Forecasting Methods, Correlation Analysis, and Regression Models

Overview of Forecasting Methods

Role and Purpose in Business:
- Forecasts serve as the operational foundation for organizational business plans and master budgets.
- Forecasts project future performance metrics, including product demand, required inventory levels, and expected cash flows.
Qualitative Forecasting Methods:
- Rely on management's qualitative judgment, accumulated experience, and professional intuition rather than purely numerical data.
Quantitative Forecasting Methods:
- Rely on mathematical models, statistical calculations, and analytical graphs.
- Causal Relationship Forecasting: An analytical technique utilized when an environmental or operational factor is plotted along the $x$ -axis.
- Time-Series Analysis: An analytical forecasting technique utilized when chronological time periods are plotted along the $x$ -axis.

Principles of Correlation Analysis

Foundational Role:
- Correlation analysis represents the quantitative foundation underlying predictive forecasting methods.
Definition:
- Correlation measures the strength and direction of the linear relationship existing between two variables.
- Mathematically expressed as the coefficient of correlation ( $r$ ).
Graphical Representation (Scatter Diagrams):
- Constructed by plotting paired numerical observations of two variables on a Cartesian plane.
- The degree to which the plotted scatter pattern approximates a single straight line reflects the absolute value of the coefficient of correlation ( $|r|$ ).
Range and Interpretation of the Coefficient of Correlation ( $r$ ):
- The numerical value of $r$ strictly ranges from $1$ to $-1$ .
- Perfect Direct Relationship ( $r = 1$ ): All plotted data points form an exact straight line with an upward positive slope (Figure 12-1).
- Perfect Inverse Relationship ( $r = -1$ ): All plotted data points form an exact straight line with a downward negative slope (Figure 12-2).
- Strong Direct Relationship ( $r = 0.7$ ): Plotted data points form an upwardly dispersing cluster that follows a positive linear trendline.
- No Linear Relationship ( $r = 0$ ):
  - Data points may be scattered randomly without any observable pattern across the coordinate plane (Figure 12-3).
  - Alternatively, data points may follow a distinct, non-linear curvilinear pattern, such as a U-shaped or wave curve (Figure 12-4).
  - Critical Caveat for Zero Correlation: A correlation coefficient of zero ( $r = 0$ ) does not indicate an absence of any relationship; it establishes solely that any relationship present between the variables cannot be expressed in the form of a linear equation.

Scatter plots illustrating correlation coefficients: perfect direct (r = 1), perfect inverse (r = -1), strong direct (r = 0.7), and zero linear correlation (r = 0)

The Coefficient of Determination

Definition and Notation:
- Denoted as $r^2$ , the coefficient of determination is calculated by squaring the coefficient of correlation ( $r$ ).
- Evaluates the goodness of fit between the paired independent and dependent variables.
Mathematical Interpretation:
- Measures the precise proportion of total variation observed in the dependent variable that is explained by or accounted for by variation in the independent variable.
Application (Example 12-1: Disposable Income and Vehicle Sales):
- Scenario: An automotive dealership establishes that new vehicle sales depend upon disposable income, demonstrating a coefficient of correlation of $0.8$ .
- Calculation:
  - $r = 0.8$
  - $r^2 = (0.8)^2 = 0.64$
- Conclusion: Stating that $r = 0.8$ is mathematically equivalent to stating that $64\%$ of the variation in new car sales from the average can be explained by changes in disposable income.

Simple Linear Regression Analysis

Definition and Methodology:
- Also called least-squares analysis.
- The mathematical technique used to derive the linear equation describing the relationship between two variables possessing a nonzero coefficient of correlation.
- Simple Regression: Used when the model relies on exactly one independent variable.
Algebraic Structure of Simple Linear Regression:
- Formula:
  - $y = a + bx$
- Variable definitions:
  - $y$ = the dependent variable
  - $a$ = the $y$ -intercept
  - $b$ = the slope of the regression line (also termed the regression coefficient)
  - $x$ = the independent variable
Derivation via the Least-Squares Method:
- The optimal linear line is mathematically determined through calculus.
- Calculus is utilized to minimize the sum of the squared vertical distances between each empirical data point and the linear line (giving rise to the term "least-squares").
Application and Step-by-Step Projection (Example 12-2: Advertising and Sales):
- Empirical Observations Collected:
  - Observation 1: Advertising ( $x$ in thousands) = $71$ , Annual Sales ( $y$ in millions) = $26.3$
  - Observation 2: Advertising ( $x$ in thousands) = $31$ , Annual Sales ( $y$ in millions) = $13.9$
  - Observation 3: Advertising ( $x$ in thousands) = $50$ , Annual Sales ( $y$ in millions) = $19.8$
  - Observation 4: Advertising ( $x$ in thousands) = $60$ , Annual Sales ( $y$ in millions) = $22.9$
  - Observation 5: Advertising ( $x$ in thousands) = $35$ , Annual Sales ( $y$ in millions) = $15.1$
- Solved Regression Model:
  - Application of the least-squares method identifies expected sales as equal to $\$4.2\text{ million}$ plus $311.741$ multiplied by the advertising expenditure:
  - $y = \$4{,}200{,}000 + 311.741x$
- Projecting Advertising Required for $\$32{,}000{,}000$ in Sales:
  - $y = \$4{,}200{,}000 + 311.741x$
  - $\$32{,}000{,}000 = \$4{,}200{,}000 + 311.741x$
  - $311.741x = \$27{,}800{,}000$
  - $x = \$89{,}177$
  - The organization must budget and spend $\$89{,}177$ on advertising to attain the target revenue of $\$32{,}000{,}000$ .

