Prediction and Judgmental Adjustments of Supply-Chain Planning in Festive Season

Table of contents

1. I. Introduction

upply-chain holds a huge planning propaganda as a baseline to project sales and revenue generation based on it. Specially a case of modern era, where the comfort to customer can help an organization to retain the customer and thereby increase more sales and generate more business of the products. In addition to it, the planned resource production prediction helps generating less of the cost of production and more of the effort on quality productivity. Supply chain managements involve huge planning horizon for demand forecasting. For the purpose they use forecast systems for initial forecast followed by judgmental adjustment by the company experts to adjust exceptional events in the planning process. The manual adjustments made raise questions related to improvement of accuracy and type of adjustments made. Effective Short-term forecasting is important for improving supply chain management [26], irrespective of the type of business. Multiple applications of the prediction analysis and adjustment behavior in prediction accuracy can be seen in past few years [14]- [16], [18]- [20]. According to the literature of economic forecasting, accuracy of the statistical decisions can be improved when experts consider the changes in the statistical models according to the changes coming from occurrence of special events [1]- [8], [10], [11] and [22]- [23] showed that the suggested judgmental adjustments tend to improved accuracy marginally but may also introduce bias. Since it's a human added knowledge as a judgement factor, it is more likely to make error in level of adjustment and make room for error as experimental evidences suggest [24]- [26]. Forecasters make decisions on the basis of noisy and randomly fluctuating events in time series [9].

Several methods, techniques have been used in literature to forecast load demands. We used exponential smoothing technique and trend fitting for prediction.

This study presents effect of seasonal demand on prediction methodology of above mentioned models using reference data of handlooms business sector for predicting shipment load for four different Handlooms companies. The proposed methods are used to predict one month's demand. The outcome of both models is analyzed and accuracy is compared.

2. II. Time Series Models

A time series is sequential nature of data produced during a certain period of time. Assuming no major disrupting to critical parameters of a recurring event, the future prediction is always related to past data. Two time-series analysis models, namely, multiplicative decomposition and the smoothing technique use the dependency of future data to the past events, and model the behavior as follows:

3. a) Smoothing Techniques

Smoothing techniques are used to smoothen out random variations in the data due to irregular components of the time series. They provide a clearer and better view of data and it is easy to understand. 1) Moving averages: A moving average (MA) is an average of the data provided for certain number of time period. The method is called "moving" because it is obtained using summing and averaging the values from a given number of periods say n, each time deleting the oldest value and adding the new one. The moving average is calculated as:

MA t+1 =? ???1 ??=1(1)

Where t= current period.

4. c) Trend and seasonal component

To occupy the Seasonal festive pattern, time series decomposition model breaks down, analyzes and forecasts the seasonal and the trend components. The method is often referred as Time series decomposition, since the technique is analyzing seasonal indexes after decomposing the series in order to identify seasonal components called as seasonal indexes. These helps deseasonalize the series. This deseasonalized series helps in projecting trend projection line. Lastly, seasonal indexes are used to seasonalize the trend projection. [27].The steps involved are as follows:

1. Identify the quarters, months etc. and calculate centered moving averages (CMA).

5. III. Experimentation

The data is input to both the methods with and without tuning the seasonal effects. In order to fit the seasonal component, extent of seasons is fixed for a month's duration. For example, Ludhiana manufacturers tend to see a huge impact on sale during Diwali, Baisakhi etc. Data is selected and analyzed from four Ludhiana-based handloom manufacturers. For a better accuracy rate, last three years data is analyzed. In the given market trend of last three years, Each festive

6. a) Data

The data is collected for the festive season's months of Punjab for Ludhiana based four Handlooms manufacturers for the last three years. Table 1 is organized structure of observed shipment load for the festive season of Lohri (Jan), Holi (March), Vaisakhi (April), Rakhsha-Bandhan (August), Krwachauth(Sep-Oct), Diwali and E-id (Oct-Nov), Guru-Nanak Jayanti (Nov) and finally Christmas(Dec) for all four handlooms. Along with these values, table1 also contains one last entry as observed value of Lohri (Jan '17) .

The graphical representation of the observed data along with its linear trend fitting is shown in graph1. The graph shows observed shipment lad of all four handlooms over the seasonal period of last three years along with one last entry as observed shipment load of Jan'17 which is value of interest here.

7. b) Results

All three Smoothing averages and trend fitting with and without tuning the trend effect are applied on the data collected and outcome is predicted for festive season of Lohri (Jan '17). The results are compared with already observed value for Lohri (Jan '17).

