Forecasting: Predicting Future Trends

• Forecasting involves predicting future occasions through analyzing historical data, employing models for informed estimations regarding demand, trends, and behaviors and is important for planning and effective making decisions in manufacturing and service applications.
• Predicts future events using quantitative and qualitative methods.
• Quantitative uses mathematical models and qualitative uses subjective predictions based on expert opinions and intuition.
• Usually a combination of both the approaches is used to enhance accuracy

Forecasting Time Horizons

• Short-range forecasts are generally less than 3 months and up to 1 year and include purchasing, job scheduling, workforce levels, job assignments, and production levels.
• Medium-range forecasts range from 3 months to 3 years, covering sales and production planning, budgeting, cash budgeting, and operating plan analysis.
• Long-range forecasts are over 3 years and are inclusive of new product planning, facility location and expansion, expenditures, research and development.
• Medium and long range deal with planning, products, plants, and processes.
• Short-term forecasts employs different methods than longer-term forecasting.

Types of Forecasts

• Economic forecasts addresses inflation rate, money supply, and housing starts.
• Technological forecasts predict the rate of progress and its affects on new product development.
• Demand forecasts predict sales of current products and services.

Strategic Importance

• Supply Chain Management: Forecasting is important for supplier relations, product innovation, cost management, and market speed.
• Human Resources: Involves hiring, training, and laying off workers.
• Capacity: Shortages might lead to unreliable delivery, customer loss, and market share reduction.

Seven Steps in Forecasting

• Determine the use of the forecast.
• Select the items to be forecasted.
• Determine the forecast's time horizon.
• Select the forecasting model(s).
• Gather the data needed to make the forecast.
• Make the forecast.
• Validate and implement the results.

Forecasting Realities

• Forecasting assumes system stability.
• Computerized software helps the automation of predictions by firms.
• Aggregate forecasts are more accurate than individual ones, which helps balance the over and underpredictions.
• External and unmanageable events can disrupt the forecasting ecosystem, like the COVID-19 pandemic.
• Stable trends help forecasting models when leveraging on historical data, this is useful for time series, exponential smoothing, and regression analysis.

Aggregated Forecasts

• Random variations offset when forecasting product families.
• Total laptop demand is more accurate than predicting a specific model.
• Companies use this strategy for production, materials, and inventory.
• Retailers minimize risk by forecasting at the category level.
• It provides more demand prediction than individual product forecasts.

Impact of external factors

• Forecasting models are unable to predict or control Global pandemics, natural disasters, economic recessions and sudden market changes.
• Retail sector saw an online increase in shopping due to COVID-19 which resulted in traditional in-store forecasts to fail.
• Ventilators and PPE saw a surge, where factory shutdowns in China caused shortages in electronics, cars, and pharmaceuticals.
• Businesses adopted short-term forecasting and scenario planning, more focus on analyzing data in real time and machine learning to make adjustments dynamically.
• External shocks can disrupt forecasting; flexibility and new data are essential.

Forecasting Approaches

• Forecasting methods are broadly classified into Qualitative and Quantitative that suits different business scenarios.
• Qualitative methods are useful when situation is vague and little data exist.
• New products and technology involves emotions, personal experiences, and value system which is the concept behind Qualitative Methods.
• Quantitative methods are useful when the situation is ‘stable’ and historical data exist.
• Forecasting sales using existing products and technology involves mathematical techniques, which is the concept behind Quantitative Methods.
• Businesses combine both for reliable forecasting.

Qualitative Forecasting Methods

• Qualitative forecasting uses expert judgment, intuition, emotions, and experience rather than numerical data.
• It is useful when historical data is unavailable or unreliable for new products and technologies.
• It is applicable in uncertain situations, and provide flexibility for forecasting.
• Delphi Method: Experts provide independent forecasts and iterate until consensus is reached.
• Market Research: Surveys, focus groups, and customer feedback predict demand.
• Panel Consensus: Collective predictions from Managers and Professionals.
• Historical Analogy: Estimate demand by relating it with a similar product.
• The methods might be subjective, but they are valuable with little or no historical data.

Overview of Qualitative Methods

• Jury of executive opinion pools opinions of high-level experts or managers, sometimes augmented by statistical models.
• Delphi method uses a panel of experts queried iteratively.
• Sales force composite aggregates estimates from salespersons and ensures they are reasonable.
• Market surveys ask customers about future purchasing plans.
• Involves a small group of experts and managers working together.

Jury of Executive Opinion

• A small group of high-level experts and managers is used.
• Group estimates demand by working together.
• Combines experience with statistical models.
• Groupthink can be a disadvantage for the group, despite being quick.

Delphi Method

• Is a group process that runs until consensus is reached.
• Has Decision Makers, Staff, and Respondents.

Sales force composite

• Individual salespeople project their sales, combined regionally and nationally
• Reps know customers - May be overly optimistic.

Market Survey

• Customers are asked about purchasing plans
• Good for demand forecasting, product design, planning
• What consumers say and do may differ and may be overly optimistic.

Quantitative Forecasting Methods

• Forecasting uses math and stats with historical data and assumes patterns continue.
• It is best to used when trends are consistent, for products with historical sales data, and to analyze trends.
• Moving Averages: Uses past sales data to smooth fluctuations and identify trends.
• Exponential Smoothing: Greater importance is assigned to recent data for trend analysis.
• Trend Projection: Uses past growth rates to estimate future demand.
• Regression identifies relationships (e.g., advertising spend vs. sales).
• Economic indicators are used in Econometric Models are used to predict demand.
• These techniques are objective, but don't adapt well to sudden market changes.

