Forecasting
≈ 18 min read
Names in other languages
No names in other languages are held for this entry yet.
Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.
Forecasting is the process of making predictions based on past and present data. These forecasts can later be compared with actual outcomes. For example, a company might estimate their revenue in the next year, then compare it against the actual results creating a variance actual analysis. Prediction is a similar but more general term. Forecasting might refer to specific formal statistical methods employing time series, cross-sectional or longitudinal data, or alternatively to less formal judgmental methods or the process of prediction and assessment of its accuracy. Usage can vary between areas of application: for example, in hydrology the terms "forecast" and "forecasting" are sometimes reserved for estimates of values at certain specific future times, while the term "prediction" is used for more general estimates, such as the number of times floods will occur over a long period. Risk and uncertainty are central to forecasting and prediction; it is generally considered a good practice to indicate the degree of uncertainty attaching to forecasts. In any case, the data must be up to date in order for the forecast to be as accurate as possible. In some cases the data used to predict the variable of interest is itself forecast.…
Forecasting is the process of making predictions based on past and present data. These forecasts can later be compared with actual outcomes. For example, a company might estimate their revenue in the next year, then compare it against the actual results creating a variance actual analysis. Prediction is a similar but more general term. Forecasting might refer to specific formal statistical methods employing time series, cross-sectional or longitudinal data, or alternatively to less formal judgmental methods or the process of prediction and assessment of its accuracy. Usage can vary between areas of application: for example, in hydrology the terms "forecast" and "forecasting" are sometimes reserved for estimates of values at certain specific future times, while the term "prediction" is used for more general estimates, such as the number of times floods will occur over a long period. Risk and uncertainty are central to forecasting and prediction; it is generally considered a good practice to indicate the degree of uncertainty attaching to forecasts. In any case, the data must be up to date in order for the forecast to be as accurate as possible. In some cases the data used to predict the variable of interest is itself forecast. A forecast is not to be confused with a Budget; budgets are more specific, fixed-term financial plans used for resource allocation and control, while forecasts provide estimates of future financial performance, allowing for flexibility and adaptability to changing circumstances. Both tools are valuable in financial planning and decision-making, but they serve different functions.
Forecast improvements
Forecast improvement projects have been operated in a number of sectors: the National Hurricane Center's Hurricane Forecast Improvement Project (HFIP) and the Wind Forecast Improvement Project sponsored by the US Department of Energy are examples. In relation to supply chain management, the Du Pont model has been used to show that an increase in forecast accuracy can generate increases in sales and reductions in inventory, operating expenses and commitment of working capital. The Groceries Code Adjudicator in the United Kingdom, which regulates supply chain management practices in the groceries retail industry, has observed that all the retailers who fall within the scope of his regulation "are striving for continuous improvement in forecasting practice and activity in relation to promotions".
Categories of forecasting methods
Qualitative vs. quantitative methods
Qualitative forecasting techniques are subjective, based on the opinion and judgment of consumers and experts; they are appropriate when past data are not available. They are usually applied to intermediate- or long-range decisions. Examples of qualitative forecasting methods are informed opinion and judgment, the Delphi method, market research, and historical life-cycle analogy. Quantitative forecasting models are used to forecast future data as a function of past data. They are appropriate to use when past numerical data is available and when it is reasonable to assume that some of the patterns in the data are expected to continue into the future. These methods are usually applied to short- or intermediate-range decisions. Examples of quantitative forecasting methods are last period demand, simple and weighted N-Period moving averages, simple exponential smoothing, Poisson process model based forecasting and multiplicative seasonal indexes. Previous research shows that different methods may lead to different level of forecasting accuracy. For example, GMDH neural network was found to have better forecasting performance than the classical forecasting algorithms such as Single Exponential Smooth, Double Exponential Smooth, ARIMA and back-propagation neural network.
Average approach
In this approach, the predictions of all future values are equal to the mean of the past data. This approach can be used with any sort of data where past data is available. In time series notation: \hat{y}_{T+h|T} = \bar{y} = ( y_1 + ... + y_T ) / T where y_1, ... , y_T is the past data. Although the time series notation has been used here, the average approach can also be used for cross-sectional data (when we are predicting unobserved values; values that are not included in the data set). Then, the prediction for unobserved values is the average of the observed values.
