Naïve, & Moving Average
Simple Slope
Ted Mitchell
Naïve Forecast
• Simplest possible forecast • Tomorrow will be like today
• Naïve is the basis for comparison of all methods.
• Ignores any historical data previous to today
8 7 6 5 4 3 2 ? ? 1 10 0 E=A-F F A error Naïve Forecast Actual Year
Use today’s result to forecast tomorrow
8 7 6 5 4 3 2 10 ? 1 10 0 E=A-F F A error Naïve Forecast Actual Year
Use today’s result to forecast tomorrow
N = 8 8 7 6 5 4 3 2 2 10 12 1 10 0 E=A-F F A error Naïve Forecast Actual Year
Use today’s result to forecast tomorrow ME = 1.63
SE =13 23 ? N = 8 2 21 23 8 2 19 21 7 2 17 19 6 1 16 17 5 1 15 16 4 1 14 15 3 2 12 14 2 2 10 12 1 10 0 E=A-F F A error Naïve Forecast Actual Year
Momentum
• If things have momentum they are easier to predict.
• Averages are a measure of momentum • Various averages are used for prediction
– Total historical average – Moving averages – Weighted averages
Total Historical Average
• Fi+1 = (1/n)(ΣAi) where
• Fi+1 = forecast for next period • n = number of historical periods
• ΣAi = sum of the actual results for each of the n historical periods
? Error 9 8 7 6 5 4 3 10/1 = 10 ? 2 10 1
Forecast using average of the total history actual
period
Get the forecast for the next period
? 2 Error 9 8 7 6 5 4 (10+12)/2 = 11 ? 3 10/1 = 10 12 2 10 1 forecast actual period
Get the forecast for the next period
7.5 6.3 5 3.6 3.2 2.2 3 2 Error (10+12+14+15+16+17+19+21)/8 =15.5 23 9 (10+12+14+15+16+17+19)/7 =14.7 21 8 (10+12+14+15+16+17)/6 =14 19 7 (10+12+14+15+16)/5 =13.4 17 6 (10+12+14+15)/4 =12.8 16 5 (10+12+14)/3 =12 15 4 (10+12)/2 = 11 14 3 10/1 = 10 12 2 10 1 forecast actual period
Get the forecast for the next period
Using Total Historical Average
• Disadvantage
• Really lags behind a trend! because
– Uses all historical data
– Puts equal weight on every piece of historical information
Moving Average
• Pick the last n periods that are most relevant • Fi+1 = (1/n) (ΣAi)
where
• Fi+1 =the forecast for next period • n = the number of periods in the moving
average
• (ΣAi) = the sum of the last n periods
? Error 9 8 7 6 5 (10+12+14)/3 =12 ? 4 14 3 12 2 10 1
Forecast using a moving average on last 3 periods actual period ? 3 Error 9 8 7 6 (12+14+15)/3 =13.7 ? 5 (10+12+14)/3 =12 15 4 14 3 12 2 10 1
Forecast using a moving average on last 3 periods actual period 2.3 3 Error 9 8 7 (14+15+16)/3 =15 ? 6 (12+14+15)/3 =13.7 16 5 (10+12+14)/3 =12 15 4 14 3 12 2 10 1
Forecast using a moving average on last 3 periods actual period 4 3.7 3 2 2.3 3 Error (17+19+21)/3 =19 23 9 (16+17+19)/3 =17.3 21 8 (15+16+17)/3 =16 19 7 (14+15+16)/3 =15 17 6 (12+14+15)/3 =13.7 16 5 (10+12+14)/3 =12 15 4 14 3 12 2 10 1
Forecast using a moving average on last 3 periods actual
period
Moving Average
• Still lags behind a trend
• Puts equal weight on each of the historical results being used
• Gives bias when seasonal data is involved • If you want more weight on the most recent
Weighted Moving Average
• Three period average with equal weight • Fjun = (Amar +Aapr + Amay ) / 3or
• Fjun = (3Amar +3Aapr + 3Amay ) / 9 • Weighted average with more on May • Fjun = (2Amar +3Aapr + 4Amay ) / 9 • Naïve Again
• Fjun = (0Amar +0Aapr + 9Amay ) / 9 3.6
3.3 2.8 1.8 2 2.6 Error (2(17)+3(19)+4(21))/9 = 19.4 23 9 (2(16)+3(17)+4(19))/9 = 17.7 21 8 (2(15)+3(16)+4(17))/9 = 16.2 19 7 (2(14)+3(15)+4(16))/9 = 15.2 17 6 (2(12)+3(14)+4(15))/9 = 14 16 5 (2(10)+3(12)+4(14))/9 = 12.4 15 4 14 3 12 2 10 1
Forecast using a WEIGHTED moving average on last 3 periods
actual period
Weighted Moving Average
• Weighted Moving Average is better at responding to a trend because it puts more weight on recent data and less weight on old data
• They get the appropriate weights by doing a statistical fit to the data
Simple Growth and Slope For
Trends
Ted Mitchell
Trends
• Trends in the data are not handled well by moving averages or exponential smoothing methods.
