Sie sind auf Seite 1von 7

RESEARCH STARTERS

ACADEMIC TOPIC OVERVIEWS

Time Series
Statistics > Time Series Overview
Managers and other business decision makers frequently need to determine the best course of action and develop strategic plans to help the organization reach its goals and objectives. To optimize the worth of such decisions, good business strategy needs to be based on the rigorous analysis of empirical data. Sometimes the data are readily obtainable through activities such as an analysis of the organizations resources and abilities. Sometimes the data require research and analysis such as data concerning competitor capabilities and offerings. Sometimes the data require creative guesswork such as determining market needs and trends in the future. Fortunately, there are a wide range of descriptive and inferential statistical tools available for analyzing and interpreting the data that managers rely on to make decisions. One of the tools that is particularly helpful for the latter category of analysis where one needs to forecast trends as the basis for decision making is time series analysis. Uses of Time Series Data Time series data are data gathered on one or more specific characteristics of interest over a period of time at intervals of regular length. These data series are used in business forecasting to examine patterns, trends, and cycles from the past in order to predict patterns, trends, and cycles in the future. Time series analysis typically involves observing and analyzing the patterns in historical data. These patterns are extrapolated to forecast future behavior. Most statistical analysis of time series data involves model building, the development of a concise mathematical description of past events. These mathematical models are used to forecast how the pattern will continue into the future. Time series are analyzed through several techniques including nave methods, averaging, smoothing, regression analysis, and decomposition. These analysis techniques assume that the sequence of observations is a set of jointly distributed random variables. Through the analysis of time series data, one can study the structure of the correlation (i.e., the degree to which two events or variables are consistently related) over time to determine the appropriateness of the model.

Table of Contents Abstract Keywords Overview Applications Terms & Concepts Bibliography Suggested Reading

Abstract
Time series data are used in business forecasting to examine patterns, trends, and cycles from the past in order to predict patterns, trends, and cycles in the future. In general, there are three objectives for the use of time series data: Understand the mechanism underlying the observed data, extrapolate the data in order to predict their effect on future behavior or data values, and control a process or system so that its outcome is more favorable to the business. Time series data can take several distinctive forms and can be influenced both by deterministic and stochastic variables. Although there are several approaches to modeling time series data, they often yield different results. Model building and forecasting using time series data is a complicated process that is part art and part science. Human beings will always have to determine which variables to include, where the line of best fit lays, what time periods to consider, and how to interpret the results. This can be both strength and a weakness.

EBSCO Research Starters Copyright 2008 EBSCO Publishing Inc. All Rights Reserved

Time Series

Essay by Ruth A. Wienclaw, Ph.D.

Keywords
Autocorrelation Autoregression Autoregressive Integrated Moving Average Business Cycle Data Decomposition Deterministic Variables Forecasting Moving Average Seasonal Fluctuation Stationarity Stochastic Time Series Data Trend Variable

can point out areas where the organization needs to change (e.g., a new widget design that incorporates more technology might be appropriate) in order to change the trend in the future. Forms of Time Series Data As shown in Figure 1, real world time series data can take several distinctive forms. Stationarity Figure 1a shows the viscosity data on a chemical product over time. These data remain fairly constant over time and are said to be constant about the mean (an arithmetically derived measure of central tendency in which the sum of the values of all the data points is divided by the number of data points). This characteristic of the data is referred to as stationarity. Stationarity exists when the probability distribution of a time series does not change over time. Stationarity is of interest to analysts because it allows one to mathematically model the process with an equation with fixed coefficients that estimate future values from past history. If the process is assumed to be stationary, the probability of a given fluctuation in the process is assumed to be the same at any given point in time. The time series data in Figure 1b, however, do not exhibit the same degree of constancy or stationarity. It is difficult to mathematically model a non-stationary process using a simple algebraic equation. However, it can be possible to use a simple mathematical procedure to transform non-stationary processes into ones that are approximately stationary for purposes of analysis. This allows the development of models to help the analyst or decision maker better understand the underlying mechanisms in the data series. Figure 1: Three Examples of Time Series Data (From Mastrangelo, Simpson, & Montgomery, p. 828) COMPDOC

Objectives of Time Series Data In general, there are several objectives for the use of time series data. First, time series data are often analyzed in order to understand the mechanism underlying the observed data and to build a model that describes the mechanism and its influence on the variables of interest. Second, while understanding the mechanisms that influence events and trends is of interest, time series data are frequently analyzed not only to understand these mechanisms, but more importantly to extrapolate them in order to predict their affect on future behavior or data values. Third, although sometimes knowing this information is sufficient for making decisions and developing strategies, there are also situations in which the results of the analysis of time series data can also give organizations information that will enable them to control a process or system so that its outcome is more favorable to the business. For example, one might find that changes in industry technology are progressively making a current widget design obsolete and that sales are dropping of. If acquired in time, this information

