Lecture 6 Forecasting Annual RevenuesForecasting ‘A proper forecast combines a statistical projection of the past with expected changes in the future. The lecture will show how to create statistical models for projecting past behavior into the future and will demonstrate how to recognize when past trends have changed. Forecasts of both annual and seasonal revenues are essential to planning a company’s future.Importance of forecasting Y Forecasts help CFOs arrange financing for capital expenditures in plants and facilities, plan month-to-month borrowing to meet payrolls and maintain inventories, and invest surplus cash in short-term securities. They help CFOs prepare financial plans and better cope with whatever the future brings. Y Forecasts help operations managers acquire manpower, equipment, and materials in time to manufacture goods or provide services when customers want them, When demands are greater than expected, companies may need to work overtime to satisfy customers, hire temporary workers to serve customers, or reorder products so as not to run out of stock and lose sales. Conversely, when demands are less than expected, inventories of finished goods can pile up, the cost of holding inventories may incréase, facilities may sit idle, and workers may be sent home.Cont- ¥ Forecasts help sales managers keep customers happy by meeting their demands promptly. Marketing managers can better adjust prices and develop selling strategies. Y + Investors decide to buy or sell stocks and other financial securities according to their expectations of a company’s future success in the marketplace. These expectations are partly based on forecasts. Y Forecasts of a company’s future earnings are important to the decisions of banks and other lenders to grant or to refuse to make loans for capital investments.Regression Models They are popular for making both mid- and long-range forecasts. Regression models are equations that describe the relationship between dependent values, such as annual sales, and one or more independent variables, such as time, advertising budget, and the average personal income of buyers. They are useful for making statistical projections of past behavior for several months or quarters to several years into the future. Regression models include linear and several kinds of curvilinear equations that express the trends of historical data. ‘Time will be considered as the single independent variable and express annual sales as a function of time.Cont- * Time is a composite proxy for such factors as population, personal income, inflation, and other demographic and economic factors that are directly related, in a causal sense, to a company’s annual sales. However multivariate regression models express Annual Sales Revenues as a function of a number of such variables. A regression equation expresses the trend of the data. A forecast is made by extrapolating or projecting the past trend forward, into the future. IF an equation is a valid model of the past trend and IF the trend does not change in the future, the past and future values will all scatter randomly about the trend line, Note the IF conditions carefull * The technique can be applied to financial statements variables as well as the ratios.Steps in creating regression models ‘Once the data has been collected from records of the past, the steps for developing regression models to project the past cod into the future inciade the following! roots Identify the type of equation that is most suitable for fitting the trend of the data. Evaluate the equation’s parameters—that is, determine the values of the coefficients of a regression equation for fitting the data ‘Validate the model—that is, show that the model reproduces the data’ trend. Determine the model’ accuracy: Use the model to project future values of interest. Determine the accuracy of the forecast future values Present the results in useful, management quality formatsCont- * Time-series models are another way to use time as the independent variable in a forecasting model. These are autoregressive models, and they are better suited than regression models for short-term forecasts for tactical purposes. Typical applications include managing inventories of items that turn over quickly, as in supermarkets, or following the daily movement of monetary exchange rates. They can be adjusted for changes in trends and seasonalit Forecasting is an important management tool with many applications. For service facilities, such as banks, supermarkets, hospital emergency rooms, toll bridges, and information networks, the dependent variable of forecasting models can be something as simple as the arrival of customers, which can be people, cars, or bytes of data. These vary widely during a single day.Adjusting for future changes in past trends “The basic premise in using statistical projections of past behavior as forecasts of the future is that past trends and seasonal behavior will continue unchanged. If the projections are relatively short term (e.g,, a few weeks or months) and conditions are relatively stable, the projections should be accurate enough to be useful as forecasts. For long- term forecasts, however, statistical projections of the past are best treated as the starting point for forecasting the future. Making adjustments for future changes in past behavior uses various types of judgmental models, such as the Delphi technique and sales force estimates, as well as large-scale economic models of national and international economies. Spreadsheets provide many statistical and charting tools that simplify the selection and creation of forecasting models. They make it easy to create tables and charts that present results in