Intro to Supply Chain Chapter 2
Other Forecasting Models
1. Drift Method 2. Holt's Linear Trend Method 3. Holt-Winters Seasonal Method 4. Autoregressive Integrated Moving Average (ARIMA) 5. Box-Jenkins 6. X-11 7. Econometric Model 8. Input-Output Model
Exponential Smoothing is well accepted for the following reasons
1. Exponential models are surprisingly accurate. 2. Formulating an exponential model is relatively easy. 3. The user can understand how the model works. 4. Little computation is required to use the model. 5. Computer storage requirements are small. 6. Tests for accuracy are easy to compute.
The Forecasting Process (not resp. for this on test)
1. Identify the purpose of the forecast. 2. Select the items to be forecasted. 3. Determine the time horizon of the forecast. 4. Select a forecast model that seems appropriate for the item. 5. Collect the data needed to create the forecast. 6. Generate the forecast. 7. Is the accuracy of accuracy of the forecast the forecast acceptable? 8a. If yes, forecast over the planning horizon. 8b. If no, select a new forecast model, or adjust the model, or adjust the parameters of the parameters of the existing model (go back to step #5). 9. Adjust forecast based on additional qualitative information and insight. 10. Monitor results and measure the forecast accuracy
Two important considerations about a forecast
1. Statistically speaking, the forecast will be inaccurate, and although it may be inaccurate, it is still useful. 2. The forecast is the basis for most "downstream" supply chain planning decisions, so it is critical to be as accurate as possible. Good forecasting can benefit a company by facilitating more effective planning, which can lead to reduced inventories, reduced costs, reduced stockouts, and improved customer service. However, bad forecasting can be the root cause for creating just the opposite. There is a familiar adage that applies to forecasting: "garbage in = garbage out." If a forecast is bad, everything else (i.e., the supply plan) based on that forecast will also be bad.
Fundamentals of Forecasting
1. Your forecast is most likely wrong 2. The more "granular" the forecast, the less accurate it is. 3. It is easier to forecast next month more accurately than it is to forecast next year. 4. Simple forecast methodologies trump complex ones. 5. A correct forecast doesn't prove your forecast method is correct. 6. If you don't use the data regularly, trust it less when forecasting 7. All trends will eventually end (that's why they're called trends). 8. It's hard to eliminate bias, so most forecasts are biased. 9. Technology is not the solution to better forecasting. 10. Forecasting is really a blend of art and science.
Cause and Effect Forecasting
2 basic cause-and-effect models: simple linear regression, and multiple linear regression. These are a statistical measure that determines the strength of the relationship between one dependent variable and a series of independent variables. The first step is to identify the historical relationship between an independent and a dependent variable to predict the future values of the dependent variable, i.e., demand.
Collaborative Planning, Forecasting, & Replenishment (CPFR)
A business practice that combines the intelligence of multiple trading partners who share their plans, forecasts, and delivery schedules with one another in an effort to ensure a smooth flow of goods, and services across a supply chain. It can significantly reduce the Bullwhip Effect, and provide a plethora of benefits including better customer service, lower inventory costs, improved quality, reduced cycle time, and better production methods. It also requires a fundamental change in the way that buyers, and sellers work together. The real value of CPFR comes from the sharing of forecasts among firms, rather than firms relying on sophisticated algorithms and forecasting models to estimate demand.
Forecast bias
A consistent deviation from the mean in one direction; can be high, or low. In other words, it exists when the demand is consistently over, or under the forecast. A good forecast is not biased. ∑ Forecast Error = ∑ Actual Demand - ∑ Forecast Demand. If the sum of the forecast error is not zero, there is bias in the forecast. A negative result shows that actual demand was consistently less than the forecast, whereas a positive result shows that actual demand was greater than forecast demand. A forecast process with bias will eventually create significant problems in the supply chain if left unchecked. Good supply chain planners are aware of these biases. A best practice is to measure for forecast bias routinely, and then make corrections accordingly.