Naïve approach
Naïve forecasts are the most cost-effective forecasting model, and provide a benchmark against which more sophisticated models can be compared. This forecasting method is only suitable for time series data. Using the naïve approach, forecasts are produced that are equal to the last observed value. This method works quite well for economic and financial time series, which often have patterns that are difficult to reliably and accurately predict. If the time series is believed to have seasonality, the seasonal naïve approach may be more appropriate where the forecasts are equal to the value from last season. In time series notation: \hat{y}_{T+h|T} = y_T
Drift method
A variation on the naïve method is to allow the forecasts to increase or decrease over time, where the amount of change over time (called the drift) is set to be the average change seen in the historical data. So the forecast for time T+h is given by \hat{y}_{T+h|T} = y_T + \frac{h}{T-1}\sum_{t=2}^T (y_{t}-y_{t-1}) = y_{T}+h\left(\frac{y_{T}-y_{1}}{T-1}\right). This is equivalent to drawing a line between the first and last observation, and extrapolating it into the future.
Seasonal naïve approach
The seasonal naïve method accounts for seasonality by setting each prediction to be equal to the last observed value of the same season. For example, the prediction value for all subsequent months of April will be equal to the previous value observed for April. The forecast for time T + h is \hat{y}_{T+h|T} = y_{T+h-m(k+1)} where m is the seasonal period (for example, for monthly data with yearly seasonality, m=12 ) and k is the integer part of (h-1) / m . The seasonal naïve method is particularly useful for data that has a very high level of seasonality. \begin{aligned} f_n(x_t) = \dfrac{1}{(1+x_t^n)} \, , \qquad n \in {\mathbb {N}},\;x\in {\mathbb {R}}. \end{aligned} The short term behaviour x_t and the is the medium-long-term trend y_t are \begin{aligned} {\left\{ \begin{array}{ll} x_{t+1} = \alpha f_n(x_t) + \gamma \, y_t + \delta \\ y_{t+1} = \beta \, y_t - \mu \, x_{t} + \eta \end{array}\right. } \end{aligned} where \alpha, \gamma, \beta, \mu, \eta are some parameters. This approach has been proposed to simulate bursts of seemingly stochastic activity, interrupted by quieter periods. The assumption is that the presence of a strong deterministic ingredient is hidden by noise. The deterministic approach is noteworthy as it can reveal the underlying dynamical systems structure, which can be exploited for steering the dynamics into a desired regime.
Time series methods
Time series methods use historical data as the basis of estimating future outcomes. They are based on the assumption that past demand history is a good indicator of future demand. Moving average Weighted moving average Exponential smoothing Autoregressive moving average (ARMA) (forecasts depend on past values of the variable being forecast and on past prediction errors) Autoregressive integrated moving average (ARIMA) (ARMA on the period-to-period change in the forecast variable) e.g. Box–Jenkins Seasonal ARIMA or SARIMA or ARIMARCH, Extrapolation Linear prediction Trend estimation (predicting the variable as a linear or polynomial function of time) Growth curve (statistics) Recurrent neural network
Relational methods
Some forecasting methods try to identify the underlying factors that might influence the variable that is being forecast. For example, including information about climate patterns might improve the ability of a model to predict umbrella sales. Forecasting models often take account of regular seasonal variations. In addition to climate, such variations can also be due to holidays and customs: for example, one might predict that sales of college football apparel will be higher during the football season than during the off season. Several informal methods used in causal forecasting do not rely solely on the output of mathematical algorithms, but instead use the judgment of the forecaster. Some forecasts take account of past relationships between variables: if one variable has, for example, been approximately linearly related to another for a long period of time, it may be appropriate to extrapolate such a relationship into the future, without necessarily understanding the reasons for the relationship. Causal methods include: Regression analysis includes a large group of methods for predicting future values of a variable using information about other variables. These methods include both parametric (linear or non-linear) and non-parametric techniques. Autoregressive moving average with exogenous inputs (ARMAX) Quantitative forecasting models are often judged against each other by comparing their in-sample or out-of-sample mean square error, although some researchers have advised against this. Different forecasting approaches have different levels of accuracy. For example, it was found in one context that GMDH has higher forecasting accuracy than traditional ARIMA.
