Summary of Corporate Financial Analysis: Methods and Forecasting
Corporate Financial Analysis: Methods and Forecasting Guide
Introduction
Forecasting is the process of using past information to estimate future values of a variable of interest — for example, revenue, demand, or cost. Accurate forecasts support planning, budgeting, and decision-making across business functions. This material focuses on two broad analytic approaches: Associative Models and Time Series Models. We break down concepts, show examples, and compare methods so you can apply the right model to a real problem.
Definition: Forecasting is the systematic use of historical data and relationships among variables to estimate future outcomes.
Big-picture distinction
- Associative Models treat the target variable as linked to other explanatory variables. Forecasts are built from those relationships.
- Time Series Models use only the historical values of the target variable itself, identifying patterns (trend, seasonality, cycles, noise) to project forward.
Associative Models
What they are
Associative models (also called causal or explanatory models) assume the variable to be forecasted is related to one or more other variables in the system. The goal is to estimate how changes in those explanatory variables affect the forecasted variable.
Definition: An associative model estimates a target variable by modeling its functional relationship with one or more explanatory variables.
Common forms
-
Simple linear regression: one explanatory variable. Example: sales $y$ related to advertising spend $x$ via $y = \beta_0 + \beta_1 x + \varepsilon$.
-
Multiple regression: several explanatory variables. Example: revenue $R$ as a function of price $p$, marketing $m$, and GDP $g$: $$R = \beta_0 + \beta_1 p + \beta_2 m + \beta_3 g + \varepsilon$$
-
Nonlinear models: relationships that are exponential, logistic, or power functions, e.g. $y = \alpha e^{\beta x}$.
Practical example
A retailer models weekly sales as a function of price, number of promotions, and local unemployment rate. After estimating coefficients using historical data, the retailer inputs projected prices and macro forecasts to produce future weekly sales.
Strengths and limitations
| Strengths | Limitations |
|---|---|
| Can incorporate causal drivers and scenario analysis | Requires good explanatory variables and quality data |
| Useful for policy or price-impact questions | Misspecified relationships produce biased forecasts |
| Enables sensitivity analysis ("what-if") | More complex estimation and data requirements |
When to use associative models
- You have strong theory about drivers of the target variable.
- You need to test scenarios or estimate causal impacts.
- You can access reliable explanatory variable forecasts.
Time Series Models
What they are
Time series models use only the historical sequence of the variable being forecasted to identify recurring patterns and structure. Common components are trend, seasonality, cycle, and residual (noise).
Definition: A time series model forecasts future values by modeling patterns within the historical sequence of the target variable itself.
Common forms
-
Naive methods: the simplest forecasting methods using recent values (e.g., current period = next period).
-
Moving average and exponential smoothing: smooth past values to remove noise and reveal underlying patterns. Example of simple exponential smoothing: $$\hat{y}_{t+1} = \alpha y_t + (1-\alpha)\hat{y}_t$$ where $0<\alpha<1$ is the smoothing parameter.
-
Seasonal decomposition: separate series into trend, seasonal, and residual components (additive or multiplicative decomposition).
-
ARIMA (AutoRegressive Integrated Moving Average): models autocorrelation and differencing to handle non-stationarity. A general ARIMA$(p,d,q)$ can be written conceptually as polynomial operators on lagged values and errors.
Practical example
An airline forecasts monthly passenger numbers. The historical series shows an upward tren
Sign up for the full summaryAlready have an account? Sign in
Forecasting Models Overview
Klíčové pojmy: Forecasting uses history to estimate future values, Associative models use explanatory variables to forecast the target, Time series models rely only on the target's historical values, Simple linear regression is an example of an associative model: $y = \beta_0 + \beta_1 x + \varepsilon$, Exponential smoothing forecast: $\hat{y}_{t+1} = \alpha y_t + (1-\alpha)\hat{y}_t$, Use associative models when causal drivers and their forecasts are available, Use time series models when historical patterns (trend/seasonality) dominate, Validate forecasts with holdout samples and metrics like MAE or RMSE, Combine models (hybrids) when complementary strengths exist, Document assumptions about explanatory variable forecasts and re-evaluate after structural changes