d id you k now
play

D ID YOU K NOW ? Predicting (forecasting) the weather plays an - PowerPoint PPT Presentation

Jan 03, 2023 •772 likes •1.54k views

F ORECASTNG AND R SK M ANAGEMENT P ROFESSOR D AVID G ILLEN (U NIVERSITY OF B RITISH C OLUMBIA ) & P ROFESSOR B ENNY M ANTIN (U NIVERSITY OF W ATERLOO ) Istanbul Technical University Logistic Management in Air Transport Air Transportation

  1. F ORECASTİNG AND R İSK M ANAGEMENT P ROFESSOR D AVID G ILLEN (U NIVERSITY OF B RITISH C OLUMBIA ) & P ROFESSOR B ENNY M ANTIN (U NIVERSITY OF W ATERLOO ) Istanbul Technical University Logistic Management in Air Transport Air Transportation Management Module 9 M.Sc. Program 18 December 2015

  2. L EARNING O BJECTIVES • Understand the following concepts: FORECASTING Why do we forecast? How do we forecast? What is a good forecast? RISK MANAGEMENT Risk Estimation Risk Evaluation Risk Identification Strategies: Allocating Risk/Risk Pooling 2

  3. D ID YOU K NOW ? • Predicting (forecasting) the weather plays an enormous role in the world of advertising and marketing • In the world of marketing, retailers have always had a fundamental knowledge of weather because they had to navigate conditions to transport products from manufacturing plants to retail locations. – “Sears spotted a pattern in their auto parts department. They realized that car batteries more than five years old tend to die after three consecutive nights of sub-zero temperatures - so they began to place ads on the day after the third freeze” 3

  4. W EATHER I NFLUENCES A DVERTISING WHICH INFLUENCES D EMAND & L OGISTICS • By crunching data and sales results from hundreds of categories, the Weather Channel has the ability to spot patterns. – For example , it learned that bug repellent sells well in Dallas during the spring when there was a below-average dew point (the temperature at which dew begins to form) - but in Boston bug repellent only sells when the dew-point is above average. – The Weather Channel discovered that the first day of above- average heat in Chicago results in a surge of air-conditioner sales. But in Atlanta, people will sweat it out for two days before making the same run to the appliance store. 4

  5. T HE FASCINATING "P ROFIT OF O NE D EGREE " LIST . S OURCE : WXTRENDS . COM • when the temperature drops one degree colder - soup sales go up 2%. • When the temp goes up one degree - beer and soft drink sales go up 1.2%. • One degree colder in the fall equals a 4% increase in children's apparel sales. • Just one degree hotter in summer translates to a 10% increase in sun care products. • Just one degree colder in the fall means a 25% increase in mousetrap sales. • And just one degree hotter in summer - just one degree - results in a 24% increase for air conditioners. 5

  6. W HAT IS F ORECASTING ? • A ‘statistical’ estimate of future demand, that can be used to plan current activities • Often based on past sales (activity), while considering issues like seasonality, trends in demand, etc 6

  7. O PERATIONS AND I NFORMATION @W AL -M ART • Wal- Mart manages one of the world’s largest data warehouses • Wal-Mart tracks sales, inventory, shipment for each product at each store – Wal- Mart’s demand forecasting system tracks 100,000 products, and predicts which products will be needed in each store • The data warehouse is made available to store managers and suppliers 7

  8. F ORECASTING IS V ITAL Finance and Accounting: Marketing Forecasts provide the basis for Relies on sales forecasting to plan budgetary planning and cost control new products and promotions Production and Operations Strategic Planning: Use forecasts to make decisions Forecasting is one of the basis for involving capacity planning, process corporate long-run planning selection and inventory control 8

  9. F ORECASTING P RICE OF O IL 9

  10. H OW DO WE FORECAST ? E XAMPLE 1 • You are working for BIM • Your first assignment Determine the number of units of the latest iPad model to order for Christmas sales • How would you approach this problem? 10

  11. H OW DO WE FORECAST ? Q UALITATIVE M ETHODS Delphi Method Executive Judgment 1. Choose experts to participate Based on experience and representing a variety of points of history view 2. Obtain forecasts (and reasoning) from all participants 3. Summarize the results and Market Research redistribute them to the participants Surveys, interviews, etc along with appropriate new questions 4. Summarize again, refining forecasts and conditions, and again develop new questions to distribute to all Panel Consensus participants Meetings of executives, 5. Repeat the previous step as necessary salespeople and customers and distribute the final results. 11