Deterministic Variables A third general form that can be taken by time series data is illustrated in Figure 1c. Time series data that can be influenced by various deterministic variables are those for which there are specific causes or determiners. This type of variable includes trends, business cycles, and seasonal fluctuations. Trends are persistent, underlying directions in which a factor or characteristic is moving in either the short, intermediate, or long term. Trends tend to be linear rather than cyclic, and grow or shrink steadily over a period of years. For example, a trend might be an increasing tendency for business to outsource and offshore technical support and customer service in many high tech companies. Trends are not necessarily linPage 2

EBSCO Research Starters Copyright 2008 EBSCO Publishing Inc. All Rights Reserved

Time Series

Essay by Ruth A. Wienclaw, Ph.D.

ear, however. For example, trends in new industries tend to be curvilinear as the demand for the new product or service grows after its introduction then declines after the product or service becomes integrated into the economy. Another type of deterministic factor is business cycles. These factors are continually recurring variation in total economic activity. Business cycles tend to occur across most sectors of the economy at the same time. For example, several years of a boom economy with expansion of economic activity (e.g., more jobs, higher sales) are frequently followed by slower growth or even contraction of economic activity. Another category of deterministic factor is seasonal fluctuations. These are changes in activity that occur in a fairly regular annual pattern and which are related to seasons of the year, the calendar, or holidays. Figure 1c shows a seasonal fluctuation in soft drink sales, with sales going up during the warmer months and down during the colder months over a period of four years. Stochastic Variables In addition to deterministic variables, time series data can be influenced by stochastic variables. These variables are caused by randomness or include an element of chance or probability and are caused by unpredictable factors. Stochastic variables can be either irregular or random. Examples of irregular stochastic variables include natural disasters such as earthquakes or floods, political disturbances such as war or changes in the political party in charge, strikes, and other external factors. Other unpredictable or random factors include situations such as high absenteeism due to an epidemic that could affect a businesss profitability. The real world includes many variables both deterministic and stochastic that can affect time series data, and most real world situations and models include both types of variables.

Smoothing Techniques Nave Forecasting Another family of methods frequently used in building models from time series data is smoothing techniques. One approach to smoothing time series data is through nave forecasting models. These are simple models that assume that future outcomes are best predicted by the more recent data in the time series. Under this philosophy, for example, one would assume that last months sales were a better predictor of next months sales than were the sales from six months ago. It should be borne in mind, however, that nave forecasting models do not consider the possibility of trends, business cycles, or seasonal fluctuations. For example, if ten gross of widgets were sold last month, a nave model would conclude that ten gross of widgets would also be sold next month. Because of this assumption, nave forecasting models work better on data that are reported more frequently (e.g., daily or weekly) or in situations without trends or seasonality. One danger in nave model forecasts lies in the fact that they are often based on the observations of one time period. As a result, they can easily become a function of irregular fluctuations in data (e.g., Acme corporation needed a one-time purchase of widgets to set up their new operations facility which accounts for the one-time high demand for widgets in the previous month). Averaging Models Smoothing of time series data can also be done using averaging models. These techniques take into account data from several time periods, thereby neutralizing the problem of nave models in which the forecast is overly sensitive to irregular fluctuations as illustrated in the previous example. In the simple average model, forecasts for an upcoming time period are made using the average of the values for a specified number of previous time periods (e.g., the forecast of widgets sales for next month might be the average number of widgets sold per month over the past six months). In the moving average approach, however, the average value from previous time periods is used to forecast future time periods and is updated in each ensuing time period by including the new values not available in the previous average and dropping out the date from the earliest time periods. Although the moving average approach has the advantage of taking into account the most recent data available, it can be difficult to choose the optimal length of time over which to compute the moving average. Another drawback of the moving average approach is that it does not take in to account the effects of trends, business cycles, and seasonal fluctuations. Because of this fact, analysts often use a weighted moving average which gives more weight to some time periods in the series than to others. For example, if three months ago the company introduced a newly redesigned product into the marketplace, the analyst might believe that the past three months reflect the markets reaction to the new design and be better able to forecast the continuing reaction than if s/he did not have this information. Similarly, prediction of sales from last years holiday season may be better predictors of next years holiday sales than sales during the summer months. A third approach to smoothing time series data comprises exponential smoothing
Page 3

Applications
Approaches to Time Series Modeling Regression Methods There are several approaches to modeling time series data. One of the primary tools for analyzing time series data and developing a mathematical model of real world situations is regression methods. These are a family of statistical techniques that are used to develop a mathematical model for use in predicting one variable from the knowledge of another variable. Regression methods include both linear and nonlinear techniques. In linear regression, a line of best fit is fitted to a data set to estimate the effect of a single independent variable. The slope of the line shows the impact of the independent variable on the dependent variable. Other regression methods are available for use in multivariate and nonlinear situations.