8. i. Smoothing Technique

? Moving averages: Here the moving average (MA) is an average of the data provided for observed shipment load of all four Handlooms for festive seasons of past three years. Graph 2 shows Observed Vs Forecasted shipment load with 3period -moving average for the festive season of past three years for Ludhiana based four Handlooms supply-chain companies. The graph illustrates that with 3 period moving average the next forecasted value that is Jan'17 reduced than observed value. Where as in graph 3 shows with the 4-period moving average the forecast increases. handlooms in the market hence, adjusted exponential smoothening (F t ) adj is obtained as a result. Therefore using eq 3, trend adjusted forecast is calculated with ? = 2/(n+1) i.e.= 2/(27+1) = 0.1, and initial T t = 0 and ?= 0.1. The adjusted forecast in table4 gives final (F t ) adj =2280.922 which is close to simple exponential smoothing without any tuning for trend effects F t =2312.219 Accuracy of forecast is better judged by finding mean square error for different values of smoothing constant ? (0< ??1). In order to get which smoothing factor gives better result, comparison between forecasts for ?=0.1 and ?=0.8 is shown in table 5. Result shows for ?=0.8 is relatively gives more root mean square error hence less accurate forecast for large data set. It is observed that the data with larger fluctuations over the period of time more than a year does not predict accurate using exponential smoothing. ii. Trend Fitting: Using eq6 the model can be fitted using table1 data. To occupy the Seasonal festive pattern, time series decomposition model breaks down, analyzes and forecasts the seasonal and the trend components. The given data set has distinct nine seasons hence the forecast is effected by the trend and seasonal component. Table 6 shows before and after seasonal and trend decomposition effect comparison of trend fitting. The forecasted value comes out to be T = 2378and 2322 resp. for Jan'17.

The accuracy so measure for all the methods applied are shown in table 7. Accuracy is compared by calculating MSE and RMSE of all the forecasts so far applied in this work. The lesser the RMSE better is the forecast. As shown in table 7, Trend fitting after trend deseasonalization gives least RMSE and hence is the best forecast seen.

9. IV. Conclusion

The techniques used in this case study shows following results based on forecast and the measure of error based on MSE and RMSE:

10. SEASONS IN ALL THREE YEARS

11. O B S E R V E D S H I P M E N T L O A D ( U N I T S P E R P A C K O F H A N D L O O M S ) ( Y )

Prediction and Judgmental Adjustments of supply-Chain Planning in Festive Season

? Using Weighted moving averages, given the understanding of the owner of the sales-market head, the weights assigned to the various months of certain period 'n' hugely impacts the forecast value accuracy level. In the given data set, the weights assigned by the stakeholder proves to give best outcome of all forecasts. WMA is the best suited outcome for the dataset.

12. SEASONS IN ALL THREE YEARS O B S E R V E D S H I P M E N T L O A D ( U N I T S P E R P A C K O F H A N D L O O M S ) ( Y )

