Index Numbers: Foundations, Weighted Aggregates & Construction Methods

An index number is a specialized statistical measure designed to quantify the relative change in the magnitude of a group of related variables across time, geographic space, or economic sectors.

The Fundamental Index Problem

Let an economy encompass $N$ distinct commodities indexed by $i = 1, 2, \dots, N$. In base period $0$, each commodity $i$ is transacted at unit price $p_{i0}$ and traded in quantity $q_{i0}$. In current period $t$, the same commodity is transacted at unit price $p_{it}$ and quantity $q_{it}$. The individual price relative for commodity $i$ is:

Because individual price relatives diverge dramatically—some prices rise rapidly due to supply bottlenecks, others remain stable, and others fall due to technological automation—we cannot simply sum prices across commodities ($p_{\text{steel}} + p_{\text{wheat}}$ is meaningless). The foundational challenge of index number theory consists in finding a continuous aggregator function $I(p_t, q_t, p_0, q_0)$ that maps the $4N$-dimensional vector of prices and quantities into a meaningful scalar price index $P_{0t}$ and quantity index $Q_{0t}$.

Key Economic Applications of Index Numbers

1. Inflation Monitoring and Monetary Policy: Central banks (such as the Federal Reserve, the European Central Bank, and Bangladesh Bank) monitor the Consumer Price Index (CPI) and Personal Consumption Expenditures (PCE) deflator to calibrate interest rate decisions, inflation-targeting mandates, and quantitative easing operations.

2. Real Wage Indexation and Cost-of-Living Adjustments (COLA): Labor unions, public service pay commissions, and social security administrations index nominal wages, pensions, and minimum wage thresholds to price indices to protect workers' purchasing power from inflationary erosion.

3. National Accounts Deflation: Macroeconometricians divide nominal Gross Domestic Product (GDP) by the GDP deflator to isolate real physical output growth from pure monetary inflation.

4. Terms of Trade (ToT) in International Commerce: The net barter terms of trade, defined as the ratio of the export price index to the import price index ($P_X / P_M \times 100$), determines whether a sovereign nation's international purchasing power is appreciating or deteriorating.

The numerical value of any index number depends fundamentally on the benchmark against which comparisons are executed: the Base Period.

Criteria for Selecting a Valid Base Period

1. Economic Normalcy: The base period must be an economically normal year, free from severe macroeconomic dislocations, hyperinflation, wartime devastation, harvest failures, or global pandemics. Choosing an abnormal crisis year creates a distorted benchmark that skews subsequent comparisons.

2. Temporal Proximity: The base period should not be too distant in the past. Over time, consumer preferences evolve, new electronic products emerge, and older products become obsolete. If a base year is thirty years old, its fixed commodity basket no longer reflects modern consumption patterns.

Fixed-Base vs Chain-Base Systems

In economic statistics, temporal comparisons are structured according to two distinct operational architectures:

1. Fixed-Base Index System:

A single, designated historical period $0$ serves as the permanent denominator for all subsequent periods $t$:

• Advantages: Computationally straightforward and simple to explain to the public; comparisons between any two adjacent years are consistent with a fixed reference basket.
• Disadvantages: As the baseline recedes into the past, new goods cannot be easily incorporated, and structural shifts in consumer spending are ignored.

2. Chain-Base Index System:

The base period shifts continuously, with each period compared directly to its immediately preceding period, generating a sequence of link relatives:

The cumulative chain index relative to the initial origin is obtained by successive multiplication (chaining) of the link relatives:

• Advantages: Highly flexible; new technological commodities can be added and obsolete goods retired each year, and quantity weights reflect contemporary spending patterns.
• Disadvantages: If prices oscillate cyclically without a secular trend, chain indices can suffer from chain drift (where the chained index fails to return to unity after prices and quantities return to their starting values).

Shifting and Splicing of Index Series

When a statistical agency updates its fixed base year (e.g., rebasing a Consumer Price Index from 2010 = 100 to 2020 = 100), it creates two disjoint, non-comparable historical series: the old series terminating in 2020 and the new series beginning in 2020. To establish a continuous, unbroken longitudinal historical series, demographers and econometricians perform splicing.

