Regression Models in Demand Estimation

1. Introduction

Demand estimation is the process of determining how much of a good or service consumers are willing and able to purchase at different prices and under different economic conditions. Regression models are among the most widely used quantitative techniques for estimating demand because they allow economists, businesses, regulators, and policymakers to measure the relationship between quantity demanded and factors such as price, income, population, substitute prices, consumer preferences, and time.

In energy law and electricity regulation, regression-based demand estimation is particularly important. Electricity demand depends not only on tariff levels but also on industrial activity, weather, household income, population, seasonal patterns, technological changes, and regulatory interventions. Accurate demand forecasts can therefore influence tariff determination, generation planning, transmission investment, power procurement, and consumer protection.

2. Meaning of Regression Models

A regression model statistically estimates the relationship between a dependent variable and one or more independent variables.

For demand estimation, a simple model can be expressed as:

\[ Q_d = \beta_0+\beta_1P+\beta_2Y+\beta_3P_s+\beta_4Z+\epsilon \]

Where:

• \(Q_d\) = quantity demanded;
• \(P\) = price of the product;
• \(Y\) = consumer income;
• \(P_s\) = price of a substitute;
• \(Z\) = other relevant demand factors;
• \(\beta_0\) = intercept;
• \(\beta_1,\beta_2,\beta_3,\beta_4\) = estimated coefficients; and
• \(\epsilon\) = error term.

Normally, economic theory predicts that the coefficient of own price will be negative:

\[ \beta_1<0 \]

because an increase in price generally reduces quantity demanded.

\(\hat{y} = b_0 + b_1x\)

\(\hat{y} = 8.49 - 0.54x\)

R² = 0.72 · b₀ = intercept · b₁ = slope · least squares minimizes squared vertical residual gaps.

Focus

DataLinePredictionsResiduals

Generate new data

Drag a point to refit the line.

246810246810

Give feedback

3. Simple Linear Regression

The simplest demand model contains only price as the explanatory variable:

\[ Q_d=a+bP \]

For example, suppose electricity consumption is estimated as:

\[ Q_d=10,000-500P \]

An increase in the electricity tariff would, according to the model, reduce demand.

The major advantage of this model is its simplicity. However, it can be misleading if other factors affecting demand are omitted.

For example, electricity consumption may increase even when tariffs increase because industrial production, population, or temperature also increases.

4. Multiple Regression Models

A multiple regression model includes several explanatory variables:

\[ Q_d=\beta_0+\beta_1P+\beta_2Y+\beta_3P_s+\beta_4Pop+\beta_5T+\epsilon \]

where:

• \(P\) = electricity price;
• \(Y\) = income;
• \(P_s\) = substitute energy price;
• \(Pop\) = population;
• \(T\) = technological or seasonal factor.

Multiple regression is usually more appropriate for energy demand because electricity consumption is determined by several interacting variables.

For instance, household electricity demand may depend on:

1. electricity tariffs;
2. household income;
3. number of household members;
4. appliance ownership;
5. temperature;
6. availability of alternative fuels; and
7. government subsidies.

5. Log-Linear Demand Models

A particularly useful model is the logarithmic regression:

\[ \ln Q=\beta_0+\beta_1\ln P+\beta_2\ln Y+\epsilon \]

The advantage is that the coefficients can often be interpreted as elasticities.

For example, if:

\[ \beta_1=-0.4 \]

then a 1% increase in price is associated, other things being equal, with approximately a 0.4% reduction in demand.

This is highly relevant to tariff regulation because regulators frequently need to understand how consumers may respond to tariff changes.

6. Price Elasticity of Demand

Regression analysis can be used to estimate:

\[ E_p=\frac{\%\Delta Q}{\%\Delta P} \]

The price elasticity indicates how sensitive demand is to price.

Elastic demand

If:

\[ |E_p|>1 \]

demand is elastic.