Prediction

Figure 1. Table 2
2
shows 3-month
Figure 2. Table 1 :
1
Year 1 2 3
Observed Observed Observed
Festive Month Season (t) Shipment Load (Units per pack of Festive Month Season (t) Shipment Load (Units per pack of Festive Month Shipment Load (Units per pack of
Handlooms)(Y ) Handlooms)(Y) Handlooms)(Y)
(Jan'14) 1 2500 (Jan'15) 1 2200 (Jan'16) 1 2300
(March'14) 2 1130 (March'15) 2 1145 (March'16) 2 1130
(April'14) 3 2200 (April'15) 3 2500 (April'16) 3 2400
(August'14) 4 2250 (August'15) 4 2300 (August'16) 4 2250
(Sep-Oct'14) 5 3450 (Sep-Oct'15) 5 3400 (Sep-Oct'16) 5 3350
(Oct-Nov'14) 6 3000 (Oct-Nov'15) 6 2800 (Oct-Nov'16) 6 3000
(Oct-Nov'14) 7 3330 (Oct-Nov'15) 7 2850 (Oct-Nov'16) 7 3150
(Nov'14) 8 1100 (Nov'15) 8 1200 (Nov'16) 8 1400
(Dec'14) 9 1700 (Dec'15) 9 1950 (Dec'16) 9 1900
Year 4 (Jan'17) 1 2250
? Weighted moving Average (WMA): The expert
planner/ analysts of the companies decide to weigh
the past three month's sales. WMA calculated using
Note: average weightage given to past values for the combined data is shown in table 3. Using observed Shipment load for the last three months from table1, WMA is calculated for festive season of Lohri (Jan'17) as follows: WMA for Lohri'17= 2233.33.
Figure 3. Table 2 :
2
Year 2017 ) ? Exponential smoothing Technique: Since trend is expected out of festive season demands of the
Observed Shipment Load
( Festive Month Year Season(t) (Units per pack of Handlooms) MA 3 MA 4 CMA 4
(y)
(Jan'14) 1 2500
(March'14) 2 1130 1943.333
(April'14) 3 2200 1860 2020 2138.75
(August'14) 4 2250 2633.333 2257.5 2491.25
(Sep-Oct'14) 1 5 3450 2900 2725 2866.25
(Oct-Nov'14) 6 3000 3260 3007.5 2863.75
(Oct-Nov'14) 7 3330 2476.667 2720 2501.25
(Nov'14) 8 1100 2043.333 2282.5 2182.5
(Dec'14) 9 1700 1666.667 2082.5 1809.375
(Jan'15) 1 2200 1681.667 1536.25 1711.25
(March'15) 2 1145 1948.333 1886.25 1961.25
(April'15) 3 2500 1981.667 2036.25 2186.25
(August'15) 4 2300 2733.333 2336.25 2543.125
(Sep-Oct'15) 2 5 3400 2833.333 2750 2793.75
(Oct-Nov'15) 6 2800 3016.667 2837.5 2700
(Oct-Nov'15) 7 2850 2283.333 2562.5 2381.25
(Nov'15) 8 1200 2000 2200 2137.5
(Dec'15) 9 1950 1816.667 2075 1860
(Jan'16) 3 1 2300 1793.333 1645 1795
Figure 4. Table 3 :
3
.Season (t) Weights (w) Values (y) weights*value Festival
Last Month 1/2 1900 950 Christmas
Two months ago 1/6 1400 233.33 Gurunanak Jayanti
Three Months ago 1/3 3150 1050 Eid
Forecasted value 2233.33 Lohri(Jan'17)
Observed shipment load for the festive season of 2016 for Ludhiana based four Handlooms supply-chain
companies.
OBSERVED SHIPMENT LOAD (UNITS PER PACK OF HANDLOOMS)
1000 2000 3000 4000 Graph 2: 0 Observed Shipment load (Jan'14) (March'14) (April'14) (August'14) (Sep-Oct'14) (Oct-Nov'14) (Oct-Nov'14) (Nov'14) (Dec'14) SEASONS IN ALL THREE YEARS (Jan'15) (March'15) (April'15) (August'15) (Sep-Oct'15) (Oct-Nov'15) (Oct-Nov'15) (Nov'15) (Dec'15) (Jan'16) (March'16) (April'16) (August'16) (Sep-Oct'16) (Oct-Nov'16) (Oct-Nov'16) (Nov'16) (Dec'16) Jan'17
SHIPMENT LOAD 0 500 1000 1500 2000 2500 3000 3500 4000
OBSERVED
Figure 5. Table 4 :
4
SHIPMENT LOAD 0 1000 2000 3000 4000
OBSERVED
with such huge variation. Setting ?=0.8 gives poor
forecast of the observed shipment load of Lohri
(Jan'17).
? The trend adjustments made in data due to
seasonal effect of festivals dramatically improves
projection using trend line projection. RMSE
comparison between other techniques and trend
projection shows trend projection with trend
? Due to large variation of shipment load in various deseasonalization gives best of all results and
sequential months, Exponential smoothing closest forecast to actual observation.
technique could not predict better results for data
Festive Month Year Season(t) Observed Shipment Load (Units per pack of Handlooms) (y) Old Forecast F t-1 New Forecast (F t = F t-1 + 0.1(y t-1 -F t-1 )) Adjusted forecast (F t ) adj = F t + (1-?)/? * T t
(Jan'14) 1 2500 2500 2500 2500
(March'14) 2 1130 2500 2363 2239.7
(April'14) 3 2200 2363 2346.7 2221.06
(August'14) 4 2250 2346.7 2337.03 2215.251
(Sep-Oct'14) 5 3450 2337.03 2448.327 2438.893
(Oct-Nov'14) 6 3000 2448.327 2503.494 2544.654
(Oct-Nov'14) 7 3330 2503.494 2586.145 2697.575
(Nov'14) 8 1100 2586.145 2437.53 2404.064
(Dec'14) 1 9 1700 2437.53 2363.777 2267.28
(Jan'15) 1 2200 2363.777 2347.4 2245.812
(March'15) 2 1145 2347.4 2227.16 2027.515
(April'15) 3 2500 2227.16 2254.444 2099.319
(August'15) 4 2300 2254.444 2258.999 2123.487
(Sep-Oct'15) 5 3400 2258.999 2373.099 2353.828
(Oct-Nov'15) 6 2800 2373.099 2415.789 2436.867
(Oct-Nov'15) 7 2850 2415.789 2459.211 2517.259
(Nov'15) 2 8 1200 2459.211 2333.289 2272.204
Figure 6. Table 5 :
5
Year 2017
28
)
(
Figure 7. Table 7 :
7
Error measure Smoothing techniques Exponential smoothing MA4 Moving averages MA3 ?= 0.1 ?= 0.8 Normal Trend fitting Deseasonalized
SSE 21470355.56 15913144.53 16949382.13 22536138.72 15230927.11 6568052.846
MSE 795198.3539 589375.7234 627754.8936 834671.8046 564108.4115 243261.2165
RMSE 891.7389494 767.7080978 792.309847 913.6037459 751.0715089 493.2151828
Forecasted value 2150 2544 2358 1745 2378 2322
1
2