Simple Trend Projection
• Before the era of simple statistical tools on every PC managers used simple calculations of trends based on the naïve forecast. • The naïve forecast is
sales in next period t = sales in the last period (t-1) or
Do a Forecast for Period 2
time 0 1 2 Sales Revenue Last period Naïve Forecast Two Periods AgoLast period plus X Percent
• The sales in the last period plus the percentage growth over the last two periods
Do a Forecast for Period 2
time 0 1 2 Sales Revenue Last period Naïve Forecast Two Periods Ago Simple Percentage Projection uses the same growth as between the last two periods
g
Getting the slope
• The percentage growth over the last two periods = g
• Prediction for the last period would be R1 = g R0
• We know R1 and R0 so we can calculate g
Example Calculate historical g
• R1 = g R0 where • R0 = $150 • R1 = $175 then calculate g • 175 = g(150) • g = 175 / 150 • g = 1.17 or growth is 117%g = Growth rate between 0 and 1
time Sales Revenue Last period Naïve Forecast Two Periods Ago
Simple last period plus percent growth Projection uses the same slope as the last two periods
Prediction of Revenue in period 2
• (Revenue in period 2) = g (Revenue in period 1) • R2 = gR1 Where • R1 = revenue in 1 = 175 • g = 1.17 Then R2 = 1.17(175) = 204.17Prediction for R2
time 0 1 2 Sales Revenue R1 = 175 Naïve Forecast = 175 R0 = 150 R2 = 1.17(175) = 204 g = 1.17Prediction for R2 in period 2
time 0 1 2 Sales Revenue R1 = 175 Naïve Forecast = 175 R0 = 150 R2 = 204.17 g = 1.17
The Problem with
• The “last period result + percent improvement” method
• Very dependent on the base used in the percentage. If you use the same percentage as time passes then the method inflates the forecasted values
• But it is simple and very popular!
Examples: Naïve Method
& Last Period Plus Rate of
Change Method
Ted Mitchell
New Shoes Home Market Spring 478
• Home Market in this example is
experiencing a long run decline in sales as it nears the end of the Product Life Cycle
7 6 5 1,193,000 4 1,193,000 3 E=A-F F A error Naïve Forecast Actual Units Sold Period
Use today’s result to forecast tomorrow
7 6 5 170,000 1,193,000 1,023,000 4 1,193,000 3 E=A-F F A error Naïve Forecast Actual Units Sold Period
Use today’s result to forecast tomorrow
7 6 1,023,000 ? 5 170,000 1,193,000 1,023,000 4 1,193,000 3 E=A-F F A error Naïve Forecast Actual Units Sold Period
Use today’s result to forecast tomorrow
What to do Next?
• You have two pieces of information • Industry Sales in period 3 = 1,193,000 • Industry Sales in period 4 = 1,023,000 • And the idea that the market is in decline
phase of the Product Life Cycle (PLC) • Do you naïve or last period + decline %
Last period + change %
• Consider the last period plus the decline rate from the two previous periods
• What is the decline rate
• Sales in 4 = decline rate (Sales in 3) 1,023 = decline rate (1,193 )
• Decline rate = 1,023 / 1,193 = 85.75%
Forecasting period 5
• Sales in 5 = decline rate (sales in 4) • Sales in 5 = 85.75% (1,023,000) • Sales in 5 = 877,225 units
7 6 877.225 ? 5 170,000 1,193,000 1,023,000 4 1,193,000 3 E=A-F F A error Naïve Forecast Actual Units Sold Period
Use last period and decline rate to forecast period 5 7 6 Smallest error is naive 877.225 or the naïve method 1,023,000 1,000,000 5 170,000 1,193,000 1,023,000 4 1,193,000 3 E=A-F F A error Naïve Forecast Actual Units Sold Period
Use last period and decline rate to forecast period 5 or naïve method
7 Smallest error is last period + decline rate 977,517 or the naïve method 1,000,000 885,000 6 Smallest error is naive 877.225 or the naïve method 1,023,000 1,000,000 5 170,000 1,193,000 1,023,000 4 1,193,000 3 E=A-F F A error Naïve Forecast Actual Units Sold Period
Use last period and decline rate to forecast period 5