Scatter plot and fitted regression line of annual sales versus advertising expenditures

Multiple Regression and Logistic Regression

Multiple Regression Analysis:
- Employed whenever the predictive model incorporates more than one independent variable.
- Practical Necessity: Modeling annual corporate sales strictly as a function of advertising expenditures (as in Example 12-2) is unrealistic because organizational sales depend upon numerous concurrent internal and market variables.
- Functionality: Allows an organization to isolate multiple independent factors and assign an individualized mathematical weight to each factor corresponding to its distinct influence on the dependent variable.
- Formula Structure:
  - $y = a + b_1x_1 + b_2x_2 + b_3x_3 + b_4x_4 + \dots$
Logistic Regression Analysis:
- A specialized statistical modeling technique applied in classification problems and predictive analytics (covered further in Study Unit 20, Subunit 2).

Assumptions of the Linear Regression Model

Relevant Range Constraints:
- The linear association formulated between $x$ and $y$ is strictly valid solely within the designated relevant range.
- Relevant Range Definition: The bounded range of operational volume or activity within which unit variable costs remain constant and total fixed costs do not change.
- Model practitioners must determine this relevant range and ensure they do not forecast or extrapolate values beyond these boundaries.
Negative $y$ -Intercept Warning:
- The occurrence of a negative $y$ -intercept ( $a < 0$ ) in a simple regression formula typically serves as an explicit warning that the calculated intercept lies outside the relevant range.
Distributional Assumptions:
- For any given specific value of the independent variable $x$ , the values of the dependent variable $y$ dispersed around the regression line must follow a normal distribution.
- The mean of this error distribution around the regression line must equal zero.
Statistical Independence:
- The observed values of $y$ must be statistically independent of one another.

Strategic Applications and Limitations of Regression Analysis

Managerial Accounting and Budgeting Benefits:
- Regression analysis is a core quantitative tool for operational budgeting, financial planning, and managerial cost accounting.
- Essential for separating mixed costs into their constituent components when establishing flexible budgets:
  - The $y$ -intercept ( $a$ ) represents the fixed cost component.
  - The slope coefficient ( $b$ ) represents the variable cost rate per unit of activity.
Assumption of Temporal Continuity:
- Regression analysis presumes that historical cost and volume relationships can be validly projected into subsequent periods.
- Operational Limitation: Regression models can only be utilized as long as organizational cost structures and environmental conditions remain unchanged from the historical baseline.
Correlation Versus Causality:
- Regression analysis models mathematical co-movement; it does not demonstrate or verify causality.
- Even when $x$ and $y$ display strong statistical co-movement, the apparent connection may be generated by an unmodeled external third variable.
- Illustrative Example: High car wash sales volume and sunny weather are strongly correlated, but car wash sales volume does not cause sunny weather.
Review Reference:
- Subunit 12.1 review material corresponds to multiple-choice questions 1 through 5 beginning on page 404.

Overview of Forecasting Methods

Why Businesses Forecast (The "Crystal Ball"):
- Think of forecasting as planning for a huge party: you wouldn't buy snacks without estimating how many guests are showing up.
- Companies use forecasts as the backbone of their master budgets to estimate product demand, stock the right amount of inventory, and ensure they have enough cash to keep the lights on.
Qualitative Forecasting (Going with Experience and Gut):
- Relies on human intuition, executive experience, and educated judgment rather than crunching cold numbers.
- Example: A veteran coffee shop manager predicts pumpkin spice latte demand will surge early this year simply because social media vibes and autumn weather arrived early.
Quantitative Forecasting (Letting the Math Speak):
- Relies on statistical formulas, hard historical data, and plotted graphs.
- Causal Relationship Forecasting: Exploring cause-and-effect by plotting an environmental or operational trigger along the $x$ -axis (e.g., graphing local temperatures on the $x$ -axis to predict ice cream sales on the $y$ -axis).
- Time-Series Analysis: Tracking patterns purely across chronological time along the $x$ -axis (e.g., mapping sneaker sales month-by-month from January to December).

