Forecasting v01 - Lecture PDF

Preview

Summary

Lecture...

Description

Forecasting

Forecasting is an attempt to predict the future. An organisation provides outputs needed in the environment and receives certain returns in exchange. For example, a bicycle manufacturer sells its products for money. In order to most efficiently conduct this process, the organisation needs to know not only what to produce but also when, how many and at what price. Forecasts are generally used for 3 purposes in organisations: 1. To decide if demand is sufficient to generate the desired results. For example, a demand may exist but at too low a price to cover costs. 2. To determine long term capacity requirements. An accurate projection of demand for a number of years into the future will enable capacity to be tailored to meet future demand. This will minimise the chances of:  Being inefficient due to excess capacity.  Having insufficient capacity to meet demand. 3. To ascertain short term fluctuations in demand for determining the master schedule. Popular forecasting methods can be classified as per Figure 1 below:

Figure 1 Forecasting Methods

Qualitative (sometimes called judgmental) forecasting methods are often used for long range forecasts. They are usually used where historical data is limited or non-existent, such as in new product or new service introductions. The Delphi technique utilises several experts in the area to be forecast. Instead of bringing these experts together and having them influence one another, they are approached separately and individually. The forecasts of these experts are summarised and fed back to each contributor, who then revises their forecast in the light of the feedback. After several revisions it is hoped that a consensus can be reached. This method allows the benefits of multiple opinions but avoids the pitfalls of dominant behaviour and stubbornness to change one’s mind. New products and services are often subjected to extensive market research before a final decision is made regarding their introduction. Telephone and personal interview surveys as well as mail questionnaires can provide information about current attitudes and behaviour, about past actions, and about intentions for the future. Consumer panels (made up of paid or

1

EOM_19/20_TrimB_W2

volunteer participants) are provided with new and sometimes competing products, and after using them are asked to feedback their opinions about them. Test marketing involves introducing the product or service into a specific limited geographic area. This provides information about actual consumer behaviour rather than consumer attitudes, opinions or intentions. In scenario planning, the possible outcomes are identified and analysed to determine their probability and develop contingency plans for them.

Quantitative Time Series Analysis models try to predict future occurrences based on an analysis of historical data. We need a reliable set of information about the past, i.e. a sequence of observations taken at regular intervals over a period of time (hourly, daily, weekly, monthly etc.). A knowledge of past behaviour will help our understanding of (and therefore our ability to predict) future behaviour. Time series analysis assumes that history will repeat itself and past tendencies will continue. There are 4 patterns of variation (Figure 2).

Figure 2 Demand patterns

Horizontal pattern shows not change in time. The data cluster about a horizontal line Trend is the long term direction of demand. For example, there may be a steady change in demand, or a constant percentage change in demand. Many new products follow a “stretched s“ growth curve, which is characterised by a relatively slow start up, a period of rapid product acceptance and then a slowdown in the rate of adoption as the market becomes saturated. Seasonal fluctuations result primarily from nature but are also brought about by human behaviour. Sun cream and ice lollies enjoy brisk demand during the summer months, whereas warm coats and hats peak in the spring and summer months. Seasonal variation need not be tied to the seasons of the year for example traffic congestion is dictated by 9 to 5 working hours.

2

EOM_19/20_TrimB_W2

Cycles can be defined as long term oscillations or swings about the trend line over a period of at least 3 years. Economic cycles of prosperity and depression, and periods of war and peace are examples of such cycles. When the data do not show any specific assignable cause and without pattern, we say it is random. Random variations can sometimes be explained after they have happened, such as an increase in fuel consumption during a public transport strike.

Forecasting techniques Table 1 Sells of a pumps manufacturer

YEAR

QUARTER

PERIOD NUMBER

NUMBER OF PUMPS SOLD

2016

1 2 3 4 1 2 3 4 1 2 3 4 1 2

1 2 3 4 5 6 7 8 9 10 11 12 13 14

3,500 8,000 5,500 10,000 4,500 6,000 3,000 5,500 5,000 9,500 7,500 15,000 13,500 17,500