Judgmental methods
Judgmental forecasting methods incorporate intuitive judgement, opinions and subjective probability estimates. Judgmental forecasting is used in cases where there is a lack of historical data or during completely new and unique market conditions. Judgmental methods include: Composite forecasts Cooke's method Delphi method Forecast by analogy Scenario building Statistical surveys Technology forecasting
Artificial intelligence methods
Artificial neural networks Group method of data handling Support vector machines Often these are done today by specialized programs loosely labeled Data mining Machine learning Pattern recognition
Geometric extrapolation with error prediction
Can be created with 3 points of a sequence and the "moment" or "index". This type of extrapolation has 100% accuracy in predictions in a big percentage of known series database (OEIS).
Other methods
Granger causality Simulation Demand forecasting Probabilistic forecasting and Ensemble forecasting
Forecasting accuracy
The forecast error (also known as a residual) is the difference between the actual value and the forecast value for the corresponding period: \ E_t = Y_t - F_t where E is the forecast error at period t, Y is the actual value at period t, and F is the forecast for period t. A good forecasting method will yield residuals that are uncorrelated. If there are correlations between residual values, then there is information left in the residuals which should be used in computing forecasts. This can be accomplished by computing the expected value of a residual as a function of the known past residuals, and adjusting the forecast by the amount by which this expected value differs from zero. A good forecasting method will also have zero mean. If the residuals have a mean other than zero, then the forecasts are biased and can be improved by adjusting the forecasting technique by an additive constant that equals the mean of the unadjusted residuals. Measures of aggregate error:
Scaled-dependent errors
The forecast error, E, is on the same scale as the data, as such, these accuracy measures are scale-dependent and cannot be used to make comparisons between series on different scales. Mean absolute error (MAE) or mean absolute deviation (MAD): \ MAE = MAD = \frac{\sum_{t=1}^{N} |E_t|}{N} Mean squared error (MSE) or mean squared prediction error (MSPE): \ MSE = MSPE = \frac{\sum_{t=1}^N {E_t^2}}{N} Root mean squared error (RMSE): \ RMSE = \sqrt{\frac{\sum_{t=1}^N {E_t^2}}{N}} Average of Errors (E): \ \bar{E}= \frac{\sum_{t=1}^N {E_t}}{N}
Percentage errors
These are more frequently used to compare forecast performance between different data sets because they are scale-independent. However, they have the disadvantage of being extremely large or undefined if Y is close to or equal to zero. Mean absolute percentage error (MAPE): \ MAPE = 100* \frac{\sum_{t=1}^N |\frac{E_t}{Y_t}|}{N} Mean absolute percentage deviation (MAPD): \ MAPD = \frac{\sum_{t=1}^{N} |E_t|}{\sum_{t=1}^{N} |Y_t|}
Scaled errors
Hyndman and Koehler (2006) proposed using scaled errors as an alternative to percentage errors. Mean absolute scaled error (MASE): MASE = \frac{\sum_{t=1}^{N} |\frac{E_t}{\frac{1}{N-m}\sum_{t=m+1}^{N}|Y_t - Y_{t-m}|}|}{N} where m=seasonal period or 1 if non-seasonal
Other measures
Forecast skill (SS): \ SS = 1- \frac{MSE_{forecast}}{MSE_{ref}} Business forecasters and practitioners sometimes use different terminology. They refer to the PMAD as the MAPE, although they compute this as a volume weighted MAPE. For more information, see Calculating demand forecast accuracy. When comparing the accuracy of different forecasting methods on a specific data set, the measures of aggregate error are compared with each other and the method that yields the lowest error is preferred.
Training and test sets
When evaluating the quality of forecasts, it is invalid to look at how well a model fits the historical data; the accuracy of forecasts can only be determined by considering how well a model performs on new data that were not used when fitting the model. When choosing models, it is common to use a portion of the available data for fitting, and use the rest of the data for testing the model, as was done in the above examples.
Cross-validation
Cross-validation is a more sophisticated version of training a test set. For cross-sectional data, one approach to cross-validation works as follows: Select observation i for the test set, and use the remaining observations in the training set. Compute the error on the test observation. Repeat the above step for i = 1,2,..., N where N is the total number of observations. Compute the forecast accuracy measures based on the errors obtained. This makes efficient use of the available data, as only one observation is omitted at each step For time series data, the training set can only include observations prior to the test set. Therefore, no future observations can be used in constructing the forecast. Suppose k observations are needed to produce a reliable forecast; then the process works as follows: Starting with i=1, select the observation k + i for the test set, and use the observations at times 1, 2, ..., k+i–1 to estimate the forecasting model. Compute the error on the forecast for k+i. Repeat the above step for i = 2,...,T–k where T is the total number of observations. Compute the forecast accuracy over all errors. This procedure is sometimes known as a "rolling forecasting origin" because the "origin" (k+i -1) at which the forecast is based rolls forward in time. Further, two-step-ahead or in general p-step-ahead forecasts can be computed by first forecasting the value immediately after the training set, then using this value with the training set values to forecast two periods ahead, etc.