Inelastic demand

If:

\[ |E_p|<1 \]

demand is inelastic.

Electricity demand, particularly for essential household consumption, may be relatively inelastic in the short run. Consumers cannot immediately eliminate essential electricity consumption simply because tariffs rise.

However, demand can become more elastic over the longer term as consumers purchase efficient appliances, install solar generation, alter production processes, or switch fuels.

7. Regression Variables in Energy Demand Estimation

A. Electricity price

Price is normally the most important explanatory variable.

A regulator may estimate the relationship between:

• tariff changes; and
• electricity consumption.

The estimated coefficient helps determine expected consumption following a tariff revision.

B. Income

Higher household or industrial income can increase electricity consumption because consumers can purchase more appliances and businesses can expand production.

C. Weather

Temperature is extremely important for electricity demand.

Hot weather can increase:

• air-conditioning;
• refrigeration; and
• cooling demand.

Cold weather may increase:

• electric heating; and
• water heating.

D. Population

Population growth generally increases aggregate electricity demand.

E. Industrial output

Industrial electricity demand can be strongly associated with:

• manufacturing production;
• mining activity;
• economic growth; and
• capacity utilisation.

F. Substitute energy prices

Demand for electricity may be influenced by prices of:

• natural gas;
• coal;
• petroleum;
• LPG; and
• other energy sources.

8. Time-Series Regression

Energy demand data are frequently collected over time—for example:

• daily;
• monthly;
• quarterly; or
• annually.

A time-series demand model may therefore take the form:

\[ Q_t=\beta_0+\beta_1P_t+\beta_2Y_t+\beta_3Temp_t+\beta_4Q_{t-1}+\epsilon_t \]

Here \(Q_{t-1}\) represents previous-period demand.

This is important because electricity consumption frequently exhibits inertia: today's consumption can be related to yesterday's or last month's consumption.

Time-series models must, however, address problems such as:

• autocorrelation;
• seasonality;
• non-stationarity;
• structural breaks; and
• changing consumer behaviour.

9. Panel Regression

Where data are available for several regions, states, utilities, industries, or consumer groups, researchers can use panel regression.

For example:

\[ Q_{it}=\alpha_i+\beta P_{it}+\gamma Y_{it}+\epsilon_{it} \]

where \(i\) represents a particular state, utility, or consumer group and \(t\) represents time.

Panel models are useful because they allow researchers to study both:

• differences between regions; and
• changes within each region over time.

This can be especially valuable in electricity regulation where different distribution companies have different consumer characteristics.

10. Regression and Tariff Regulation

Regression models can assist regulators in determining whether proposed tariff increases are likely to cause significant changes in consumption.

A tariff-setting authority may consider:

1. historical electricity consumption;
2. historical tariff levels;
3. income growth;
4. industrial production;
5. weather conditions;
6. consumer categories; and
7. expected future economic activity.

The regression model can then generate a demand forecast.

Such forecasts can influence the regulator's assessment of:

• revenue requirements;
• power procurement;
• generation requirements;
• transmission capacity;
• distribution-system investment; and
• tariff structures.

11. Regression and Electricity Demand Forecasting

Demand estimation and demand forecasting are related but distinct.

Demand estimation asks:

What is the quantitative relationship between demand and its determinants?

Demand forecasting asks:

What will future demand be?

Regression estimation provides coefficients that can subsequently be used for forecasting.

For example:

\[ Q_{future}=\hat{\beta}_0+\hat{\beta}_1P_{future} +\hat{\beta}_2Y_{future}+\hat{\beta}_3Temp_{future} \]

The reliability of the forecast therefore depends upon the quality of both the estimated model and the assumptions concerning future variables.

12. Problems in Regression-Based Demand Estimation

12.1 Omitted-variable bias

If an important determinant of demand is excluded, the estimated price coefficient may be biased.

For example, estimating electricity demand only from tariff data while ignoring temperature can produce misleading results.

