Skip to content

Demand Projection

As a demand-driven optimization model, KiNESYS projects future useful energy demands by combining macroeconomic drivers with empirically estimated elasticities. This methodology ensures that demand projections are grounded in observed historical relationships while allowing for region-specific variations in economic structure and development pathways.

Overview

The demand projection framework operates in three stages:

+-------------------------------------------------------------------------+
|                        1. DRIVER PREPARATION                            |
+-------------------------------------------------------------------------+
|                                                                         |
|   SSP Database ------> Base Drivers ------> Sectoral Drivers            |
|   (GDP, Population,    (GDP, POP, HOU,      (GDP_AGR, GDP_IND,           |
|    Households)          GDPPCAP, GDPPHOU)    GDP_SRV, GDP_MFG)           |
|                                                                         |
|   World Bank WDI -----> Sector Shares -----> Convergence to 2100        |
|   (Agriculture,         (normalized to       (developing economies      |
|    Industry, Services)   sum to 100%)         converge to OECD          |
|                                               structure)                |
+-------------------------------------------------------------------------+
                                  |
                                  v
+-------------------------------------------------------------------------+
|                     2. ELASTICITY ESTIMATION                            |
+-------------------------------------------------------------------------+
|                                                                         |
|   Historical Data ----> Log-Log Regression ----> Quality Filtering      |
|   (sector outputs,      (OLS estimation of       (significance,         |
|    driver values)        elasticity)              fit, bounds)          |
|                                                                         |
|                               |                                         |
|                               v                                         |
|                         Driver Selection                                |
|                         (best driver per                                |
|                          region-sector by R-squared)                    |
|                                                                         |
|                               |                                         |
|                               v                                         |
|                      Hierarchical Fallback                              |
|                      (region -> aggregate -> world)                     |
+-------------------------------------------------------------------------+
                                  |
                                  v
+-------------------------------------------------------------------------+
|                      3. DEMAND PROJECTION                               |
+-------------------------------------------------------------------------+
|                                                                         |
|   Base Year Demand x (Driver_t / Driver_base)^elasticity = Demand_t     |
|                                                                         |
+-------------------------------------------------------------------------+

Driver Preparation

Macroeconomic Drivers

The primary macroeconomic drivers are sourced from the SSP (Shared Socioeconomic Pathways) Database, which provides internally consistent projections of population, economic growth, and urbanization across five socioeconomic narratives.

Base Drivers

Driver Definition Typical Application
GDP Gross Domestic Product (PPP, billion USD) Industrial output, freight transport
POP Population (millions) Residential cooking, passenger rail
HOU Households (millions) Residential space heating/cooling, refrigeration
GDPPCAP GDP per capita (thousand USD/person) Passenger car travel, residential lighting
GDPPHOU GDP per household (thousand USD/household) Residential appliances, commercial services

Households are estimated as half of the population between 18 and 65 years of age.

Sectoral GDP Drivers

Total GDP is disaggregated into sectoral components using value-added shares from the World Bank Development Indicators.

Sector Definitions

Driver WDI Indicator Application
GDP_AGR Agriculture, forestry, and fishing (% of GDP) Agricultural energy demand
GDP_IND Industry including construction (% of GDP) Industrial sectors (mining, construction, other industry)
GDP_SRV Services (% of GDP) Commercial sector demands
GDP_MFG Manufacturing (% of Industry) Manufacturing subsectors

Structural Convergence

Sector shares evolve over the projection horizon according to a structural convergence assumption: developing economies gradually shift toward the sectoral composition observed in industrialized economies. This reflects the empirical regularity that:

  • Agriculture's share of GDP declines with development
  • Industry's share rises initially, then stabilizes
  • Services' share rises continuously with income

The convergence is implemented through linear interpolation from base year shares toward long-term target shares:

\[s_{i,t} = s_{i,\text{base}} + \frac{t - t_{\text{base}}}{t_{\text{horizon}} - t_{\text{base}}} \times (s_{i,\text{target}} - s_{i,\text{base}})\]

where:

  • \(s_{i,t}\) is sector i's share of GDP at time t
  • \(s_{i,\text{base}}\) is the observed base year share
  • \(s_{i,\text{target}}\) is the long-term target share

The services share is calculated as a residual to ensure shares sum to unity.

Elasticity Estimation

Econometric Methodology

Demand elasticities are estimated using log-log ordinary least squares (OLS) regression on historical data. This specification assumes a constant elasticity relationship between demand and its driver:

\[\ln(D_{r,s,t}) = \alpha_{r,s} + \beta_{r,s} \cdot \ln(X_{r,s,t}) + \varepsilon_{r,s,t}\]

where:

  • \(D_{r,s,t}\) is demand for sector s in region r at time t
  • \(X_{r,s,t}\) is the driver value
  • \(\beta_{r,s}\) is the elasticity (the parameter of interest)
  • \(\alpha_{r,s}\) is the intercept
  • \(\varepsilon_{r,s,t}\) is the error term

The coefficient \(\beta\) represents the percentage change in demand associated with a one percent change in the driver---the demand elasticity.

Quality Filtering

Estimated elasticities undergo rigorous quality control:

Statistical Significance

Only relationships with p-value < 0.05 are retained, ensuring the estimated elasticity is statistically distinguishable from zero.