EBSCO Research Starters Copyright 2008 EBSCO Publishing Inc. All Rights Reserved

Time Series

Essay by Ruth A. Wienclaw, Ph.D.

techniques. Exponential techniques use weight data from previous time periods with exponentially decreasing importance so that the new forecast is a product of the current forecast and the current actual value. Autoregression Another approach to modeling time series data is autoregression, a multiple regression technique in which future values of the dependent variable are predicted from past values of the variable. This technique takes advantage of the relationship of values to the values of previous time periods. The independent variables are time-lagged versions of the dependent variable so that one tries to forecast a future value of a variable from knowledge of that variables value in previous time periods. This approach can be particularly useful for locating both seasonal and cyclic effects. Mixed/Integrated Techniques Times series data can be modeled using mixed or integrated techniques that utilize both moving average and autoregressive techniques. One of these approaches is the autoregressive integrated moving average (ARIMA) model (also called the Box-Jenkins model). The ARIMA model is an integrated tool for understanding and forecasting using time series data that has both an autoregressive and a moving average component. ARIMA modeling techniques are powerful and frequently result in a better model than either the use of moving averages or autoregressive techniques alone. For example, ARIMA can be used to determine the length of the weights (i.e., how much of the past should be used to predict the next observation) and the values of these weights. Autocorrelation Models based on time series data can produce spurious results when the error terms of the model are correlated with each other in a situation referred to as autocorrelation or serial correlation. Autocorrelation can be problematic in the use of regression analysis because regression analysis assumes that error terms are not correlated because they are either independent or random. When autocorrelation occurs, the estimates of the regression coefficients may be inefficient and both the variance of the error terms and the true standard deviation may be significantly underestimated. In addition, the autocorrelation effect means that the confidence intervals and t and F tests are no longer strictly applicable. A number of methods are available to determine whether or not autocorrelation is present in time series data (e.g., the DurbinWatson test). Autocorrelated data can sometimes be corrected by the addition of independent variables or transforming variables. Determining Which Forecast to Use As discussed above, there are a number of techniques available to forecast stationary time series data (i.e., those that show no significant trend, cyclic, or seasonal effects). However, these approaches often yield different results. To help determine which forecast better models a given set of data, it is necessary to determine the amount of error produced by each technique. This is

the difference between the forecasted value of a variable and the actual value of a variable. A number of techniques are available for doing this including mean error, mean absolute deviation, mean square error, mean percentage error, and mean absolute percentage error. Data that are influenced by trends can be analyzed by a number of approaches, including linear regression and regression using quadratic models. However, if these data are also influenced by seasonal fluctuations, other data analysis techniques are more appropriate. One of the most frequently used of these techniques is decomposition, in which the time series data are broken down into the four component factors of trend, business cycle, seasonal fluctuation, and irregular or random fluctuation. Modeling Technologies Model building and forecasting using time series data is a complicated process that is part art and part science. However, there are now a number of software packages available that can help in this task. The techniques and concomitant software for developing forecasting models continue to improve. However, these techniques and tools are still not highly accurate for erratic patterns with short life cycles. No matter how refined or powerful the tools, some randomness or noise will always remain in time series data. It is this noise that limits the upper limit of the forecast. In addition, forecasting cannot be done wholly based on mathematics. Human beings will always have to determine which variables to include, where the line of best fit lies, what periods to consider, and how to interpret the results. This can be both a strength and a weakness.

Terms & Concepts


Autocorrelation: A problem occurring over time in regression analysis when the error terms of the forecasting model are correlated. Also called serial correlation. Autoregression: A multiple regression technique used in forecasting in which future values of the variable are predicted from past values of the variable. Autoregressive Integrated Moving Average (ARIMA): An integrated tool for understanding and forecasting using time series data. The ARIMA model has both an autoregressive and a moving average component. The ARIMA model is also referred to as the Box-Jenkins model. Business Cycle: A continually recurring variation in total economic activity. Such expansions or contractions of economic activity tend to occur across most sectors of the economy at the same time. Data: (sing. datum) In statistics, data are quantifiable observations or measurements that are used as the basis of scientific research. Deterministic Variables: Variables for which there are specific causes or determiners. These include trends, business cycles, and seasonal fluctuations.
Page 4

EBSCO Research Starters Copyright 2008 EBSCO Publishing Inc. All Rights Reserved

Time Series

Essay by Ruth A. Wienclaw, Ph.D.

