{
  "id": "exponential-smoothing",
  "title": "Exponential Smoothing",
  "lang": "en",
  "source_lang": "en",
  "summary": "Exponential smoothing is a family of time-series forecasting methods that recursively update estimates of level, trend, and seasonality, giving progressively less influence to older observations.",
  "category": "mathematics",
  "subcategory": "probability-statistics",
  "type": "concept",
  "aliases": [
    "Exponentially weighted smoothing"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/exponential-smoothing/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/exponential-smoothing/",
      "json": "https://aiwiki.page/zh/exponential-smoothing.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Exponential smoothing is a family of methods for analyzing and forecasting a time series by recursively updating estimates of its underlying components. In its simplest form, each new observation is combined with the previous estimate of the series’ level, producing weights that decrease geometrically with the age of the observations. Extensions incorporate trend, seasonality, and damped trend. The methods are widely used for demand, sales, inventory, and operational forecasting because they require relatively little computation and can adapt as new data arrive."
    },
    {
      "heading": "Simple exponential smoothing",
      "text": "Simple exponential smoothing is appropriate for a series whose level may change over time but which has no substantial trend or seasonal pattern. Let \\(y_t\\) be the observation at time \\(t\\), and let \\(\\ell_t\\) be the estimated level after observing \\(y_t\\). The updating equation is\n\n\\[\\ell_t=\\alpha y_t+(1-\\alpha)\\ell_{t-1}, \\qquad 0\\leq\\alpha\\leq1,\\]\n\nwhere \\(\\alpha\\) is the smoothing parameter and \\(\\ell_0\\) is an initial level.\n\nThis recurrence relation can also be written as an error-correction equation:\n\n\\[\\ell_t=\\ell_{t-1}+\\alpha e_t, \\qquad e_t=y_t-\\ell_{t-1}.\\]\n\nThe new level is therefore the previous level plus a fraction of the latest one-step forecast error.\n\nRepeated substitution gives\n\n\\[\\ell_t =(1-\\alpha)^t\\ell_0 +\\alpha\\sum_{j=0}^{t-1}(1-\\alpha)^j y_{t-j}.\\]\n\nThe observation weights follow a geometric sequence. For \\(0<\\alpha<1\\), they can also be expressed using an exponential function:\n\n\\[(1-\\alpha)^j =\\exp\\!\\bigl(j\\log(1-\\alpha)\\bigr).\\]\n\nThis exponential decay gives the method its name. It does not imply that the observations themselves follow an exponential growth curve.\n\nResponsiveness and memory\n\nA larger \\(\\alpha\\) gives more weight to recent observations, allowing faster adaptation but also transmitting more short-term variation. A smaller \\(\\alpha\\) produces a smoother estimate with a longer memory, but responds more slowly to changes.\n\nAt the endpoints:\n\n- \\(\\alpha=1\\) makes the level equal to the latest observation.\n- \\(\\alpha=0\\) leaves the initial level unchanged.\n\nFor \\(0<\\alpha<1\\), the half-life of an observation’s weight, measured in observation intervals, is\n\n\\[h_{1/2}=\\frac{\\log(1/2)}{\\log(1-\\alpha)}.\\]\n\nFor example, with \\(\\alpha=0.2\\), an observation’s relative weight halves approximately every 3.1 intervals.\n\nUnlike a fixed-window moving average, exponential smoothing does not discard an observation abruptly when it reaches a particular age. Its influence decreases continuously. A recursive implementation needs only the previous level, rather than a stored window of past observations.\n\nForecasts\n\nThe forecast made at time \\(t\\) for \\(h\\) periods ahead is\n\n\\[\\widehat y_{t+h\\mid t}=\\ell_t, \\qquad h\\geq1.\\]\n\nThus, simple exponential smoothing produces a flat forecast path. The level can change when new observations arrive, but a forecast made at a fixed time contains neither an extrapolated trend nor a seasonal cycle.\n\nFor example, if \\(\\ell_{t-1}=100\\), \\(y_t=110\\), and \\(\\alpha=0.2\\), then\n\n\\[\\ell_t=0.2(110)+0.8(100)=102.\\]\n\nAll future point forecasts made at that time are 102. This does not imply that future observations will be constant; their uncertainty generally increases with the forecast horizon."