Historical Analogy
A judgmental forecasting technique based on identifying a sales history that is analogous to a present situation, such as the sales history of a similar product, and using that past pattern to predict future sales. It is an approach to sales forecasting in which the past sales results of a similar product are used to predict the likely sales of a similar new product. There is potential to provide a significant amount of info that can be used, at least initially, to create a forecast for a new product. It can also be a relatively inexpensive way to create a forecast. However, there may not be a historical comparison available, and no two historical situations are exactly identical in all respects, so it may prove to be ineffective.
Simple moving average
A mathematical result that is calculated by averaging a number of past data points. It uses a calculated average of historical demand during a specified number of the most recent time periods to generate the forecast. It is best for short-term forecasting, and is useful when demand is not growing or declining rapidly, and no seasonality is present. It also smooths out random variations. However, it fails to identify trends, or seasonal effects. It will create shortages when demand is increasing, and excess inventory when demand is decreasing, as it lags behind actual demand.
Trend variations
A movement of a variable over time. Identification of trends is a common starting point when developing a forecast. It is more easily observed by plotting actual demand on a graph over time to see whether there is a pattern in the data. Common trend types include linear, S-curve, asymptotic, and exponential.
Forecast Accuracy - Tracking Signal
A simple indicator that forecast bias is present determines if the forecast is within acceptable control limits, and provides a warning when there are significant unexpected departures from the forecast. If the tracking signal falls outside the pre-set control limits, there is a bias problem with the forecasting method, and an evaluation of the way forecasts are generated is warranted (Think of a smoke detector. It is preset to allow for a certain range of smoke, but beyond that range the alarm (tracking signal) goes off and warns you). Tracking Signal = RSFE / MAD.
Linear trend forecast
A simplistic forecasting technique that imposes a line of best fit to time series historical data. It is imposing a best fit line across the demand data of an entire time series. It is also used as the basis for forecasting future values by extending the line past the existing data, and out into the future while maintaining the slope of the line. It can provide an accurate forecast into the future even if there is random variation. However, seasonal and cyclical variations are softened, making this method more useful for annual forecasts than for monthly forecasts. (In Microsoft Excel, select a data range and select "Add Trendline" from the dropdown menu.)
Naïve forecasting
A technique in which the last period's actuals are used as this period's focus without adjustment. It sets the demand for the next time period to be exactly the same as the demand in the last time period. It works well for mature products with stable demand, and it very easy to determine. However, it works well for mature products only because any variations in demand will create inventory issues.
Weighted moving average
A technique that puts more weight on recent data, and less on past data through the use of a weighting factor. It is similar to a simple moving average except that not all time periods are valued/weighted equally. It is more accurate than a simple moving average if actual demand is increasing or decreasing, and allows unequal weighting of prior time periods. Although it is better than a simple moving average, this technique will still lag behind actual demand to some degree, and is more inconvenient, and costly than exponential smoothing.
The more "granular" the forecast, the less accurate it is.
An annual forecast for a product family (i.e., group of related products), is likely to be more accurate than a weekly forecast for an individual item within that product family. Similarly, a national forecast for an item is likely to be more accurate than the individual regional forecasts for that item.
Forecast
An estimate of future demand.
Random variations
Are instability in the data caused by random occurrences. These random changes are generally very short-term, and can be caused by unexpected, or unpredictable events such as weather emergencies, natural disasters, labor strikes, war, etc. (Ex's: spikes in demand for; wood to repair house damage; tree clean up services; water damage restoration, all resulting from the impact of a hurricane). Random variations are frequently considered to be abnormal demand, and are sometimes removed from the data set before the data set is used to generate new forecasts.
Seasonal variations
Are repeating patterns of demand that occur within one year. The pattern can be repeated from year to year, with some periods of considerably higher demand than others (Ex's: holiday shopping, swim suits sales in the summer, snow shovel sales in the winter, building construction increasing in the summer and slowing down in the winter by region).