convincing form.Linear Regression models * To develop a linear regression model for forecasting, consider the annual sales of Tak Limited. * To simplify the regression equation, the actual years are replaced by the values of X (.e., by the number of years).Identify the type of Equation that best fit the trend of data * Identifying the trend is best done by creating a scatter plot, and examining how well different types of equations fit the data. * To create this chart, use the mouse to select the data in the Cells and then select Excel’s ChartWizard. Select the “XY (Scatter)” chart type. * Accept the default chart sub-type to plot the data points without connecting lines. Y Create the Scatter Plot by choosing the XY (Scatter) Chart Type. Y Create the Scatter Chart by making sure that the data range is correct and the series of data are in columns.Evaluating the parameters of the linear regression model The general equation for a straight-line or linear relationship between two variables is Y=at+bXte where Y = the dependent variable (annual sales) X = the independent variable (here the year number) a= the value of Y when X = 0 (called the Y-intercept) and b = the rate of change of Y with respect to X (i.e., the slope of the line).Using the Model’s Parameters to Forecast * Use the model to forecast or “fit” values of Y to values of X.Calculate the errors * The next step is to calculate the errors. Errors are the differences between the data values of the dependent variable Y and the values of Y forecast by the model. Statisticians also call them deviations or residuals. By convention, errors are calculated by subtracting the forecast values from the data values rather than vice versa. Select cells and find the differences and copying the formula to the rest of the cells. Note that individual errors can be either positive or negative and their average is exactly zero.Validation of the Model Model validation is an important step and a valid model for projecting the past trend forward must be a valid model of the past. If a model does not follow the trend of the data for the past, itis not a valid model of the past—that is, it is, an invalid model because it does not satisfy the basic idea of using past data to forecast the future by projecting the past trend forward, Invalid models lead to invalid projections and should be rejected. Even valid models lead to erroneous projections if the past trend changes and projected values are not adjusted for the change. Using past history to forecast the future therefore consists of two cseeutal spe Y Creating a statistical model of the past that can be projected forward and, adjusting projected values for any changes in the past trend that can be anticipatedCont- For a model to be valid the data values must seater randomly above and below the trend line on a scatter diagram and the average error must be 2er0, ‘The key word heres random” and dsneton mis be made berwen a systematic anda random searing about the trend line. Analyzing the errors provides a more sensitive and useful test of randomness and model validity than simply looking at scatter diagram. Errors are the differences between the data and fitted values teas ‘Thus a valid model must satisfy each of two conditions: ‘The average error is zero and, the errors seatter randomly rather than systematically about the average of zero. “The first condition is automatically satisfied for the linear regression model ifthe spreadsheet entries are cortect, ‘The second condition may ot may not be satisfied. ‘Therefore, you need to check the extor pattern to see if its scatter is random of systematicCont- * One way to examine whether or not the errors are random or systematic is by scanning through the cells. * When the errors starts out largely positive get successively smaller and take on negative values as one moves down the column until a minimum value is reached near the center and then get successively larger as one continues to move down the column to a maximum at the end this is systematic behavior, not random, and the model is therefore not valid.Random error Random error is referred to as “noise”, because it blurs the true value (or the “signal”) of what’s being measured. Some common sources of random error include: natural variations in real world or experimental contexts. imprecise or unreliable measurement instruments. individual differences between participants or units. poorly controlled experimental procedures.Measuring the Model’s Accuracy * A model’s accuracy or precision is limited by the random scatter in the data, and a future value might be more than of less than what the model predicts. * Determining how far off the mark our forecasts might be is important for determining how much safety stock or how much extra service capacity we should provide to ensure that we are able to satisfy our customers promptly, for example, or to estimate the risk for breaking even on an investment. * To evaluate a model’s accuracy, we calculate its standard error of estimate.standard error of estimate ° SEE= af * The number of degrees of freedom equals the number of sets (the number of pairs) of data values minus the number of model parameters we've estimated in the model. The coefficient of correlation (R) is calculated by the entry =CORREL(Cell:Cell,Cell:Cell). This value for R is converted to the coefficient of determination (R?) by squaring,Using LINEST to Develop a Linear Regression Model Select 2 columns and 5 rows, type in the formula(1* row & 1* column cell): =LINEST(Y values;X values;Tab button[TRUE]) Simultaneously press Ctrl + Shift then ‘ENTER’ oo ee Standard error of the slope Standard error of the intercept R*-Coefficient of Standard error of