    },
    {
      "heading": "Trend methods",
      "text": "Holt’s linear method\n\nHolt’s method extends simple exponential smoothing by maintaining both a level \\(\\ell_t\\) and a trend \\(b_t\\):\n\n\\[\\ell_t =\\alpha y_t+(1-\\alpha)(\\ell_{t-1}+b_{t-1}),\\]\n\n\\[b_t =\\beta^*(\\ell_t-\\ell_{t-1}) +(1-\\beta^*)b_{t-1}.\\]\n\nHere, \\(b_t\\) estimates the change per observation interval, and \\(\\beta^*\\) controls how quickly the trend estimate adjusts. The forecast is\n\n\\[\\widehat y_{t+h\\mid t}=\\ell_t+h b_t.\\]\n\nThe method allows both level and trend to evolve, rather than fitting one fixed straight line to the entire history as in a basic linear regression model.\n\n“Double exponential smoothing” is sometimes used for Holt’s method, but it also refers to Brown’s method, which derives level and trend estimates by smoothing a series twice. These constructions are related but are not interchangeable.\n\nDamped trend\n\nA linear trend extrapolated indefinitely can produce implausibly large increases or decreases. Damped-trend exponential smoothing introduces a damping parameter \\(\\phi\\), conventionally with \\(0<\\phi<1\\):\n\n\\[\\ell_t =\\alpha y_t+(1-\\alpha)(\\ell_{t-1}+\\phi b_{t-1}),\\]\n\n\\[b_t =\\beta^*(\\ell_t-\\ell_{t-1}) +(1-\\beta^*)\\phi b_{t-1}.\\]\n\nThe forecast becomes\n\n\\[\\widehat y_{t+h\\mid t} =\\ell_t+\\left(\\sum_{j=1}^{h}\\phi^j\\right)b_t.\\]\n\nThe trend contribution increases with the horizon but approaches a finite limit:\n\n\\[\\sum_{j=1}^{h}\\phi^j =\\frac{\\phi(1-\\phi^h)}{1-\\phi}.\\]\n\nWhen \\(\\phi=1\\), the equations reduce to Holt’s undamped method. Damping changes the extrapolation of trend; it is distinct from the smoothing parameters that govern adaptation to observations."
    },
    {
      "heading": "Seasonal methods",
      "text": "Holt–Winters methods add a seasonal component \\(s_t\\) with a known period \\(m\\), such as \\(m=12\\) for annual seasonality in monthly data. Their usual forms distinguish additive from multiplicative seasonality.\n\nAdditive seasonality\n\nAdditive seasonality is appropriate when seasonal fluctuations have roughly constant magnitude in the units of the observations. One common formulation is\n\n\\[\\ell_t =\\alpha(y_t-s_{t-m}) +(1-\\alpha)(\\ell_{t-1}+b_{t-1}),\\]\n\n\\[b_t =\\beta^*(\\ell_t-\\ell_{t-1}) +(1-\\beta^*)b_{t-1},\\]\n\n\\[s_t =\\gamma(y_t-\\ell_{t-1}-b_{t-1}) +(1-\\gamma)s_{t-m}.\\]\n\nHere, \\(\\gamma\\) controls seasonal updating. Define the index of the most recently estimated seasonal state for the forecasted position as\n\n\\[r(t,h)=t+h-m\\left(\\left\\lfloor\\frac{h-1}{m}\\right\\rfloor+1\\right).\\]\n\nThen\n\n\\[\\widehat y_{t+h\\mid t} =\\ell_t+h b_t+s_{r(t,h)}.\\]\n\nSeasonal effects are usually normalized to sum to zero over a cycle so that they can be distinguished from the level.\n\nMultiplicative seasonality\n\nMultiplicative seasonality describes fluctuations whose magnitude is approximately proportional to the series’ level. A common formulation is\n\n\\[\\ell_t =\\alpha\\frac{y_t}{s_{t-m}} +(1-\\alpha)(\\ell_{t-1}+b_{t-1}),\\]\n\n\\[b_t =\\beta^*(\\ell_t-\\ell_{t-1}) +(1-\\beta^*)b_{t-1},\\]\n\n\\[s_t =\\gamma\\frac{y_t}{\\ell_{t-1}+b_{t-1}} +(1-\\gamma)s_{t-m},\\]\n\nwith forecast\n\n\\[\\widehat y_{t+h\\mid t} =(\\ell_t+h b_t)s_{r(t,h)}.\\]\n\nSeasonal factors are commonly normalized to average one. These equations require suitable nonzero denominators and are ordinarily used for strictly positive series.\n\nAlternative Holt–Winters parameterizations use different seasonal updating equations. Consequently, parameter values from different implementations need not be directly comparable. Seasonal methods can also be combined with damped trend or used without a trend component."