  12. I LLUSTRATION OF D ELPHI P ROCESS 12

  13. H OW DO WE FORECAST ? E XAMPLE 2 • You are working for MIX (a bakery in Vancouver) • Your first assignment Determine the number of units of bread to order for next week’s sales • How would you approach this problem? • What about unit of whole wheat bread? How is this a different problem? 13

  14. H OW DO WE FORECAST ? Q UANTITATIVE M ETHODS Time Series Analysis Times series forecasting models try to predict future based on past data Some common approaches o Moving averages o Exponential smoothing 14

  15. A T YPICAL T IME -S ERIES OF P AST D EMANDS Seasonal Variation x x x Linear x x Trend x x x x x x x x x x x Sales x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x 1 2 3 4 Year 15

  16. S IMPLE M OVING A VERAGE : E XAMPLE Week Demand 1 650 The average can be taken over a 2 678 number of weeks of previous data 3 720 4 785 5 859      A A A A      1 2 3 6 920 t t t t n F t n 7 850 8 758 9 892 n-period moving average forecast 10 920 for period t 11 789 12 844 16

  17. S IMPLE M OVING A VERAGE : E XAMPLE Week Demand 3-Week Moving 6-Week Moving Average Forecast Average Forecast 1 650 N/A N/A 2 678 N/A N/A F 4 =(650+678+720)/3 =682.67 3 720 N/A N/A 4 785 682.67 N/A 5 859 727.67 N/A F 7 =(650+678+720+785+859+920)/6 =768.67 6 920 780.00 N/A 7 850 854.67 768.67 8 758 876.33 802.00 9 892 842.67 815.33 17

  18. E XAMPLE OF F ORECASTING IN R EVENUE M ANAGEMENT M ODELS Revenue Data Historical Booking Data Actual Bookings No Show Data Forecasting Model Optimization Model Overbooking Model Recommended Booking Limits 18

  19. P ICKUP OR S TANDARD F ORECASTING 19

  20. P LOTTING THE M OVING A VERAGES 1000 950 900 850 Demand 800 3-Week MA 6-Week MA 750 700 650 600 1 2 3 4 5 6 7 8 9 10 11 12 • Which is a better forecast? • How many past weeks should we consider? 20

  21. W HAT IS A G OOD F ORECAST ?   forecast error forecast v alue actual value  F  A t t • The smaller the errors, the better the forecast • One approach is to evaluate a forecast method is to compute the mean absolute deviation : Why not just MEAN deviation? n   F A What is the ideal MAD? t t   1 t MAD Sold Out Newsvendor example n What does a high MAD indicate? 21

  22. C ALCULATING THE MAD • Using some forecasting method, you have the following forecasts. What is the MAD? Month Demand Forecast Abs Error 1 220 2 250 255 5 3 210 205 5 4 300 320 20 5 325 315 10 n    F A    t t 5 5 20 10    1 t 10 MAD 4 n 22

  23. W HICH FORECAST HAS THE LOWER MAD? 800 950 900 750 850 800 700 750 700 650 650 600 600 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Demand 3-Week MA 6-Week MA Demand 3-Week MA 6-Week MA The 6-week moving average has The 3-week moving average has a lower error a lower error 23

  24. T ESTING THE R OBUSTNESS OF A F ORECAST : T HE T RACKING S IGNAL The running sum of forecast errors (RSFE) considers the nature of the error. Tracking signal is a measure that indicates whether the forecast average is keeping pace with any genuine upward or downward changes in demand. 24

  25. H OW MANY PERIODS SHOULD BE USED ? Disadvantages of More Advantages of More Periods Periods • More data points give a better • A large number of observations estimate will cause the moving average • The effect of randomness is to respond slowly to permanent reduced by averaging together changes a number of observations • When there is a trend in the • When there is no trend in the data, using more observations data, using more observations results in a forecast with high results in a forecast with lower error error 25

  26. O THER W AYS OF I NTRODUCING T REND • Weighted Moving Average – Place a greater weight on more recent observations           F w A w A w A w A     1 1 2 2 3 3 t t t t n t n W k = Weight given to the     period that is k periods  1 w w w 1 2 n ago (Weights must add to one) • Exponential Smoothing (very common) 26

  27. E XPONENTIAL S MOOTHING             ( 1 ) ( ) F A F or F F A F      1 1 1 1 1 t t t t t t t F t Forecast for period t More recent F t-1 Forecast for period t-1 observations are given A t-1 Actual demand in period t-1 more weight α Parameter (between 0 and 1) 27

  28. F ORECASTS OF I NBOUND L ADEN C ONTAINERS (TEU) 1400000 1200000 1000000 800000 600000 400000 200000 0 Month 2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12 2 4 6 8 10 12 2 4 Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 28

Recommend

More recommend