Demand Forecasting Part I and II

Demand forecasting

Process in which managers predict demand (sales) expected of goods or services over a future time period.

Economic Forecasts

Address business cycle: inflation rate, money supply, housing starts, etc.

Technological Forecasts

Predict the rate of technological progress. Impacts development of new products.

Uses for Forecasts

Design the System, Use of the System, Schedule the System

What are the elements of a Good Forecast

Accurate and in writing, Reliable, Meaningful (units), Cost-effective, Useful time horizon, Simple to understand & use

What are the The Realities of Forecasting

• Forecasts are seldom perfect; unpredictable outside factors may impact the forecast

• Product family and aggregated forecasts are more accurate than individual product forecasts

• the more specific you get for your forecast, the harder it will be

• Short term forecasts tend to be more accurate than long-term forecasts

• Forecast for dependent-demand items is more accurate than for independent-demand items

• Most techniques assume underlying stability in the system

What are the 2 groups of forecasting methods

Qualitative and Quantitative

What are the Qualitative Methods for Demand Forecasting?

Jury of executive opinion, Sales force composite, Market survey, Delphi method, Historical life cycle analogy

What are the Quantitative Methods for Demand Forecasting?

Time-Series Forecasting

What are the time series components?

Trend patterns, Seasonal component, Cyclical component, Irregular variations, Random variations

What are the Time Series Models?

• Naive methods - stables times series, seasonal variaions, data with trend

• Averaging methods - Simple moving average, weighted moving average, exponential smoothing

• Trend models - Linear and non-linear, trend-adjusted exponential smoothing

• Techniques for seasonality and cycles

What is and when should you use the naive method?

the next periods forecast is equal to last periods actual forcast

What is and when should you use the simple moving average?

Average of the most recent periods used for forecast, its best when no strong trend is present

What is and when should you use the weighted moving average?

Assigns more weight to recent values

What is and when should you use the weighted exponential smoothing?

Weighted average of all previous observations - each of them will their own weights, you will need data at least for the previous period - so you need to have an initial value

Whats the relation between the data and the alpha value?

if you want to give more weight to the most recent data, your alpha should be bigger, if your data is not showing any trend, then choose a smaller alpha

What are some ways to determine alpha?

a) average of several periods of actual data

b) subjective estimate (for this example, use 60)

c) first actual value (naïve approach)

What are the impact of smoothing constants (alpha value)

α = 0: Forecast never changes (stable), α = 1: Forecast = Last observed demand (responsive) - Choose low α for stable averages, high α for changing trends

What is and when should you use the Trend-Adjusted Exponential Smoothing (TAES)

A variation of exponential smoothing used when a trend is present.

What are the components of Trend-Adjusted Exponential Smoothing (TAES)

• A forecast based on **simple exponential smoothing

• A smoothed trend factor

What is and when should you use the Linear Trend Equation

• Fit a trend line to historical data using regression (Least Squares Line)

• with this, you can forecast for multiple periods, whereas you can only forecast for the next period using the previous equations

What are the models of seasonality

• Additive Model: Demand = Trend + Seasonality

• Multiplicative Model: Demand = Trend × Seasonality

What is the procedure for seasonal forecasting

1. Compute the seasonal relatives

2. De-seasonalize the demand data

3. Fit a model (e.g., moving average or trend)

4. Forecast using the de-seasonalized data

5. Re-seasonalize the forecasts

What are 2 ways to calculate seasonal relatives?

Simple Average Method (SAM) and the Centered Moving Average Method (CMAM)

How can you calculate the seasonal relative using the SAM (Simple Average Method)

• Step 1: Compute average demand for each season

• Step 2: Compute total average demand

• Step 3: Seasonal index = (Season Avg) / (Total Avg)

How can you calculate the seasonal relative using the CMAM (Centered Moving Average Method)

• Step 1: Calculate moving average equal to season length

• Step 2: Center the moving averages

• Step 3: Compute actual / centered average ratios

• Step 4: Average ratios by period

• Step 5: Adjust total of seasonal indices = number of seasons

What is the forecast error formula and what does the output mean?

Forecast Error = Actual − Forecast; (+ve = forecast too low (your actual is higher than the forecast)), −ve = too high)

What are the accuracy methods and how do they compare with each other?

MAD (Mean Absolute Deviation) - Simple; linear weight

MSE (Mean Squared Error ) - Penalizes large errors

MAPE Mean Absolute Percent Error (gives more information) - Scales to actual demand (we want this to be small as possible)

What is Cumulative Sum Error and what does the output mean?

• its the average of your error - Cumulative Sum Error (CSE) = ∑(A − F), Bias = CSE / n

• (+ve bias = underestimation (The actual is higher than the forecast), −ve = overestimation (The actual is lower than the forecast)

Explain the Tracking Signal

• Measures how well forecast matches actuals

• Formula: Tracking Signal = Cumulative Error / MAD

• Good signal has small absolute value

Explain Control Charts

• Visualize errors and check control

• Control limits: ±2σ or ±3σ, where σ = √(MSE)

Explain the interpretation of Control Charts

• If errors exceed control limits → Take corrective action

• All positive/negative errors → Bias in forecast

• Random pattern within limits → OK

What a main difference between a seasonality patterns and cycle patterns?

seasonality patterns are patterns that regularly repeats itself and is constant in length, the other one is called a cycle and its not necessary consistent