Simple Linear Regression
Attempts to model the relationship between a single independent variable, and a dependent variable (demand) by fitting a linear equation to the observed data. The equation describes the relationship between the independent variable, and dependent variable as a straight line.
Multiple linear regression
Attempts to model the relationship between two, or more independent variables, and a dependent variable (demand) by fitting a linear equation to the observed data. Depending on the data and the number of independent variables, the mathematics involved can be complex.
Qualitative forecasting
Based on opinion, and intuition. They are generally used when data are limited, unavailable, or not currently relevant. The best use is for long-range forecasts, and for new-products. Forecast depends on skill and experience of forecaster(s), and available info. The 5 qualitative models used are personal insight, jury of executive opinion, Delphi Method, historical analogy, and customer survey.
Delphi Method
Basically the same as the Jury of Executive Opinion except that the input of each of the participants is collected separately so that people are not influenced by one another. This is done in several rounds until a consensus forecast is achieved. Decisions are enriched by the experience of competent experts, and are not likely a product of groupthink. They are very useful for new products. However, experts may introduce some bias, and companies must spend time and resources collecting data by survey. If external experts are used, there is a risk of loss of confidential information. The Delphi Method can be time-consuming, and is therefore best for long-term forecasts.
The Bullwhip Effect
Because customer demand is not perfectly stable, businesses must forecast demand in order to properly position inventory and other resources. Forecasts are based on statistics, and they are rarely 100% accurate, therefore, companies often carry an inventory buffer called safety stock. Moving from the end-consumer(s) backward across the supply chain to raw material supplier(s), each supply chain participant is farther removed from the end demand and may have less information about what is happening with demand, creating a greater need to maintain higher levels of safety stock. In the absence of any other info, or visibility, individual supply chain participants are second-guessing what is happening with ordering patterns, and potentially over-reacting, creating the bullwhip effect. In periods of rising demand, down-stream participants increase orders, whereas in periods of falling demand, orders decrease or stop, and inventory accumulates. When a small demand ripple in the market place is felt by the retailer at the end of the supply chain, the retailer will then start adjusting their orders to the wholesalers, and the wholesaler in turn will adjust its orders to the distributor, the distributor to the factory, and so on back up the supply chain. When the new demand reaches the material or components supplier at the other end of the supply chain, the magnitude of fluctuation becomes unrecognizable. An overreaction due to uncertainty occurs throughout the entire supply chain.
A correct forecast doesn't prove your forecast method is correct
Because it could've just been by chance. If you only question your methods when there is a large variance in the data, you'll miss all those times your forecast was just lucky.
Impact of Social Media on Forecasting
Companies are using social media insights to drive improvements in forecasting. Analysis of social sentiment can be used to: 1. Evaluate the health of a brand- an understanding of how your target market feels about your company, products and services. 2. Improve demand prediction- companies can use the Voice of the Customer (VOC) to drive improvements in forecasting and inventory positioning. 3. Address a crisis - social sentiment analysis might reveal a spike in negative posts, and provide an early warning to a potential product or service issue. 4. Research the competition- social sentiment analysis can help you understand how to position against the competition.
Customer survey
Customers are directly approached, and asked to give their opinions about the particular product. Customer surveys can be done in person (e.g., one-on-one, focus group), over the phone, by mail, email, or online. It is a direct method of assessing info from the primary sources, and is simple to administer and comprehend. Also, it doesn't introduce any bias, or value judgment particularly in the census method if the questions are constructed carefully. However, poorly formed questions may lead to unreliable info. Customers don't always answer the questionnaire, and it is time-consuming and costly to survey a large population.
Demand planning process (not resp. for this on test)
Demand Planning requires a holistic process that includes the following steps: 1. Profiling products to determine the appropriate treatment. 2. Validating of qualitative forecasts. 3. Estimating the magnitude of unmet demand. 4. Predicting underlying causal factors 5. Development of the quantitative forecast 6. Development of a consensus expectation 7. Planning for the commercialization of new products. 8. Coordinating of demand shaping requirements 9. Determining the confidence level of the expected demand. 10. Collaborating with customers on future requirements 11. Monitoring actual sales and adjusting the demand plan. 12. Identification of sources of forecast inaccuracies.