Decomposition: The process of breaking down time series data into the component factors of trends, business cycles, seasonal fluctuations, and irregular or random fluctuations. Forecasting: In business, forecasting is the science of estimating or predicting future trends. Forecasts are used to support managers in making decisions about many aspects of the business including buying, selling, production, and hiring. Moving Average: A method used in forecasting in which the average value from previous time periods is used to forecast future time periods. The average is updated in each ensuing time period by including the new values not available in the previous average and dropping out the date from the earliest time periods. Seasonal Fluctuation: Changes in economic activity that occur in a fairly regular annual pattern. Seasonal fluctuations may be related to seasons of the year, the calendar, or holidays. Stationarity: The condition of a random process where its statistical properties do not vary with time. Stochastic: Involving chance or probability. Stochastic variables are random or have an element of chance or probability associated with their occurrence. Time Series Data: Data gathered on a specific characteristic over a period of time. Time series data are used in business forecasting. To be useful, time series data must be collected at intervals of regular length. Trend: The persistent, underlying direction in which something is moving in either the short, intermediate, or long term. Identification of a trend allows one to better plan to meet future needs. Variable: An object in a research study that can have more than one value. Independent variables are stimuli that are manipulated in order to determine their effect on the dependent variables (response). Extraneous variables are variables that affect the response but that are not related to the question under investigation in the study.

Mastrangelo, C. M., Simpson, J. R., & Montgomery, D. C. (2001). Time series analysis. In Saul I. Gass, S. I. & Harris, C. M. (eds), Encyclopedia of Operations Research and Management Science (pp. 828-833). New York: Wiley. Retrieved September 11, 2007, from EBSCO Online Database Business Source Complete. http://search.ebscohost.com/login.aspx?direct=true&db=bth&AN=21891965 &site=ehost-live Nazem, S. M. (1988). Applied time series analysis for business and economic forecasting. New York: Marcel Dekker. Pindyck, R. S. & Rubinfeld, D. L. (1998). Econometric models and economic forecasts. Boston: Irwin/McGraw-Hill.

Suggested Reading
Craighead, C. W. (2004). Right on target for time-series forecasting. Decision Sciences Journal of Innovative Education, 2(2), p207-212. Retrieved September 12, 2007, from EBSCO Online Database Business Source Complete. http://search.ebscohost.com/login.aspx?direct=t rue&db=bth&AN=13279969&site=ehost-live De Luna, X. (2001). Guaranteed-content prediction intervals for non-linear autoregressions. Journal of Forecasting, 20(4), 265-72. Retrieved September 12, 2007, from EBSCO Online Database Business Source Complete. http://search.ebscohost.com/login.aspx?direct=true&db=bt h&AN=13638439&site=ehost-live Jarvis, C. H. & Stuart, N. (2001). Accounting for error when modelling with time series data: Estimating the development of crop pests throughout the year. Transactions in GIS, 5(4), 327-343. Retrieved September 12, 2007, from EBSCO Online Database Business Source Complete. http://search.ebscohost.com/login.aspx?direct=true&db=bt h&AN=5439610&site=ehost-live Ng, S. & Vogelsang, T. J. (2002). Forecasting autoregressive time series in the presence of deterministic components. Econometrics Journal, 5(1), 196-224. Retrieved September 12, 2007, from EBSCO Online Database Business Source Complete. http://search.ebscohost.com/ login.aspx?direct=true&db=bth&AN=6719925&site=eho st-live
Page 5

Bibliography
Black, K. (2006). Business statistics for contemporary decision making (4th ed.). New York: John Wiley & Sons. Gilliland, M. & Leonard, M. (2006). Forecasting software the past and the future. Country Commerce, 33-36. Retrieved September 11, 2007, from EBSCO Online Database Business Source Complete.http://search.ebscohost.com/ login.aspx?direct=true&db=bth&AN=21164212&site=eh ost-live

EBSCO Research Starters Copyright 2008 EBSCO Publishing Inc. All Rights Reserved

Time Series

Essay by Ruth A. Wienclaw, Ph.D.

Essay by Ruth A. Wienclaw, Ph.D.


Ruth A. Wienclaw holds a Doctorate in industrial /organizational psychology with a specialization in organization development from the University of Memphis. She is the owner of a small business that works with organizations in both the public and private sectors, consulting on matters of strategic planning, training, and human /systems integration.

EBSCO Research Starters Copyright 2008 EBSCO Publishing Inc. All Rights Reserved

Page 6

Das könnte Ihnen auch gefallen