Let $P_{0t}^{\text{old}}$ denote the old series with base year $0$, and let $P_{1t}^{\text{new}}$ denote the new series with base year $1$. At the overlapping splicing year $k$, both indices are recorded: $P_{0k}^{\text{old}}$ and $P_{1k}^{\text{new}} = 100$. To splice the old series onto the new base (converting historical pre-$k$ values into new-base equivalents):

Conversely, to splice the new series onto the old base:

This linear scaling factor ensures seamless mathematical continuity across the base-year transition.

The earliest historical index formulations aggregated prices without incorporating commodity quantities or expenditure shares.

1. Simple Aggregative Price Index

The ratio of the unweighted sum of commodity prices in the current period to the unweighted sum in the base period:

Fatal Mathematical Flaw (The Unit Dependence Problem): The simple aggregative index is extraordinarily sensitive to the arbitrary physical units in which prices are quoted. If the price of gold is quoted per metric ton instead of per gram, its numerical magnitude will completely dominate the sum, rendering price changes in all other commodities mathematically irrelevant. The simple aggregative index violates the fundamental Unit Test of index number theory.

2. Simple Average of Price Relatives

To eliminate unit dependence, one can compute individual dimensionless price relatives $R_i = p_{it} / p_{i0}$ and average them:

• Arithmetic Mean of Relatives:

• Geometric Mean of Relatives (Jevons Index, 1865):

• Harmonic Mean of Relatives (Coggeshall Index, 1887):

While the Jevons geometric index eliminates unit dependence and satisfies the time-reversal property, all unweighted averages suffer from a crippling economic defect: they treat every commodity as equally important. A hundred-percent price increase in table salt (which accounts for $0.05\%$ of household expenditure) exerts the identical statistical impact as a hundred-percent increase in residential rent (which accounts for $35\%$ of expenditure). To reflect true cost-of-living impacts, commodities must be weighted by their economic expenditure.

In modern economic statistics, weighted aggregative index numbers weight each commodity price by physical quantities transacted in either the base period, the current period, or a composite combination.

1. The Laspeyres Price Index (Étienne Laspeyres, 1871)

The Laspeyres Price Index ($P_L$) weights prices by the base-period quantities ($q_{i0}$):

The numerator represents the cost of purchasing the original base-period consumption basket at contemporary current-period prices. The denominator represents the actual expenditure incurred to purchase that same basket in the base period.

Expression as a Weighted Average of Relatives:

Multiply and divide each term in the numerator by $p_{i0}$:

where $w_{i0} = \frac{p_{i0} q_{i0}}{\sum p_{i0} q_{i0}}$ represents the base-period expenditure share of commodity $i$, satisfying $\sum w_{i0} = 1$. Thus, the Laspeyres index is mathematically equivalent to the arithmetic mean of price relatives weighted by base-period budget shares.

The Laspeyres Upward Substitution Bias:

Microeconomic consumer theory proves that when prices rise unevenly, rational consumers do not consume a rigid, fixed physical basket; instead, they substitute away from commodities whose relative prices have escalated toward goods that have become relatively cheaper. Because the Laspeyres index assumes a completely frozen physical basket ($q_{i0}$), it completely ignores consumer substitution. Consequently, the Laspeyres index systematically overstates the true cost-of-living inflation experienced by households.

2. The Paasche Price Index (Hermann Paasche, 1874)

The Paasche Price Index ($P_P$) weights prices by the current-period quantities ($q_{it}$):

The numerator represents the actual contemporary expenditure incurred in period $t$. The denominator represents what that exact same contemporary basket would have cost if purchased at base-period prices.

Expression as a Weighted Harmonic Mean of Relatives:

The Paasche index is mathematically equivalent to the harmonic mean of price relatives weighted by current-period expenditure shares:

where $w_{it} = \frac{p_{it} q_{it}}{\sum p_{it} q_{it}}$.

The Paasche Downward Substitution Bias:

Because the current basket $q_{it}$ reflects consumer substitution toward cheaper alternatives, evaluating that basket at base prices ($p_{i0}$) overstates base-period expenditure. Consequently, the Paasche index systematically understates true cost-of-living inflation.