12.2 Multicollinearity

Independent variables may be strongly correlated.

For example:

• income;
• GDP; and
• industrial production

may move together.

This makes it difficult to isolate the independent effect of each variable.

12.3 Autocorrelation

In time-series data, errors from one period may be correlated with errors from another period.

12.4 Heteroskedasticity

The variance of the error term may differ across observations.

12.5 Endogeneity

Price and demand may influence one another.

For example, a utility may increase tariffs partly because it anticipates particular consumption or revenue conditions. Simple regression may then incorrectly interpret correlation as causation.

12.6 Structural breaks

Energy markets can experience sudden changes caused by:

• regulatory reforms;
• economic crises;
• pandemics;
• fuel-price shocks;
• technological changes; or
• major changes in tariff policy.

A historical regression relationship may therefore become unreliable.

13. Causation Versus Correlation

One of the most important legal and economic issues is the distinction between correlation and causation.

Suppose regression results demonstrate that tariff increases are associated with lower electricity consumption. That does not automatically establish that the tariff increase caused the entire reduction.

Other factors may have contributed.

Consequently, regulators and courts should examine:

• the model specification;
• data quality;
• assumptions;
• statistical significance;
• alternative explanations;
• sensitivity analysis; and
• expert methodology.

14. Role in Regulatory Decision-Making

Regression models should normally be treated as decision-support tools rather than substitutes for legal judgment.

A regulator must remain within the authority granted by legislation. Statistical evidence can help answer factual and economic questions, but it cannot itself create regulatory power.

For example, a demand model may demonstrate that a proposed tariff could reduce consumption. It cannot by itself establish that the regulator has statutory authority to impose that tariff.

15. Indian Legal and Regulatory Context

In India, demand estimation is particularly relevant to electricity regulation under the Electricity Act, 2003.

The Act establishes a regulatory framework involving institutions such as:

• the Central Electricity Regulatory Commission (CERC);
• State Electricity Regulatory Commissions (SERCs);
• transmission and distribution licensees; and
• system operators.

Demand forecasts can form part of the broader economic and technical material considered when determining tariffs and planning electricity supply.

The legal framework also emphasizes principles such as:

• consumer interests;
• recovery of reasonable costs;
• efficiency;
• transparency;
• reliable electricity supply; and
• development of electricity markets.

16. Important Case Laws

1. West Bengal Electricity Regulatory Commission v. CESC Ltd. (2002)

The Supreme Court dealt extensively with electricity tariff regulation and the role of regulatory authorities in determining reasonable tariffs.

The case demonstrates that electricity tariff determination involves technical, economic, and regulatory considerations. Quantitative evidence concerning costs, consumption and revenue can therefore be relevant, but the regulator must exercise its statutory authority according to the governing legal framework.

Relevance to regression demand estimation:
Demand estimates can provide economic evidence supporting tariff and revenue determinations, but statistical models must operate within the statutory regulatory framework.

2. PTC India Ltd. v. Central Electricity Regulatory Commission (2010)

This is a major Supreme Court decision concerning electricity regulation and the powers of the CERC under the Electricity Act, 2003.

The Court examined the statutory architecture of electricity regulation and the distinction between regulatory powers and subordinate legislation.

Relevance:
Economic modelling, including demand forecasting, cannot independently expand the statutory authority of a regulator. The model is evidence used within the legal framework.

3. BSES Rajdhani Power Ltd. v. Delhi Electricity Regulatory Commission

The litigation concerning tariff determination by electricity regulators illustrates the importance of regulatory scrutiny of expenditure, revenue requirements, projections and performance of distribution licensees.

Relevance:
Demand projections are important components of the financial and operational assumptions used in tariff proceedings. A regulator may examine whether projected demand and corresponding revenue assumptions are reasonable and supported by evidence.

