Chapter 2 Terms/Definitions - Forecasting Demand

Demand forecasting

forecasting the demand for a particular good, component, or service

Simple Moving Average (time series method)

last 3 months divided by 3, very easy. This is a moving average because it is recalculated using the most recent set of months (or other periods), dropping the oldest month and adding the just-ended month to the list.

Weighted Moving Average (time series method)

places weights on the periods being averaged, usually to put greater emphasis on the more recent periods and relatively less emphasis on the more distant periods. When calculating the average, you divide by the sum of the weights rather than the number of periods. For example, if the third month out is given a weight of 1, the second month out a weight of 2, and the most recent month a weight of 3, you would divide this three-month weighted moving average by (1 + 2 + 3) = 6, for example: Thus it allows trends to have more of an impact on the forecast. The weights are usually selected using expert judgment and trial and error

Forecast error

the difference between actual and forecast demand. Forecast error is calculated using an absolute value of Actual - Forecast. To find as a percentage, just divide by the actual.

Tracking Signal

the ratio of the cumulative algebraic sum of the deviations between the forecasts and the actual values to the mean absolute deviation

Forecasting

the business function that attempts to predict sales and use of products so they can be purchased or manufactured in appropriate quantities in advance.

Mean Absolute Deviation (MAD)

A common way of tracking the extent of forecast error is to add the absolute period errors for a series of periods and divide by the number of periods. The resulting number you get, implies that the forecasts are off on average by about +/- x. Used in the bell curve to determine safety stock. 1 MAD vs 2 MAD vs 3 MAD

Leading vs Lagging Indicators

A leading indicator is a predictive measurement, for example; the percentage of people wearing hard hats on a building site is a leading safety indicator. A lagging indicator is an output measurement, for example; the number of accidents on a building site is a lagging safety indicator. Book defintions: Leading: a specific business activity index that indicates future trends. Examples: The level of consumer optimism about the economy (Consumer expectations often indicate future changes in spending.) Initial unemployment insurance claims (Initial claims for unemployment are more dependent upon business conditions than other unemployment metrics.) Lagging: are the economic and financial factors that reflect the changes that have already occurred in the economy Examples: Outstanding business and commercial loans (Demand for loans generally peaks about a year after a peak in the overall economy.) Changes in company profits (Decreases in profitability have a domino effect often felt by many members of a supply chain.)

Main principles of Forecasting

Almost always wrong (for this reason they require regular review) Should include an estimate of error (how far off from the average) Forecasts are more accurate for groups than for single items. Forecasts of near-term demand are more accurate than long-term forecasts.

Mean Squared Error (MSE)

Another method of calculating error rates, the mean squared error (MSE), magnifies the errors by squaring each one before adding them up and dividing by the number of forecast periods. Squaring errors effectively makes them absolute, since multiplying two negative numbers always results in a positive number.

Standard Deviation

Another way to calculate forecast error would be to use standard deviation, which is commonly provided in most software programs. An approximation for standard deviation when you know the MAD

associative forecasting (quantitative method)

Associative forecasting (also called causal, correlation, explanatory, or extrinsic forecasting) uses data gathered from one or more internal or external sources as a predictor of something that is presumed to be correlated. The predictor is called the independent variable. The element being predicted is called the dependent variable, and it could be demand for a product family or for total organizational sales. - Best for long term forecasts. Google: Associative forecasting models include identifying variables that can be useful in estimating another variable that has some type of association with each other

Bias (qualitative estimate)

Bias. Bias is "a consistent deviation from the mean in one direction (high or low). A normal property of a good forecast is that it is not biased." Bias exists when the cumulative actual demand differs from the cumulative actual forecast. Companies often measure it with Mean Percentage Error (MPE). This does not use absolute values as the + or - sign shows the direction of the bias. Any answer that does not result in zero reflects a bias. This can be the result of estimators may be motivated to estimate too high or too low depending on their incentives. For example, a salesperson could estimate too low to make a sales target easy to reach, while a culture of optimism might lead to aggressive sales goals.