2017

2018

2019

The simplest (naïve) forecasting technique is to use the last period’s demand as a predictor of the next period’s demand. However, this can produce erratic forecasts. Take, for example, the data presented in Table 1. This shows, for an engineering manufacturer, the numbers sold of a particular range of pumps. Based on a demand of 3,500 in period 1, to predict a demand of 3,500 for period 2 would be a gross underestimate. Likewise, a forecast of 8,000 for period 3 would be a gross overestimate. Let us look at using the average of all the demand data to predict the next period’s demand. For the first quarter of 2017 (or period 5) we can calculate a demand of 6,750 ([3,500 + 8,000 + 5,500 +10,000] / 4) when the actual demand was 4,500. For period 10 we can calculate a demand of 5,666 (the sum of the first 9 periods divided by 9) and 9500 occurred. We can see that this averaging method smooths out the fluctuations but does not adequately respond to any growth or reduction in the demand trend. The moving average technique generates the next period’s forecast by averaging the demand for only the last n (usually between 4 to 7) time periods. Any data older than n are thus ignored. For example, to forecast demand for period 15 with n being 4:

Demand = (7,500 + 15,000 + 13,500 + 17,500) / 4 = 13,375 The moving average technique is a compromise between the previous two techniques.

3

EOM_19/20_TrimB_W2

The weighted moving average technique is where the forecaster assigns more weight to certain time periods. Normally, recent demand data is weighted so as to influence the forecast more than older demand data. The n period weighted moving average is calculated as:

Weighted Moving Average = Wt Dt + Wt−1 Dt−1 + …..+ Wt−(n−1) Dt−(n−1) Where: W is the weight, decided by the forecaster D is the demand 0 ≤ W ≤ 1.0 Wt + Wt−1 + …..+ Wt−(n−1) = 1.0 Let us take the same example as above to forecast demand for period 15. So demand for the last 4 quarters is:

D11 = 7,500; D12 = 15,000; D13 = 13,500; D14 = 17,500. The demands are weighted as follows:

W11 = 0.1; W12 = 0.2; W13 = 0.3; W14 = 0.4 Therefore, the 4 period weighted moving average for D t+1 or D15 is:

Demand (D15) = D11 W11 + D12 W12 + D13 W13 + D14 W14 = (7,500 * 0.1) + (15,000 * 0.2) + (13,500 * 0.3) + (17,500 * 0.4) = 750 + 3,000 + 4,050 + 7,000 = 14,800 Exponential Smoothing Sophisticated weighted moving average that gives recent demands more weight than earlier demands, all the way back to the first period in the history file.

𝐹𝑡+1 = 𝛼𝐷𝑡 + ሺ1 − 𝛼ሻ𝐹𝑡 Ft+1 = forecast for period t+1 Ft = this period forecast Dt = this period actual demand α = smoothing parameter (between 0 and 1.0) Example: weeks 48 49 50 51 52 1

mobile phones 858 814 871 1255 980

Period t-4 t-3 t-2 t-1 t t+1

Dt = 980 – actual demand reported by the mobile phone sales department α = 0.1 – defined by the mobile phone sales department Ft = this period forecast -> moving average 4 weeks = (1255+871+814+858)/4 = 949.5 ≈ 950

𝐹𝑡+1 = ሺ0.1ሻሺ980ሻ + ሺ1 − 0.1ሻ ∗ 950=98 + 855 𝐹𝑡+1 = 953

4

EOM_19/20_TrimB_W2

Causal forecasting concentrates upon the reasons for and the relationships between the factors that influence demand. A variable that is determined by some other factor or factors is referred to as a dependant variable. An independent variable is one whose values are determined outside of the system being modelled. Independent variables are used in causal forecasting to forecast values of a dependant variable. Causal forecasting assumes that the dependant variable to be forecast has a relationship with one or more independent variables. For example, the forecaster may want to predict demand for a product (dependant variable) as a function of these independent variables:  Advertising expenditure for the product.  Price of the product.  Consumer income. Linear regression The forecaster determines the underlying relationship between the dependant variable and the independent variable(s) and then uses this relationship to forecast future values of the dependent variable. 𝑌 = 𝑏𝑋 + 𝑎 Where: Y = the dependent variable to be predicted a = the y-axis intercept b = the slope of the regression line, and X = the independent variable used to predict Y Example How much money will you spend in advertising in month 6 if the expected sales are 300 000 units? Advertising Sales Month σ Δ𝑦 𝑏= (£1000s) (1000s units) σ Δ𝑥 x y dx dy 1 2.5 264 2 1.3 116 -1.2 -148 3 1.4 165 0.1 49 4 1 101 -0.4 -64 5 2 209 1.0 108 6 ? 300 sum -0.5 -55 b 110 y = 110x – 8 300=110x-8 x=(300+8)/100 x=2.8 => £2800

5

EOM_19/20_TrimB_W2...