. Permanent link to score

season

the season is the daily shape of the load

Defined in Why a statistical model comes next, chapter 4. In the book since Edition 2026.11.11. Permanent link to season

seasonal naive also SeasonalNaive

Seasonal naive says that each future value equals the last observed value from the same season.

y(t - 24)

  • A kind of: baseline
  • Implemented by: statsforecast 2.1.1

Defined in Forecast the next day with seasonal naive, chapter 1. In the book since Edition 2026.11.11. Permanent link to seasonal naive

seasonal scale

under the seasonal scale, the denominator is seasonal naive's error on the client's own history

scale = mean of |y(t) - y(t - 24)|

Defined in Scale errors with MASE, chapter 2. In the book since Edition 2026.11.11. Permanent link to seasonal scale

share

A client's share is its readings over its own history, the 1,344 hours before the cutoff, divided by the total's readings over the same hours.

Defined in Split a total by share, chapter 7. In the book since Edition 2026.11.11. Permanent link to share

shift test

The shift test compares a feature computed from the history cut at the cutoff with the same feature computed from the whole series.

Defined in Check a backtest for leakage, chapter 3. In the book since Edition 2026.11.11. Permanent link to shift test

start rule

The start rule says that a series begins at its first reading.

Defined in Start each series at its first reading, chapter 4. In the book since Edition 2026.11.11. Permanent link to start rule

statistical model

A statistical model holds an estimate of each part of the state of a series, updates the estimates as each reading arrives, and projects them over the horizon.

Defined in Why a statistical model comes next, chapter 4. In the book since Edition 2026.11.11. Permanent link to statistical model

step back operator

B the step back operator, so that B y_t is y_(t-1)

Defined in Fit a fixed seasonal ARIMA, chapter 4. In the book since Edition 2026.11.11. Permanent link to step back operator

sustained drift

The first retirement trigger is sustained drift: flags on consecutive days with no calendar cause.

Defined in Know when to retire a model, chapter 10. In the book since Edition 2026.11.11. Permanent link to sustained drift

T

target

Each row of a feature table holds the value to predict, which the book calls the target

Defined in Turn history into features, chapter 5. In the book since Edition 2026.11.11. Permanent link to target

target half

The target half compares the 24 hours that each cutoff forecasts, computed from the readings before the cutoff only.

Defined in Check a backtest for leakage, chapter 3. In the book since Edition 2026.11.11. Permanent link to target half

time series

A time series is a sequence of readings in time order.

Defined in Load one client's hourly demand, chapter 1. In the book since Edition 2026.11.11. Permanent link to time series

top-down also top-down split

Top-down forecasts the top, the total, and splits it down by proportions.

Defined in Why the total needs a check, chapter 7. In the book since Edition 2026.11.11. Permanent link to top-down

trailing mean

a trailing mean: the average of the 24 hours up to and including each hour, which reads nothing after its hour

Defined in Check a backtest for leakage, chapter 3. In the book since Edition 2026.11.11. Permanent link to trailing mean

training row also training rows

Training rows are the hours of the history before the cutoff, each with its own features and its own reading as the target.

Defined in Turn history into features, chapter 5. In the book since Edition 2026.11.11. Permanent link to training row

trend

the trend is the slow drift of the level

Defined in Why a statistical model comes next, chapter 4. In the book since Edition 2026.11.11. Permanent link to trend

U

uncovered band

The second retirement trigger is an uncovered band: the 80% band covers well under 80% of readings on several days.

Defined in Know when to retire a model, chapter 10. In the book since Edition 2026.11.11. Permanent link to uncovered band

W

window

A window is one cutoff with its horizon and the forecast it produces.

Defined in The case and its data, chapter 1. In the book since Edition 2026.11.11. Permanent link to window