Seasonality and cyclic behaviour
Seasonality
Seasonality is a characteristic of a time series in which the data experiences regular and predictable changes which recur every calendar year. Any predictable change or pattern in a time series that recurs or repeats over a one-year period can be said to be seasonal. It is common in many situations – such as grocery store or even in a Medical Examiner's office —that the demand depends on the day of the week. In such situations, the forecasting procedure calculates the seasonal index of the "season" – seven seasons, one for each day – which is the ratio of the average demand of that season (which is calculated by Moving Average or Exponential Smoothing using historical data corresponding only to that season) to the average demand across all seasons. An index higher than 1 indicates that demand is higher than average; an index less than 1 indicates that the demand is less than the average.
Cyclic behaviour
The cyclic behaviour of data takes place when there are regular fluctuations in the data which usually last for an interval of at least two years, and when the length of the current cycle cannot be predetermined. Cyclic behavior is not to be confused with seasonal behavior. Seasonal fluctuations follow a consistent pattern each year so the period is always known. As an example, during the Christmas period, inventories of stores tend to increase in order to prepare for Christmas shoppers. As an example of cyclic behaviour, the population of a particular natural ecosystem will exhibit cyclic behaviour when the population decreases as its natural food source decreases, and once the population is low, the food source will recover and the population will start to increase again. Cyclic data cannot be accounted for using ordinary seasonal adjustment since it is not of fixed period.
Limitations
Limitations pose barriers beyond which forecasting methods cannot reliably predict. There are many events and values that cannot be forecast reliably. Events such as the roll of a die or the results of the lottery cannot be forecast because they are random events and there is no significant relationship in the data. When the factors that lead to what is being forecast are not known or well understood such as in stock and foreign exchange markets forecasts are often inaccurate or wrong as there is not enough data about everything that affects these markets for the forecasts to be reliable, in addition the outcomes of the forecasts of these markets change the behavior of those involved in the market further reducing forecast accuracy. The concept of "self-destructing predictions" concerns the way in which some predictions can undermine themselves by influencing social behavior. This is because "predictors are part of the social context about which they are trying to make a prediction and may influence that context in the process". For example, a forecast that a large percentage of a population will become HIV infected based on existing trends may cause more people to avoid risky behavior and thus reduce the HIV infection rate, invalidating the forecast (which might have remained correct if it had not been publicly known). Or, a prediction that cybersecurity will become a major issue may cause organizations to implement more security cybersecurity measures, thus limiting the issue.
Performance limits of fluid dynamics equations
As proposed by Edward Lorenz in 1963, long range weather forecasts, those made at a range of two weeks or more, are impossible to definitively predict the state of the atmosphere, owing to the chaotic nature of the fluid dynamics equations involved. Extremely small errors in the initial input, such as temperatures and winds, within numerical models double every five days.
See also
Accelerating change Cash flow forecasting Cliodynamics Collaborative planning, forecasting, and replenishment Computer supported brainstorming Earthquake prediction Economic forecasting Energy forecasting Forecasting bias Foresight (future studies) Frequency spectrum Futures studies Futurology Kondratiev wave Least squares Least-squares spectral analysis Optimism bias Planning Prediction Predictive analytics Risk management Scenario planning Spending wave Strategic foresight Technology forecasting Thucydides Trap Time series Weather forecasting Wind power forecasting
References
Sources
.
External links
Forecasting Principles: "Evidence-based forecasting" International Institute of Forecasters Introduction to Time series Analysis (Engineering Statistics Handbook) - A practical guide to Time series analysis and forecasting Time Series Analysis Global Forecasting with IFs Earthquake Electromagnetic Precursor Research Forecasting Science and Theory of Forecasting
Text derived from https://en.wikipedia.org/w/index.php?oldid=1374936049 under CC BY-SA 4.0; modifications have been made.