In almost all empirical economies, the fundamental Laspeyres-Paasche inequality holds:

3. Fisher's Ideal Index (Irving Fisher, 1922)

Recognizing that Laspeyres overstates inflation while Paasche understates it, the Yale economist Irving Fisher proposed the geometric mean of the two indices as the Fisher Ideal Index ($P_F$):

Fisher designated this formula "ideal" because, as proven in Unit 2, it satisfies both the Time Reversal Test and the Factor Reversal Test—a rigorous mathematical achievement that neither Laspeyres nor Paasche can attain.

In addition to Laspeyres, Paasche, and Fisher, economic statisticians developed several compromise formulas designed to balance base-period and current-period baskets without requiring geometric square roots.

1. The Marshall–Edgeworth Index (Alfred Marshall, 1887; Francis Y. Edgeworth, 1925)

The Marshall–Edgeworth index ($P_{ME}$) weights prices by the arithmetic average of base-period and current-period quantities:

The Marshall–Edgeworth index provides an exceptional numerical approximation to Fisher's Ideal Index without requiring non-linear root evaluations. Furthermore, it satisfies the Time Reversal Test.

2. The Dorbish–Bowley Index (F. L. Dorbish, 1899; Arthur L. Bowley, 1901)

The Dorbish–Bowley index ($P_{DB}$) is the arithmetic mean of the Laspeyres and Paasche indices:

Because the arithmetic mean of two positive numbers always exceeds or equals their geometric mean ($AM \ge GM$), the Dorbish–Bowley index is always slightly larger than Fisher's Ideal Index:

3. The Walsh Price Index (Correa Moylan Walsh, 1901)

The Walsh index ($P_W$) weights prices by the geometric mean of base and current quantities:

The Walsh index belongs to the class of superlative index numbers (in the sense of Erwin Diewert), providing an exact second-order Taylor approximation to arbitrary consumer expenditure functions.

The Economic Theory of Index Numbers: Konüs True Cost-of-Living Index

While mathematical statisticians approach index numbers through axiomatic properties and aggregative formulas, economic theorists ground index numbers in microeconomic consumer optimization. In 1924, the Russian mathematical economist Alexander A. Konüs formulated the True Cost-of-Living Index (COLI).

Let u = U(q) represent a consumer's continuous, quasi-concave utility function over consumption bundles q. The minimum expenditure required to attain utility level u at price vector p is given by the consumer's Expenditure Function (or cost function):

Konüs defined the True Cost-of-Living Index as the ratio of minimum expenditures required to attain a specified base reference utility level u_0 under current prices p_t versus base prices p_0:

The Fundamental Konüs Bounds:

Konüs established that the observable Laspeyres and Paasche index numbers provide rigorous upper and lower empirical bounds for the unobservable true cost of living:

1. The Laspeyres Upper Bound:

Because the base-period bundle q_0 was chosen by the consumer under base prices p_0, it achieves utility U(q_0) = u_0. At current prices p_t, the consumer could choose to purchase the same bundle q_0, incurring cost p_t^T q_0. However, by substituting away from relatively expensive goods, the consumer can attain utility u_0 at a cost strictly less than or equal to p_t^T q_0:

Dividing both sides by base expenditure E(p_0, u_0) = p_0^T q_0 yields the celebrated Konüs Theorem:

The Laspeyres index is a strict mathematical upper bound to the true cost-of-living index!

2. The Paasche Lower Bound:

Similarly, evaluating the true cost of living relative to current-period utility u_t yields:

The Paasche index is a strict mathematical lower bound!

Combining the two bounds establishes that the true economic cost-of-living index is bounded between Paasche and Laspeyres:

Because Fisher's Ideal Index P_F is the geometric mean of Laspeyres and Paasche, it falls squarely inside this theoretical band, serving as an exceptional empirical approximation to the true cost of living.

Superlative Index Numbers and Diewert's Exact Quadratic Approximation

In 1976, W. Erwin Diewert revolutionized index number theory by formulating the concept of Superlative Index Numbers. An index number formula is defined as superlative if it is exact for a flexible consumer expenditure function—a function capable of providing a second-order Taylor series approximation to an arbitrary, twice-continuously differentiable utility function.

Diewert established two foundational theorems:

1. The Fisher Ideal Index is Superlative: Fisher's index P_F is exact for the flexible quadratic mean-of-order-r expenditure function when r = 2.