Correlation versus Causation (referring to associative forecasting)

Correlation is an observation that the change in an independent variable has a measurable effect on a dependent variable. However, just because the effect can be reliably observed over time does not mean that the one thing caused the other thing. It could be that some third force is affecting both of them, or the correlation could be a coincidence that would be proven incorrect after a longer period of study. -> So, how do we tell whether there is enough correlation to use this predictor in our sales forecast? The statistical term that defines the strength of correlation is called the coefficient of correlation (r). Basically, it is a number between -1.0 and +1.0 where -1.0 is perfect negative correlation: An increase in the predictor causes an equal decrease in the predicted element. +1.0 is perfect positive correlation: An increase in the predictor causes an equal increase in the predicted element, and a decrease in the predictor causes an equal decrease in the predicted element. (They rise and fall together.) 0.0 is not correlated at all.

Judgmental/expert judgment forecasting (qualitative method)

Executives, salespersons, market analysts, and others can use their detailed knowledge of their products and their customers, along with their memory of the differences between prior forecasts and actual results, to generate a forecast or, more often, to adjust a quantitative forecast. You can track these adjustments to see if these modifications are increasing or decreasing accuracy and if they are showing bias being high or low.

Using Quantitative and Qualitative

Forecasts are best done with a combination of both. For example, a forecaster for an electronics company might use a mathematical model to estimate demand for future periods. The forecaster should then modify the projection with all pertinent and available intelligence. This could include knowledge of competitor sales promotions or product launches, the state of the economy, trends in discretionary spending, and so on.

MSE and MAD comparison

MSE gives a much wider range of error, giving you a more sensitive measure of the error rate which is useful if the absolute numbers are relatively close together and reduction of errors is important.

Simple Average (qualitative estimate)

One way to mitigate bias is to ask estimators to provide a pessimistic estimate, a most likely estimate, and an optimistic estimate. The three estimates can be combined and divided by three Qualitative because youre not actually using data, you are estimating them.

Deseasonalizing

Process of removing the seasonality. Step 1: This is done by first, finding a seasonal index: 1) Calculate the month average for each month. 2) Calculate the year average 3) Calculate the seasonal index - Divide each month average by the year average. Step 2: Apply the seasonal index to the raw data which will result in deseasonalized data: 1) Divide the raw data by the seasonal index for a given month (or year i would think).

Qualitative Forecasts (what are they and why use them)

Qualitative forecasts rely on judgment rather than math. These methods lack scientific precision but can be used on their own in volatile situations or when there are no historical data available, such as for a new product

Reseasonalizing

Reseasonalizing involves multiplying the deseasonalized data by the given period's seasonal index to find the seasonalized forecast values. Note that the deseasonalized data would normally not be shown in a chart at all since it is not useful for making decisions in and of itself. It is simply used to produce the reseasonalized forecast for demand planning.

Service Sector Forecasting

Services such as the resturaunt business have special forecasting challenges as their demand fluctuates so rapidly, such as hours rather than months. This requires sophisticated demand planning as they have to plan for aggregate demand but also the demand for each menu item.

Forecast Accuracy

Simple calculation of: 1 - Forecast error as a percentage.

Weighted Average (qualitative estimate)

The most likely estimate can be given more weight. For example, with the pessimistic estimate, most likely estimate and optimistic estimate, you can give more weight to the most likely estimate. A common way is to multiply the most likely result by four but then divide the total result by six Qualitative because youre not actually using data, you are estimating them.

Demand Planning (planning demand)

The purpose of forecasting is to engage in demand planning. the process of combining statistical forecasting techniques and judgment to construct demand estimates for products or services (both high and low volume; lumpy and continuous) across the supply chain from the suppliers' raw materials to the consumer's needs. Items can be aggregated by product family, geographical location, product life cycle, and so forth, to determine an estimate of consumer demand for finished products, service parts, and services. Numerous forecasting models are tested and combined with judgment from marketing, sales, distributors, warehousing, service parts, and other functions. Actual sales are compared with forecasts provided by various models and judgments to determine the best integration of techniques and judgment to minimize forecast error.