Dependent demand
Demand for an item that is directly related to other items or finished products, such as a component or material used in making a finished product. Demand for these items is calculated.
Independent demand
Demand for an item that is unrelated to the demand for other items, such as a finished product, a spare part, or a service part. Demand for these items is forecasted.
Forecast Error Value and Forecast Error %
Forecast Error Value = Actual demand - forecast demand. The error can be quantified as an absolute value, or as a percentage. Ignore than + and - signs. (Ex. Actual Demand = 95. Forecast Demand = 110. Forecast Error Value = 95 - 110 = -15 units. Forecast Error % = (Forecast error value / Actual demand) * 100 = (-15 / 95) * 100 = 15.8% (moved two places to right and rounded))
The Role of Forecasting
Forecasting is a vital function that impacts every significant management decision. Finance, and accounting use forecasts as the basis for budgeting and cost control. Marketing relies on forecasts to make key decisions such as new product planning, and personnel compensation. Production uses forecasts to select suppliers, determine capacity requirements, and drive decisions about purchasing, staffing, and inventory.
Forecasting Horizon (time frames)
Forecasting over the short-term, medium-term and long-term serves different purposes. In the short-term, forecasting is less than three months- is used mainly for tactical decisions. In the medium-term, forecasting is between three months to two years- it is used to detect general trends, and identify major turning points. In the long-term, forecasting greater than two years- it is used to detect general trends, and identify major turning points.
It is easier to forecast next month more accurately than it is to forecast next year.
If we know what we sold last month, we typically have a good idea what we will sell next month. However, 12 months from now, a lot of things can happen that might can affect sales, so forecasting next year accurately is less probable.
How to Choose a Forecasting Method
In order to determine what forecasting method to use, answer the following questions: 1. How will the forecast be used? 2. How critical is the need for accuracy? 3. What is the forecast horizon? 4. How often will our forecasts be forecasts be reviewed/revised? 5. Is the necessary data available, and accurate? 6. How predictable is the product the product demand? 7. Are there independent factors that affect the that affect the product? 8. What is the level of aggregation or disaggregation? 9. At what point in the product lifecycle is the product?
Personal insight
In this, the forecast is based on the insight of the most experienced, most knowledgeable, or most senior person available. Sometimes, this approach is the only option, but methods that include more people are generally more reliable. It is the fastest and cheapest forecasting technique, and can provide a good forecast. However, it relies on one person's judgement and opinions, as well as their prejudices and ignorance. The major weakness is unreliability- someone who is familiar with the situation often provides a worse forecast than someone who knows nothing.
Types of Demand - Bicycle Example
Independent Demand- bicycle (finished product). Dependent Demand- frame, seat, handle bar, wheels, tires, pedals(component parts).
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 results in a positive number. (MSE = ∑ (A-F) ^2 / n, where a = actual demand, F = forecast demand, and n = number of time periods.)
All trends will eventually end
Many factors will affect the pattern you're trying to forecast. It doesn't matter how accurately you predict the trend. In the future, the variables will change, and the forecast will be wrong.
Measures of Forecasting Accuracy - MAPE
Measures the size of the error in percentage terms. It is calculated as the average of the unsigned percentage error. Many companies use the MAPE, as it is easier for most people to understand forecast error and forecast accuracy in percentage terms rather than in actual units. It is also 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. (MAPE = ∑ ((|A - F|)/ A)) / n (expressed as a percentage), where A = Actual demand, F = Forecast demand, and n = number of time periods (Note: In parentheses is an absolute value sign).)