Principles of Correlation Analysis

What is Correlation?
- It measures how tightly two variables dance together and in what direction.
- Represented mathematically by the coefficient of correlation ( $r$ ), which strictly spans between $-1$ and $1$ .
Visualizing with Scatter Plots:
- You plot paired data points on a coordinate grid. The closer the cluster of dots hugs a single straight path, the stronger the relationship ( $|r|$ ).
- Perfect Direct Relationship ( $r = 1$ ):
- Plotted points line up in a flawless, upward-pointing line.
- Example: Every extra hour a tutor works directly yields an exact, predictable jump in their paycheck.
- Perfect Inverse Relationship ( $r = -1$ ):
- Plotted points line up in a flawless downward line.
- Example: For every degree the outside temperature climbs, sales of wool winter coats drop in a straight line.
- Strong Direct Relationship ( $r = 0.7$ ):
- Dots form an unmistakable upward cloud. More study hours generally yield higher test scores, though minor personal variations exist.
- No Linear Relationship ( $r = 0$ ):
- The points could look like spilled salt across the page with no direction.
- The Big Trap: $r = 0$ does not mean there is no relationship at all! It just means a straight line cannot describe it. The data could form an arch or a U-curve (e.g., your test performance increases with caffeine up to a sweet spot, but crashes if you drink too much).
Coefficient of Determination ( $r^2$ ):
- Found by simply squaring the correlation coefficient ( $r$ ).
- Tells you what percentage of the changes in the dependent variable can actually be explained by the independent variable.
- Example: If car sales and customer disposable income have an $r = 0.8$ , squaring it gives $r^2 = 0.64$ . That means $64\%$ of the swings in car sales come directly from changes in income, while the remaining $36\%$ come from other factors like gas prices or styling.

Simple Linear Regression Analysis

Connecting the Dots (The Least-Squares Method):
- If correlation shows two variables are linked, regression gives you the actual recipe (the equation) to make predictions.
- Think of stretching a rubber band through a cloud of data points. Calculus finds the one unique line that minimizes the sum of all the squared vertical distances between the real points and the line.
The Line Formula:
- $y = a + bx$
- $y$ = The target outcome you want to predict (dependent variable, like total sales).
- $a$ = The starting line or baseline when $x$ is zero ( $y$ -intercept).
- $b$ = The slope/speed boost: how much $y$ increases for every single unit increase in $x$ .
- $x$ = The lever you pull (independent variable, like money spent on marketing).
Real-World Walkthrough (Ad Spend vs. Annual Sales):
- Suppose a retail store tracks ad spending ( $x$ in thousands) and annual sales ( $y$ in millions) over five periods:
- Observation 1: Advertising = $71$ , Sales = $26.3$
- Observation 2: Advertising = $31$ , Sales = $13.9$
- Observation 3: Advertising = $50$ , Sales = $19.8$
- Observation 4: Advertising = $60$ , Sales = $22.9$
- Observation 5: Advertising = $35$ , Sales = $15.1$
- The Fitted Model:
- Running the least-squares formula yields the equation:
- $y = \$4{,}200{,}000 + 311.741x$
- Translation: Even with zero advertising, baseline brand awareness brings in $\$4{,}200{,}000$ . For every extra $\$1{,}000$ put into advertising ( $x = 1$ ), sales increase by approximately $\$311{,}741$ .
- Setting a Goal ( $\$32{,}000{,}000$ in Sales):
- Plug sales into the equation to find the required ad budget:
- $\$32{,}000{,}000 = \$4{,}200{,}000 + 311.741x$
- $311.741x = \$27{,}800{,}000$
- $x = \$89{,}177$
- The team needs to budget $\$89{,}177$ in advertising to hit their revenue target.
Beyond One Variable (Multiple Regression):
- In reality, sales do not rely solely on advertising. Competitor prices, foot traffic, and season all play a role.
- Multiple regression handles several inputs at once: $y = a + b1 x1 + b2 x2 + b3 x3 + \dots$

Rules, Assumptions, and Classic Traps

Stay Inside the Relevant Range:
- Your formula is only valid within the operational boundaries where cost structures do not change.
- Example: If your bakery regression is based on baking between $100$ and $1{,}000$ cupcakes a day, you cannot use it to forecast baking $50{,}000$ cupcakes. At that level, you'd need a whole new factory, blowing up your fixed costs.
The Negative Intercept Red Flag ( $a < 0$ ):
- If your math suggests baseline sales are negative when activity is zero, it is a warning sign that you're attempting to extrapolate outside your relevant range.
Distribution and Independence:
- For any given value of $x$ , the random errors in $y$ should balance out to an average of $0$ and follow a bell-shaped normal curve.
- Observations must be independent of one another rather than cascading domino effects.
Correlation Does NOT Prove Causation:
- Just because two variables climb together does not mean one caused the other.
- Example: Car wash revenues and sunny weekend weather have a sky-high correlation. However, washing your car does not magically cause the sun to shine; good weather encourages people to get outside and clean their cars.
Budgeting Superpower:
- Regression is essential for splitting mixed utility or factory costs into fixed elements ( $a$ ) and variable per-unit elements ( $b$ ) when preparing flexible budgets.