Mean absolute percentage error (MAPE)

There is a drawback to the MAD calculation in that it is an absolute number that is not meaningful unless compared to the forecast. Mean absolute percentage error (MAPE) is a useful variant of the MAD calculation because it shows the ratio, or percentage, of the absolute errors to the actual demand for a given number of periods

Quantitative Methods: Time-Series forecasting

Time series forecasting occurs when you make scientific predictions based on historical time stamped data. It involves building models through historical analysis and using them to make observations and drive future strategic decision-making. Less complex mathematically and easier to explain to decision makers. Time-series methods assume that the factors that influenced the past will continue on into the future. When that trend is unlikely to be stable, associative forecasting may be needed. Time series includes: Naive forecasting (Naive forecasting simply assumes that the last period's demand will be this period's forecast.) Simple and weighted moving averages and exponential smoothing Some steps: Visualizing, Deseasonalizing/Reseasonalizing

Factors Affecting Demand

Trends - a general direction in which a variable (eg Demand) is developing or changing. Trends can change direction at any time as a result of internal or external forces. Trends are long term shifts. Linear trends are upward sloping, neutral or negative, whereas exponential trends are skyrocketing up or down. Cycles and other external drivers - periodic upward, neutral, or downward shifts in demand lasting longer than one year. Primary example: economic cycles of recession and growth that form the wave pattern. These could be one of the causes of the trend. External drivers are economic cycles as well, population growth, major events. Seasonality - Demand may fluctuate depending on the time of year, e.g., holidays, weather, or other seasonal events. Seasonality is a predictable repetitive pattern of demand measured within a year where demand grows and declines. These patterns are calendar related and can appear annually, quarterly, monthly, weekly, daily and/or hourly. Can be hours, weeks, first day of the month, weekends, etc) -> Side note: Differences between seasonality and cycles: seasonality is a demand pattern that, based on history, will repeat itself on a calendar basis such as month, week, day of the week, hour of the day, etc., and therefore can be predicted. Cycles are demand patterns that repeat but follow a wavelike pattern that can span multiple years and may change at any time; therefore, they cannot be predicted easily. Promotions or other internal drivers - Promotions ex: Discounts/Advertising. Internal Drivers Ex: deals to gain favorable product placement. For many industries, promotions can explain 50 to 80 percent of sales variation. They are worked into forecasts using associative forecasting, for example, by using marketing spend as a driver. Random (irregular) variation - fluctuation in data that is caused by uncertain or random occurrences. Random variation is what is left after seasonality is removed. Not predictable and cannot be explained. The idea is to minimize this component by finding more and more explainable factors. -> Variations are categorized as either common causes (general cause) or special cause (assignable cause). Random variation is a common cause. These are many small factors that affect demand but cannot be added to the model due to need for simplicity.

Exponential Smoothing (time series method)

Used 3 inputs: the last periods forecast, the last periods demand, and a smoothing constant. Smoothing constant: a # greater than 0 and less than 1, represented by the Greek letter alpha (α), which is basically a percentage weighting where 1 = 100 percent. The alpha is multiplied by the last periods demand, so the higher the alpha, the more emphasis you are putting on last periods demand over the last periods forecast. The constant value is selected by experience, trial and error, and testing against historical data. Most companies use .3.

Mix forecast

forecast of the proportion of products that will be sold within a given product family, or the proportion of options offered within a product line...Even though the appropriate level of units is forecasted for a given product line, an inaccurate mix forecast can create material shortages and inventory problems.

Delphi Method (qualitative Method)

surveying experts and collating their responses into a document that keeps the responses anonymous, until there is a consensus thought. This removes group think (independent thinkers becoming committed to others opinion) and stake in the ground mentality (when someone publicly has an opinion and doesn't want to lose face).

Simple Regressions (also called linear regression) (associative forecasting)

uses a formula to make an association between the dependent variable y (the element being predicted) and the independent variable x (the predictor), with two other elements, alpha and beta. Beta (β) is the slope, which is a value used as a multiplier to find the correct placement of the forecast result. Alpha (α) is the intercept, which is where the slope intercepts 0 on a chart. In this example. Starting at 0 housing starts with what looks to be a previous years sales of 600k. The only value then needed is x which is the value of each housing start, then the regression line is formed. Formula is y = a+ Bx Example: Roofing Sales (the predicted) = a (starting point) + (B * Prior Months Housing Starts (the predictor)).