Measures of Forecasting Accuracy (MAD)
Measures the size of the forecast error in units. It is calculated as the average of the unsigned, i.e., absolute, errors over a specified period of time. Absolute errors for a series of time periods are added together, and then divided by the number of time periods (finding the mean). The resulting value is the MAD measure of forecast inaccuracy. Whether the forecast is over or under, the actual demand is irrelevant, and only the magnitude of the deviation matters in the MAD calculation. (MAD = ∑(IA - F|) / n, where A = Actual demand, F = Forecast demand, and n = Number of time periods (Note: In parentheses is an absolute value sign).)
Jury of Executive Opinion
People who know the most about the product, and the marketplace would likely form a jury to discuss, and determine the forecast. Generally, the panel conducts a series of forecasting meetings to discuss the forecast until the panel reaches a consensus agreement. Decisions are enriched by the experience of competent experts, and companies don't have to spend time and resources collecting data by survey. However, experts may introduce some bias, or may become biased by their colleagues or a strongly opinionated leader.
Running Sum of Forecast Errors (RSFE)
Provides a measure of forecast bias, indicates the tendency of a forecast to be consistently higher, or lower than actual demand. A positive RSFE indicates that the forecasts were generally too low, underestimating the demand. Stock-outs are likely to occur as companies are unable to meet customers' actual demand. However, a negative RSFE indicates that the forecasts were generally too high, overestimating demand. Excess inventory and higher carrying costs are likely to occur. (RSFE = ∑e , where e = forecast error for period t t t)
The 2 basic forecasting techniques used in most businesses today
Qualitative, and quantitative forecasting.
Technology is not the solution to better forecasting
Robust forecasting comes from sound logic in the methodology. Create an appropriate strategy and then use technology to make it more successful. Technology is not the answer, it's a tool to help you make the forecast better.
Forecast Error
Since forecasts are almost always inaccurate, companies need to track the forecast against actual demand, and measure the size and type of the forecast error. The size of the forecast error can be measured in units, or percentages. Error measurement plays a critical role in tracking forecast accuracy, monitoring for exceptions, and benchmarking the forecasting process. Interpretation of these statistics can be tricky, particularly when working with low-volume data, or when trying to assess accuracy across multiple items.
Forecasting and Demand Planning Challenge
Supply chain organizations routinely rank demand planning immaturity as a major obstacle in meeting their supply chain goals. Accurate Forecasts are the foundation for profitable business growth. Optimal demand planning, and forecasting requires comprehensive modeling capabilities plus the flexibility to shift methods as product life cycles progress, and market conditions change. A best-in-class forecasting system provides flexibility for users to weight elements, and override key parameters in the forecast calculation based on their knowledge, and market expertise.
Forecasting
The business function that estimates future demand for products so that they can be purchased, or manufactured in appropriate quantities in advance of need. It is the process of mathematically predicting future demand, and uses historic data to determine the direction of future trends. In the simplest terms, it is the attempt to predict future outcomes based on past events, and management insight.
Weighted Moving Average: Selecting Weights
The challenging part of using a weighted moving average is deciding on the weight for each time period. Experience, and/or trial-and-error are the simplest approaches. Recent past is often the best indicator of the future, so weights are generally higher for more recent data. If the data are seasonal, weights should reflect this appropriately (Ex: To forecast swimsuit sales for August, the sales in July should be weighted more heavily than the sales from last December).
Forecasting and Demand Planning
The key building blocks from which all supply chain planning activities are derived and are crucial components of customer satisfaction. The first step is forecasting, where the forecast is developed through data analysis and judgement. Organizations must have a formal forecasting process to develop an agreed upon set of numbers that becomes the driver for demand planning. The second step is demand planning, which is the process of combining statistical forecasting techniques, and/or judgment to construct demand estimates for products or services. Management, and other experts within the company review the forecast to ensure that it is aligned with the company's strategy, and make adjustments if necessary, using market intelligence which may be known to them, but outside the scope of the forecasting model.