Overview of Quantitative Approaches

• These include Naive, Moving averages, Exponential smoothing, Trend projection, Linear regression, Time-series models and Associative models

Time-Series Forecasting

• It is a quantitative method used to predict future values with the assumption that past patterns are to continue in the future.
• It's commonly used to apply to business, economics, and operations manageent.
• It uses evenly spaced numerical data over consistent time intervals and includes weather temperatures, economic indicators, stock prices, sales data.
• A response variable is observed over time where the dependent variable is measure over a fixed period without any external factors.
• Purely focuses on prior observation to detect patterns without any external variables.
• Assumes past influences and patterns continue in future.

Common Time-Series Forecasting Methods

• Naïve Approach: Assumes stability and equals to last period's value.
• Simple and effective for short-term forecasts in stable conditions
• Moving Averages: Averages past observations to smooth fluctuations.
• It helps identify trends by reducing noise
• Exponential Smoothing: Similar to moving averages, it assigns more weight to recent data and is more responsive to new trends and changes.
• Trend Projection: Model's trend on past data using linear regression.
• It's useful for long-term when data has a consistent trajectory.
• Seasonal Adjustments: Accounts for repeating seasonal patterns in recurring spikes and drops.

Advantages of Time-Series Forecasting

• The method Is data-driven, objective and uses historical data without judgment, ideal for stable environments and patterns in inventory, production, and marketing.
• Limitations of are that it ignores external factors and assumes patterns continue, less effective for new products, may struggle with new products as lacks historical data.

Time-Series Components

• Consists of Trend, cyclical, seasonal, and random factors.
• Trend: Long-term data movement like increase/decrease.
• Cyclical: Long-term patterns affected by economic conditions (recessions, expansions).
• Seasonal variations in short period.
• Random factors result in unpredictable fluctuations due to events like natural disasters or economic shocks.

Components of Demand

• Actual Demand: A live line that displays real demand, fluctuating due to various factors
• Small and Unforedeictable changes in demand caused by external factors
• Overall line that shows the direction of the demand of a product
• Recurring spikes of demand during specific times
• Display of demand over a four year schedule

Trend Component

• Persistent, overall pattern.
• Typically happens for several years and happens due to population, cultural changes, age, or the advent of new technology.

Seasonal Component

• A pattern that happens due to weather, or due to the custom of different regions.
• PERIOD LENGTH “SEASON” LENGTH NUMBER OF “SEASONS” IN PATTERN
• Week Day 7
• Month Week Four to four and a half
• Month Day 28 - 31
• Year Quarter 4
• Year Month 12
• Year Week 52

Cyclical Component

• Can be affected by business, economic, and the political climate.
• Movements of up and down with multiple years of planning.

Random Component

• Chance and unusual situations
• Follows no trend
• Can't be predicted

Naive Approach

• It assumes that the next period mirrors the one before.
• Good because efficient and cost effective.
• Can be a good starting point to compare more sophisticated models.

Moving Averages

• Series of math means
• Used when markets are steady
• Smoothes data
• Moving average =  demand in previous n periods /n

Weighted Moving Average

• Older data is less important, and is based on experience which affect its weights.
• The moving average gets weighted average =  ( ( Weight for period n )(Demand in period n ) )/  Weights

Potential Problems With Moving Average

• Increasing n smooths forecasts but makes it less sensitive
• Doesn't predict trends well
• Needs large datasets, too much dependence on historical data

Exponential Smoothing

• A type of weighted moving average.
• It exponentially declines weights, and prioritizes recent data
• Used to adjust smoothing constant
• a:0-1
• Has little record keeping of past data

Smoothing Constants - α

• Values lies between 0.05 0.50
• As α increases, older values becomes less important
• High a(0.7-0.9) yields more responsiveness to changes and useful where there is significant fluctuations.
• Low a (0.1-0.3) ensures smoother projection, less responsiveness to changes, where better for stable predictions and long-term planning.

Measuring Forecast Error

• Uses accuracy by obtaining the lowest error forecast with three means.

1. MAD - Mean Absolute Deviation
2. MSE - Mean Squared
3. MAPE - Mean Absolute Percent Error

• Minimize forecast errors by testing different values of a:

Mean Absolute Deviation (MAD)

• Computes the average absolute dissimilarity between predicted vs actual data

Mean Squared Error (MSE)

• It gives more weight to large deviations.

Mean Absolute Deviation (MAPE):

• Computes error as a percentage of values.

Comparison of Measures

• MAD is the average of forecasting error where MSE emphasizes large errors.
• MAPE measures its scales to show how it varies.

Exponential Smoothing with Trend Adjustment

Exponential Smoothing with Trend Adjustment (Holt's Method)

• It incorporates both level and trend which makes it efficient for trending data
• Better predicts and focuses on growth over time instead of focusing on the demand being stable thus making the product inaccurate.

Key Benefits

• Reacts better to trends and produces long-term trend predictions.
• Great for cases when you have to steady to a specific amount and is used is new tech.

Level Component

• Baseline

Trend Component

• Adjustment factor.