Appendix A

Appendix A.1

Appendix B

  1. Factors affecting the nature and effectiveness of subjective revision in sales forecasting: An empirical forecast selection. A Diamantopoulos , B P Mathews . Journal of Forecasting 1989. 9 p. .
  2. Forecasting for the ordering and stock-holding of spare parts. Ahc Eaves , B G Kingsman . J Opl Res Soc 2004. 55 p. .
  3. Managerial intervention in forecasting: An empirical investigation of forecast manipulation. B P Mathews , A Diamantopoulos . International Journal of Research in Marketing 1986. 3 p. .
  4. Alternative indicators of forecast revision and improvement. Marketing Intelligence and Planning, B P Mathews , A Diamantopoulos . 1987. 5 p. .
  5. Judgmental revision of sales forecasts -a longitudinal extension. B P Mathews , A Diamantopoulos . Journal of Forecasting 1989. 8 p. .
  6. Author's retrospective on 'Forecasting seasonals and trends by exponentially weighted moving averages. Charles C Holt . International Journal of Forecasting January-March 2004. 20 (1) p. .
  7. Decomposition for judgemental forecasting and estimation. D Macgregor . Principles of forecasting, J S Armstrong (ed.) (Norwell, MA
    ) 2001. Kluwer. p. .
  8. The role of judgment in macroeconomic forecasting. D S Turner . Journal of Forecasting 1990. 9 p. .
  9. Exponential smoothing: The state of the art -Part II. E S Gardner . International Journal of Forecasting 2006. 22 p. .
  10. Receiving other people's advice: Influence and benefit. I Yaniv . Organizational Behavior and Human Decision Processes 2004. 93 p. .
  11. Judgmental adjustment of initial forecasts -its effectiveness and biases. J S Lim , M Connor . Journal of Behavioral Decision Making 1995. 8 p. .
  12. Judgmental forecasting with time series and causal information. J S Lim , M Connor , R L Goodrich . International Journal of Forecasting 1996. 2000. 12 p. . (International Journal of Forecasting)
  13. Conducting a sales forecasting audit. M A Moon , J T Mentzer , C D Smith . International Journal of Forecasting 2003. 19 p. .
  14. Judgmental forecasting: A review of progress over the last 25 years. M Lawrence , P Goodwin , M O'connor , D &onkal . International Journal of Forecasting 2006. 22 p. .
  15. Evaluating the role judgment plays in forecast accuracy. M R Donihue . Journal of Forecasting 1993. 12 p. .
  16. Why are judgments less consistent in less predictable task situations?. N Harvey . Organizational Behavior and Human Decision Processes 1995. 63 p. .
  17. Understanding the use of forecasting systems: An interpretive study in a supply-chain company, P Goodwin , W Y Lee , R Fildes , K Nikolopoulos , M Lawrence . 2007. Bath University Management School working paper
  18. Principles of Forecasting: A Handbook for Researchers and Practitioners, P S Fader , Bgs Hardie . Armstrong J.S. (ed.) 2001. Norwell, MA: Kluwer. (Forecasting trial sales of new consumer packaged goods)
  19. Telecommunications demand forecasting-A review. R Fildes . Int J Forecasting 2002. 18 p. .
  20. Against your better judgment? How organizations can improve their use of management judgment in forecasting. R Fildes , P Goodwin . Interfaces 2007. 37 p. .
  21. Effective forecasting and judgmental adjustments: an empirical evaluation and strategies for improvement in supply-chain planning. R Fildes , P Goodwin , Lawrence M Nikolopoulos , K . Int J Forecasting 2008. 24.
  22. Effective forecasting and judgmental adjustments: An empirical evaluation and strategies for improvement in supply-chain planning. R Fildes , P Goodwin , M Lawrence , K Nikolopoulos . International Journal of Forecasting 2009. 25 p. .
  23. CHAN4CAST: A multichannel, multiregion sales forecasting model and decision support system for, S Divakar , Ratchford Bt , V Shankar . 2005. p. .
  24. Short-Term transaction forecasting using time series analysis: A case study for India. S Sharma , M P Bhatia . Advances in Information mining, 2012. 4 p. .
Notes
1
© 20 7 Global Journa ls Inc. (US) 1
2
© 20 7 Global Journa ls Inc. (US)
Home 
Date: 2017-01-15