Time Series Forecasting
The main purpose of a time series model is to collect, and study the past data of a given time series in order to generate probable future values for the series. In other words, forecasts for future demand rely on understanding past demand. Accordingly, time series forecasting can be characterized as the act of predicting the future by understanding the past. The 5 types of time series models are naïve, simple moving average, weighted moving average, exponential smoothing, and linear trend.
Demand
The need for a particular product, or component. It could come from various sources such as a customer order, a forecast, the manufacturing of another product, etc. 2 types: independent demand, and dependent demand.
Forecasting error
The only thing that we can say for sure about a forecast is that the forecast will likely be wrong, the best that we can hope for is to be as consistently accurate as possible. Therefore, the goal of the forecasting and demand planning process is to minimize forecast error. The factors that influence demand must also be considered when forecasting (market changes, seasonality, competitive activity, pricing, changing consumer preferences, etc.). Generally, the farther out into the future you forecast, the greater the deviation will likely be.
Your forecast is most likely wrong
The question is, "How wrong is the forecast?" Forecasting is difficult mainly because people know it is likely to be wrong, and nobody likes to be publicly and visibly wrong. These fundamentals of forecasting in business can be easily forgotten at times, to the detriment of the quality and accuracy of your forecasts. You must be willing to recognize, and adapt to changing conditions- don't fall in love with your forecast, and ignore evidence that it may be wrong. Just be open to the first signs of change, and be prepared to react quickly and decisively.
Simple forecast methodologies trump complex ones
There is danger in complexity. Complicated forecast methods often hide key assumptions built into the model. On the other hand, simple forecast methods are easy to understand, analyze and work out why it went wrong.
How can the Bullwhip Effect be Alleviated?
There is no single remedy, but there are some actions that supply chain participants can take collectively: 1. Collaboration- sharing info through the use of electronic data interchange (EDI), point of sale (POS) data, and web-based systems can facilitate collaboration. 2. Synchronizing the supply chain- supply chain participants coordinate planning, and inventory management to minimize the need for reactionary corrections. 3. Reducing inventory- through the use of just in time (JIT), vendor managed inventory (VMI), and quick response (QR). If the various participants work together to get closer to customers through collaborative planning, forecasting, and replenishment (CPFR), then the bullwhip effect can be greatly reduced.
Forecasting is really a blend of art and science
Today, the best practice is a combination of qualitative, and quantitative methods in addition to collaboration with supply chain partners, providing better visibility downstream in the demand chain.
Quantitative forecasting
Uses mathematical models, and historical data to make forecasts. 2 types: 1. Time series- based on the assumption that the future is an extension of the past. Historical data is used to predict future demand. Time series are the most frequently used among all the forecasting models. 2. Cause and effect- assumes that one or more factors predict future demand. It is generally recommended to use a combination of quantitative, and qualitative techniques.
Cyclical variations
Wavelike pattern that last longer than 1 year, and can extend over multiple years. They are not easily predicted. Examples include business cycles, China growth, GDP, bull markets, and bear markets.
Exponential smoothing
Weighs past observations with exponentially decreasing weights to forecast values. It is a more sophisticated version of the weighted moving average, and requires 3 basic elements: last period's actual demand, last period's forecast, and a smoothing factor, which is a number greater than 0 and less than 1 (used as a weighting percentage). The smoothing factor is not a given. It has to be determined based on the best judgment of a company's experts. Although exponential smoothing will create a forecast more responsive to trends than previous methods, it will still lag behind trends, especially upward trends since the smoothing factor would need to be greater than 1.0 to approach an accurate forecast.
Variations in Quantitative Forecasting
When creating a quantitative forecast, data should be evaluated to detect for the following variations: trend variations, random variations, seasonal variations, and cyclical variations.
If you don't use the data regularly, trust it less when forecasting
When info is not regularly used, errors often remain undetected. Regular use of data helps identify mistakes, and smooths out inconsistencies over time.
It's hard to eliminate bias, so most forecasts are biased
When you have to make a range of assumptions (which factors to include, how strongly to weight them etc.), it's likely that you will be adding some bias to the forecast.