Leontief input-output Model

Wassily Leontief published the first input-output table for the US economy in 1936 and received the Nobel Prize in 1973 for the framework. The core idea: each industry both produces output and consumes output from other industries as intermediate inputs. Steel goes into cars, which go into rental fleets, which provide services to tourists. A shock to final demand for any good ripples backward through these supply chains. The input-output model traces these ripples by solving a system of linear equations derived from the observed inter-industry flow table.

The mechanics rest on one accounting identity and one behavioral assumption. The identity: each sector's total output equals its intermediate sales to all sectors plus its final demand (consumption, investment, government, exports). The assumption: the ratio of intermediate input from sector $i$ to total output of sector $j$---the technical coefficient $a_{ij}$---is fixed. This fixed-proportions assumption (no substitution between inputs) is the model's engine and its main limitation. Given the technical coefficient matrix $A$, the total-output vector $x$ satisfies $x = Ax + d$, where $d$ is final demand. Solving: $x = (I - A)^{-1} d$, where $(I - A)^{-1}$ is the Leontief inverse.

The Leontief inverse is remarkable. Each element $(I - A)^{-1}_{ij}$ gives the total output sector $i$ must produce---directly and through all layers of intermediate demand---to deliver one unit of final demand in sector $j$. The column sums are output multipliers: the total economy-wide production triggered by a unit of final demand in each sector. These multipliers capture the full supply-chain cascade, including indirect effects that propagate through third, fourth, and deeper tiers of suppliers.

Input-output tables are compiled by national statistical offices worldwide. The Bureau of Economic Analysis (BEA) publishes US tables roughly every five years with annual updates. The OECD maintains harmonized IO tables for 60+ countries. WIOD (World Input-Output Database) and GTAP provide multi-country tables that trace international supply chains. Central banks, finance ministries, and development organizations use these tables for fiscal multiplier analysis, trade-policy evaluation, carbon-footprint accounting, and disaster-impact assessment. The World Bank uses IO models to estimate the economic impact of natural disasters on small island economies.

The primary input is an $n \times n$ inter-industry flow matrix $Z$, where $z_{ij}$ records the dollar value of output from sector $i$ purchased by sector $j$ as an intermediate input during a reference year. Alongside $Z$, the model uses a vector of total outputs $x$ (one per sector) and a final-demand vector $d$ (consumption + investment + government + exports minus imports). The accounting identity is $x_i = \sum_j z_{ij} + d_i$ for each sector $i$. These data come from national accounts and supply-use tables.

The technical coefficient matrix $A$ is computed as $a_{ij} = z_{ij} / x_j$: the share of sector $j$'s total output spent on inputs from sector $i$. Each column of $A$ sums to less than one (the remainder is value added: wages, profits, taxes). The Leontief inverse $L = (I - A)^{-1}$ captures all rounds of inter-industry demand. Its power-series expansion $L = I + A + A^2 + A^3 + \cdots$ converges because the column sums of $A$ are strictly less than one, ensuring the spectral radius $\rho(A) < 1$.

Extensions enrich the basic framework. The Ghosh supply-driven model transposes the logic, treating output as given and allocating it across buyers. Environmentally extended IO (EEIO) augments $A$ with emission intensity vectors to compute carbon footprints of final-demand categories. Multi-regional IO (MRIO) stacks country-level tables and links them through bilateral trade flows, enabling analysis of global value chains. Hypothetical extraction (removing a sector from $A$ and solving for the reduced output) quantifies how much total output depends on a specific sector.

The Bureau of Economic Analysis (BEA) maintains the benchmark US input-output tables and uses them to estimate GDP by industry, compute industry-specific multipliers, and reconcile the production and expenditure sides of the national accounts. When the US government evaluates a fiscal stimulus package, IO multipliers provide a first-pass estimate of which sectors benefit most from different spending categories. Defense spending has different upstream linkages than infrastructure spending, and the IO model quantifies these differences sector by sector.

Environmental input-output analysis (EEIO) traces greenhouse gas emissions through supply chains. By augmenting the technical coefficient matrix with emission intensities (tons of CO2 per dollar of output), the model computes the total carbon footprint of any final-demand category. The EPA's USEEIO model and the EXIOBASE multi-regional model are standard tools. They reveal that the majority of a product's carbon footprint often lies in upstream supply chains, not in the final production step---information that is invisible without the IO framework.

Disaster-impact assessment uses IO models to estimate how a natural disaster in one sector propagates through the economy. If a hurricane destroys 30% of a port's capacity, the direct output loss is clear, but the indirect losses---manufacturers who cannot receive imported inputs, retailers who cannot get products---require the IO cascade. The World Bank and UNDRR (United Nations Office for Disaster Risk Reduction) use IO-based models for post-disaster needs assessment in developing countries.

The model fails for any question involving price adjustment, substitution, or equilibrium feedback. A carbon tax changes relative prices, causing firms to substitute away from carbon-intensive inputs. The IO model, with its fixed coefficients, cannot capture this. For price-sensitive analysis, computable general equilibrium (CGE) models embed IO accounting within a system of supply-and-demand equations with flexible prices. IO also fails for long-run structural change: the technical coefficients of the 2025 US economy differ substantially from those of 2010, so multipliers computed from old tables decay in accuracy.