    },
    {
      "heading": "Statistical state-space formulation",
      "text": "The updating equations alone specify point forecasts but do not completely specify the distribution of future observations. A state-space model adds an explicit probabilistic description of observations and evolving latent states.\n\nFor example, simple exponential smoothing with additive errors can be written as\n\n\\[y_t=\\ell_{t-1}+\\varepsilon_t,\\]\n\n\\[\\ell_t=\\ell_{t-1}+\\alpha\\varepsilon_t,\\]\n\nwhere \\(\\varepsilon_t\\) is a zero-mean innovation. A common specification assumes independent innovations with a normal distribution and constant variance \\(\\sigma^2\\). The same innovation enters both the observation and state equations.\n\nThis framework underlies the ETS classification, whose three positions describe:\n\n- Error: additive or multiplicative.\n- Trend: none, additive, or a supported damped or multiplicative form.\n- Seasonality: none, additive, or multiplicative.\n\nThus ETS(A,N,N) denotes additive errors with no trend or seasonality, while ETS(A,Ad,A) denotes additive errors, damped additive trend, and additive seasonality. Not every conceivable combination is statistically or numerically suitable.\n\nThe error specification is separate from the seasonal specification. Multiplicative errors describe uncertainty relative to the predicted level; multiplicative seasonality describes how the seasonal pattern scales with level. Different error models can give the same point-forecast equations but different likelihoods and uncertainty intervals.\n\nPrediction intervals\n\nA prediction interval describes uncertainty about a future observation, rather than uncertainty about an estimated mean alone.\n\nFor the additive simple exponential smoothing model above, conditional on known parameters and the state at time \\(t\\),\n\n\\[\\operatorname{Var}(y_{t+h}-\\widehat y_{t+h\\mid t}) =\\sigma^2\\left[1+\\alpha^2(h-1)\\right].\\]\n\nUnder Gaussian innovations, this gives the interval\n\n\\[\\widehat y_{t+h\\mid t} \\pm z_{1-\\eta/2}\\, \\sigma\\sqrt{1+\\alpha^2(h-1)},\\]\n\nwhere \\(z_{1-\\eta/2}\\) is a standard-normal quantile and \\(1-\\eta\\) is the nominal coverage probability. This expression does not include uncertainty from estimating the parameters or current state.\n\nMore complicated ETS models may require recursive calculations or simulation. Their intervals depend on the selected model, including whether uncertainty is additive or multiplicative."
    },
    {
      "heading": "Estimation and model selection",
      "text": "Smoothing parameters and initial states may be fixed in advance or estimated jointly through mathematical optimization. For additive-error models, a common objective is to minimize the sum of squared one-step forecast errors:\n\n\\[\\sum_{t=1}^{n} \\left(y_t-\\widehat y_{t\\mid t-1}\\right)^2.\\]\n\nUnder independent Gaussian innovations with constant variance, minimizing this objective is closely connected to maximum likelihood estimation. Models with multiplicative errors generally require a likelihood that accounts for scale-dependent uncertainty.\n\nInitialization matters, especially for short series or small smoothing parameters. Possible approaches include:\n\n- Using an early observation or an average of early observations for the initial level.\n- Estimating an initial trend from differences or a regression.\n- Estimating seasonal states from several complete cycles.\n- Optimizing initial states jointly with smoothing parameters.\n\nSeasonal normalization constraints prevent the level and seasonal components from representing the same variation redundantly. Parameter restrictions may also be needed to ensure stable, admissible model behavior; conventional componentwise bounds do not by themselves describe every statistical admissibility condition.\n\nModels can be compared using likelihood-based criteria and out-of-sample forecast performance. Time-series cross-validation preserves temporal order, repeatedly fitting to past data and forecasting subsequent observations. Randomly shuffling observations can leak future information into model fitting.\n\nA model with small one-step errors is not necessarily best for a longer forecast horizon. Evaluation therefore depends on both the horizon and the chosen loss function. Measures such as mean squared error emphasize large errors, while percentage measures can behave poorly when observations are zero or near zero."