the y determination: ‘estimate F-Statistic-determinates Degrees of freedom df (n- whether variable is by chance. parameters) L__PrSatiunmnieouscir___ Regression sum of squares (SS Residual sum of squares (6S of ”__| regression) residual)Quadratic Regression Models * Quadratic regression models are similar to linear regression models but include the square of the independent variable as a third term in the equations. Depending on one’s choice of symbols for the parameters, the general form of the equation for a quadratic regression model can be written as either * Y=atbX +cX? or Y=by +b, X+b,X? * The subscripts on the coefficients in equation indicate the power of the independent variable X associated with it.Cont- * Although mathematically, a quadratic equation expresses the dependence of the variable Y on a single independent variable, X, a spreadsheet, however, treats the squared value of independent variable as a second independent variable. Iris the squared value, of course, that deflects the trend line from a straight-line path. If its coefficient is positive, the deflection is upward to higher values of Y; if negative, the deflection is downward to lower values. Click on the data and repeat the steps for adding a linear trend line, but choose a second-order polynomial as shown in the Add Trendline dialog box. Use the option tab to display the equation and R-squared value on the chart.Using LINEST to Develop a Quadratic Regression Model Insert an additional Column for X? With the mouse, select appropriate Cells and type =LINEST(D6D16,B6:C16,1,1) (typical cells), and press Ctrl/Shift/ Enter. The result is shown in the 3-column/5-row matrix labeled “LINEST Output for Quadratic Model.” The values in Cells would be the values of ¢, b, and a. Use the model’s parameters to make forecasts, enter as follows: =$F$23+$E$23*B6+$D$23+C6 in Cell 6 and copy the entry, Calculate the errors by finding the difference. To validate the model, first show that the average error is zero by summing the errors then show that the errors scatter randomly rather than systematically (error pattern at lower tight).Cont- * Comparing the quadratic to the linear model, you should note, first of all, that the systematic behaviour of the errors with the linear model has been eliminated and the errors scatter more or less randomly about the mean value of zero. Comparison the quadratic regression model to the linear regression model, in terms of the standard error of estimate for the quadratic model and the coefficient of determination.LINEST output meaning EEE Ce Standard error of b2 R?-Coefficient of determination F Statistic-determinates whether variable is by chance. Regression sum of squares (GS regression) Standard error of bi Standard error of the model Degrees of freedom df (n- parameters) Residual sum of squares (SS of residual) Standard error of the intercept Standard error of the intercept ENIZ #N/AExponential Regression Model (follows FV model) A common business example of a curvilinear trend is the exponential curve for the increase in the value of money with time when the principal and interest are reinvested at a compound rate of interest. Curves for such behavior are concave upwards; the increase in value from one year to the next gets greater each year because of the increasing size ofthe accumlated funds on which interest is paid. The general equation for such behavior is: F=Pa+i* where F = future value, after n periods from the start P = present value of deposit or investment (ie., the value of F when n = 0) i= periodic rate of interest (eg, the annual rate of interest, compounded annually) and n = number of interest-bearing periods (eg,, the number of years in the future). ‘The formula is used in financial analysis to project present values to the future or, by using its inverse, to find the present value of a stream of future cash flows. Besides its use for calculating future financial values, equation is also useful for projecting many other business trends where the rate of change is a percentage of the base valueRe-writing the FV formula For the purposes of forecasting the equation can be rewritten as follows: Y = ABS Y is the dependent variable (for instance the annual sales revenue) whose value depends on X (where X is the year or year number). Comparing with the previous equation (F = P(1 + i)"), Y corresponds to the future value of money in the previous equation, and X to the number of interest-bearing periods. Equating X to zero, Y=A, the parameter B in equation would correspond to one plus the rate of increase per unit change of X—that is, to 1+i in the previous equation. Copy the spreadsheet for the linear model for analyzing the use of an exponential model to fit the data. Choose the exponential model, shown in the Add Trendline dialog box. Use the Option Tab to display the equation and R-squared value on the chart.Using LOGEST to Evaluate an Exponential Model’s Parameters of data values to an exponential regression equation, we need to evaluate the two s Aand B. Once we have determined the value of B, we can subtract one from it to find the relative rate of change (eg, the percentage rate of increase or decrease). Excel provides the LOGEST function for evaluating the parameters A and B in the exponential regression model. Its syntax (language rule) is: =LOGEST (range of known y values, range of known x values, 1, 1) ‘The parameters of the exponential model are evaluated by selecting Cells and entering the LOGEST tule.LOGEST Output meaning -Y = AB* B ig Standard error of B Standard error of A R2 Standard error of the y estimate FSuatistic Degrees of freedom Regression sum of squares (SS Residual sum of squares (SS of regression) residual)