Model Fit

Adjusted R-squared thresholds filter out poorly fitting models. The threshold is configurable by time period, allowing stricter criteria for periods with more stable data.

Elasticity Bounds

Elasticity values are evaluated against economic reasonableness:

  • 0 < beta <= 2: Accepted without review
  • 2 < beta <= 3: Flagged for expert review
  • beta > 3 or beta <= 0: Rejected as implausible

Data Exclusions

The years 2020 and 2021 are excluded from elasticity estimation due to the anomalous economic and energy consumption patterns induced by the COVID-19 pandemic. Including these years would distort the estimated long-run relationships.

Driver Selection

A key feature of the methodology is that drivers are not assigned a priori. Instead, multiple candidate drivers are evaluated for each sector, and the best-performing driver is selected based on statistical fit.

For each region-sector combination:

  1. All candidate drivers are regressed against historical demand
  2. Regressions passing quality filters are ranked by Adjusted R-squared
  3. The highest-ranking driver is selected

This data-driven approach ensures that driver assignments reflect actual historical relationships rather than theoretical assumptions. The result is that different regions may use different drivers for the same sector---for example, passenger car travel may be driven by GDP per capita in developed economies but by population growth in rapidly urbanizing developing economies.

Hierarchical Fallback

When a region has insufficient historical data or poor model fit at the individual region level, the system employs hierarchical fallback:

  1. Model Region: First attempt estimation at the finest regional granularity
  2. Aggregate Region: If unsuccessful, pool data across similar regions
  3. World: As a last resort, use global average elasticity

This ensures every region-sector combination receives a defensible elasticity estimate, even for regions with sparse data.

Demand Projection

Projection Formula

Future demands are projected using the constant elasticity formula:

\[D_{r,s,t} = D_{r,s,\text{base}} \times \left( \frac{X_{r,s,t}}{X_{r,s,\text{base}}} \right)^{\beta_{r,s}}\]

where:

  • \(D_{r,s,t}\) is projected demand for sector s in region r at time t
  • \(D_{r,s,\text{base}}\) is the calibrated base year demand
  • \(X_{r,s,t}\) is the projected driver value from SSP scenarios
  • \(X_{r,s,\text{base}}\) is the base year driver value
  • \(\beta_{r,s}\) is the estimated elasticity

This formulation preserves the base year calibration while allowing demands to evolve according to projected driver growth and empirically observed responsiveness.

Interpretation

Consider a region with:

  • Base year residential appliance demand: 100 PJ
  • GDP per household elasticity: 0.8
  • Projected GDP per household growth: 50% by 2050

The projected 2050 demand would be:

\[D_{2050} = 100 \times (1.50)^{0.8} = 100 \times 1.38 = 138 \text{ PJ}\]

The elasticity of 0.8 implies that a 1% increase in GDP per household leads to a 0.8% increase in appliance demand---reflecting efficiency improvements and saturation effects that prevent demand from growing proportionally with income.

Demand Categories

KiNESYS models demands across five major sectors. The appropriate driver for each demand is determined empirically through the elasticity estimation process described above.

Residential Sector

Code Description Unit
RCK Cooking PJ
REA Appliances PJ
RHW Hot water PJ
RLI Lighting PJ
ROT Other PJ
RRF Refrigeration PJ
RSC Space cooling PJ
RSH Space heating PJ

Commercial Sector

Code Description Unit
CCK Cooking PJ
CHW Hot water PJ
CLA Lighting and appliances PJ
COE Other electric PJ
COT Other PJ
CRF Refrigeration PJ
CSC Space cooling PJ
CSH Space heating PJ

Industry Sector

Code Description Unit
IFP Food processing PJ
ILP Pulp and paper Mt
IMC Mining and construction PJ
INF Non-ferrous metals Mt
INM Non-metallic minerals Mt
IOI Other industry PJ
IPC Petrochemicals Mt
IS Iron and steel Mt
ITM Transport equipment and machinery PJ
ONO Other non-specified PJ

Transport Sector

Code Description Unit
Trd_2w Road: 2-wheelers Bvkm
Trd_3w Road: 3-wheelers Bvkm
Trd_bus Road: buses Bvkm
Trd_car Road: passenger cars Bvkm
Trd_hdt Road: heavy trucks Bvkm
Trd_lcv Road: light commercial vehicles Bvkm
TAD Aviation: domestic PJ
TAI Aviation: international PJ
TTF Rail: freight PJ
TTP Rail: passenger PJ
TWD Shipping: domestic PJ
TWI Shipping: international PJ

Agriculture and Other

Code Description Unit
AGR Agriculture PJ
NEU Non-energy uses PJ

Data Sources

SSP Database

Shared Socioeconomic Pathways projections for GDP, population, and households. Version 3 provides projections for years 2023-2100 in 5-year intervals, with annual interpolation for near-term years.

World Bank Development Indicators

GDP sectoral shares (agriculture, industry, services, manufacturing as percentage of GDP). Latest available year per country, normalized and projected to 2100 using convergence assumptions.

Historical Energy Balances

IEA and regional energy statistics for base year demand calibration and elasticity estimation.

Regional Mapping

Countries are aggregated to model regions using a configurable mapping table that supports multiple regional definitions.