4. Reliance Energy Ltd. v. Maharashtra State Road Development Corporation Ltd. (2007)

The Supreme Court emphasized principles concerning contractual and regulatory arrangements in the electricity/infrastructure context.

Relevance:
The case illustrates the importance of examining the legal character of economic arrangements rather than relying solely on economic assumptions.

5. All India Power Engineer Federation v. Sasan Power Ltd. (2016)

The Supreme Court considered issues concerning electricity procurement, tariffs, regulatory oversight and consumer interests.

Relevance:
Electricity pricing decisions can have substantial consequences for consumers and the wider power system. Quantitative evidence—including demand and price analysis—can therefore assist regulators in assessing the consequences of regulatory decisions.

17. International Case-Law Perspective

Regression-based economic evidence has also become important in competition and regulatory litigation internationally.

United States v. Microsoft Corp. (2001)

The litigation demonstrated the importance of economic evidence in assessing market behaviour and competitive effects.

Relevance to demand estimation:
Econometric evidence may assist courts in determining market relationships, consumer responses and competitive effects, although statistical evidence must be evaluated together with the broader factual record.

Daubert v. Merrell Dow Pharmaceuticals, Inc. (1993)

Although not an energy case, Daubert is highly significant for expert evidence.

The U.S. Supreme Court emphasized judicial scrutiny of scientific expert methodology, including whether methods are testable, subject to peer review, have known error rates, and are generally accepted.

Relevance:
Where regression analysis is presented as expert evidence, methodological reliability matters. A sophisticated-looking regression model is not automatically reliable merely because it contains statistical calculations.

18. Legal Standards for Reliable Regression Evidence

A regulator or court examining regression-based evidence should consider:

Data quality

Are the data accurate, sufficiently large, and representative?

Model specification

Have economically relevant variables been included?

Statistical significance

Are estimated coefficients statistically meaningful?

Economic significance

Is the estimated effect economically meaningful, rather than merely statistically significant?

Robustness

Do results remain similar when alternative specifications are used?

Causality

Does the methodology support a causal inference or merely establish correlation?

Transparency

Can the opposing party reproduce or challenge the analysis?

Consistency with law

Does the use of the model remain within the statutory authority of the decision-maker?

19. Importance for Energy Law

Regression models have increasing importance because modern electricity systems are becoming more complex.

Demand may now be affected by:

• rooftop solar;
• electric vehicles;
• battery storage;
• smart meters;
• time-of-use tariffs;
• demand-response programmes;
• distributed energy resources;
• energy-efficiency programmes; and
• changing industrial consumption.

Consequently, traditional models based only on historical consumption and price may become inadequate.

Modern demand estimation may need to incorporate:

\[ Q=f(P,Y,W,EV,RE,Storage,Technology,Policy) \]

where the variables include price, income, weather, electric-vehicle adoption, renewable generation, storage, technology and policy.

20. Conclusion

Regression models in demand estimation provide a systematic method for measuring how demand responds to price and other economic, technological, demographic and environmental factors. They are particularly important in electricity regulation because demand forecasts influence tariff determination, procurement, infrastructure investment and system planning.

However, regression analysis has limitations. Problems such as omitted variables, endogeneity, multicollinearity, autocorrelation, structural breaks and poor-quality data can produce misleading results. Consequently, regulators and courts should examine not merely the numerical outcome of a model but also its methodology, assumptions, data, robustness and legal relevance.

The Indian electricity cases, particularly **_West Bengal Electricity Regulatory Commission v. CESC Ltd.** and **PTC India Ltd. v. CERC_**, demonstrate the broader principle that technical and economic analysis operates within a statutory regulatory framework. Regression models can strengthen regulatory decision-making, but they cannot replace statutory authority, procedural fairness, reasoned decision-making, or judicial review.

In contemporary energy law, therefore, regression-based demand estimation should be understood as a quantitative evidence mechanism supporting economically rational and legally accountable regulation, rather than as an automatic formula for determining what the law requires.