    },
    {
      "heading": "Relationships to other methods",
      "text": "ARIMA models\n\nSome linear exponential smoothing methods have equivalent ARIMA representations. For simple exponential smoothing,\n\n\\[y_t-y_{t-1} =\\varepsilon_t-(1-\\alpha)\\varepsilon_{t-1},\\]\n\nwhich is an ARIMA(0,1,1) representation under the displayed sign convention. The corresponding forecasts agree when initialization and parameterization are aligned.\n\nHolt’s linear method also has an ARIMA representation under suitable conditions. However, the families are not identical: their usual parameter restrictions differ, and nonlinear ETS models do not generally reduce to linear ARIMA models.\n\nKalman filtering\n\nThe Kalman filter updates latent-state estimates using forecast errors multiplied by a gain. This resembles the error-correction form of exponential smoothing. For suitable linear state-space models, exponential smoothing can be interpreted as a constant-gain update, while a Kalman filter may use a gain that changes over time and can approach a steady value.\n\nSmoothing versus forecasting\n\nDespite its name, exponential smoothing ordinarily uses only observations available up to the current time. It is therefore a causal filter and forecasting method. It differs from retrospective smoothers that use both earlier and later observations to estimate a historical state."
    },
    {
      "heading": "Applications and limitations",
      "text": "Exponential smoothing is used in inventory control, sales forecasting, production planning, staffing, and other settings involving many regularly observed series. Simple and nonseasonal methods require only a small number of stored states; seasonal methods additionally retain seasonal states. This makes sequential updating computationally inexpensive.\n\nIts principal limitations follow from its structure:\n\n- Limited explanatory information. Basic methods use a series’ own history and do not directly represent prices, promotions, weather, or other external predictors.\n- Adaptation lag. Abrupt changes can produce repeated forecast errors before the estimated components adjust.\n- Sensitivity to unusual observations. Large errors can shift the estimated level, trend, or seasonal pattern.\n- Restricted seasonality. Standard Holt–Winters methods represent one fixed seasonal period, not several interacting or changing periods.\n- Difficult data regimes. Intermittent demand, long runs of zeros, irregular observation times, and changing seasonal calendars can require specialized models.\n- Extrapolation constraints. Additive forecasts can become negative, and undamped trends can become implausible at long horizons.\n\nStandard formulas presume equally spaced observations. Applying the same smoothing parameter to observations separated by different time gaps does not preserve a constant rate of forgetting in calendar time."
    },
    {
      "heading": "Historical development",
      "text": "Exponential smoothing emerged as a practical forecasting approach in the 1950s. Robert G. Brown developed methods associated with inventory and demand forecasting, while Charles C. Holt developed extensions for evolving trends. Peter R. Winters’ 1960 work helped establish the seasonal method now known as Holt–Winters.\n\nLater developments included damped-trend forecasting, systematic statistical state-space formulations, likelihood-based estimation, automatic model selection, and model-based prediction intervals. These developments connected simple recursive forecasting procedures with a broader theory of probabilistic time-series models."
    }
  ],
  "links": [
    "time-series",
    "recurrence-relation",
    "geometric-series",
    "exponential-function",
    "linear-regression",
    "state-space-model",
    "normal-distribution",
    "variance",
    "prediction-interval",
    "mathematical-optimization",
    "maximum-likelihood-estimation",
    "cross-validation",
    "loss-function",
    "mean-squared-error",
    "kalman-filter"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}