2. The Törnqvist Index is Superlative (Leo Törnqvist, 1936):

The Törnqvist index P_T weights logarithmic price changes by the arithmetic mean of base-period and current-period expenditure shares:

Diewert proved that the Törnqvist index is exact for the transcendental logarithmic (translog) expenditure function.

Because both Fisher and Törnqvist indices allow consumer expenditure weights to adapt smoothly to price changes, they virtually eliminate the substitution bias that plagues the fixed-basket Laspeyres Consumer Price Index.

Hedonic Quality Adjustment and Scanner Data Innovations

In modern digital economies, goods undergo rapid technological improvement: a modern personal computer or smartphone may cost the same nominal price as a model sold five years ago, but it features ten times the processing speed, memory, and photographic resolution. If statistical agencies simply record the nominal price, they drastically overstate true inflation because they ignore the colossal expansion in product quality.

To resolve this challenge, Zvi Griliches (1961, 1971) formalized Hedonic Price Regression:

where z_{ik} denotes the k-th physical characteristic of good i (e.g., processor speed, RAM, screen resolution, vehicle horsepower). The regression coefficient beta_k represents the marginal implicit shadow price of characteristic k. Statistical agencies use hedonic regressions to decompose observed price changes into a pure quality change component and a residual pure inflation component, ensuring that consumer price indices track true utility costs rather than technological upgrading.

The Economic Theory of Index Numbers: Konüs True Cost-of-Living Index

While mathematical statisticians approach index numbers through axiomatic properties and aggregative formulas, economic theorists ground index numbers in microeconomic consumer optimization. In 1924, the Russian mathematical economist Alexander A. Konüs formulated the True Cost-of-Living Index (COLI).

Let u = U(q) represent a consumer's continuous, quasi-concave utility function over consumption bundles q. The minimum expenditure required to attain utility level u at price vector p is given by the consumer's Expenditure Function (or cost function):

Konüs defined the True Cost-of-Living Index as the ratio of minimum expenditures required to attain a specified base reference utility level u_0 under current prices p_t versus base prices p_0:

The Fundamental Konüs Bounds:

Konüs established that the observable Laspeyres and Paasche index numbers provide rigorous upper and lower empirical bounds for the unobservable true cost of living:

1. The Laspeyres Upper Bound:

Because the base-period bundle q_0 was chosen by the consumer under base prices p_0, it achieves utility U(q_0) = u_0. At current prices p_t, the consumer could choose to purchase the same bundle q_0, incurring cost p_t^T q_0. However, by substituting away from relatively expensive goods, the consumer can attain utility u_0 at a cost strictly less than or equal to p_t^T q_0:

Dividing both sides by base expenditure E(p_0, u_0) = p_0^T q_0 yields the celebrated Konüs Theorem:

The Laspeyres index is a strict mathematical upper bound to the true cost-of-living index!

2. The Paasche Lower Bound:

Similarly, evaluating the true cost of living relative to current-period utility u_t yields:

The Paasche index is a strict mathematical lower bound!

Combining the two bounds establishes that the true economic cost-of-living index is bounded between Paasche and Laspeyres:

Because Fisher's Ideal Index P_F is the geometric mean of Laspeyres and Paasche, it falls squarely inside this theoretical band, serving as an exceptional empirical approximation to the true cost of living.

Superlative Index Numbers and Diewert's Exact Quadratic Approximation

In 1976, W. Erwin Diewert revolutionized index number theory by formulating the concept of Superlative Index Numbers. An index number formula is defined as superlative if it is exact for a flexible consumer expenditure function—a function capable of providing a second-order Taylor series approximation to an arbitrary, twice-continuously differentiable utility function.

Diewert established two foundational theorems:

1. The Fisher Ideal Index is Superlative: Fisher's index P_F is exact for the flexible quadratic mean-of-order-r expenditure function when r = 2.

2. The Törnqvist Index is Superlative (Leo Törnqvist, 1936):

The Törnqvist index P_T weights logarithmic price changes by the arithmetic mean of base-period and current-period expenditure shares:

Diewert proved that the Törnqvist index is exact for the transcendental logarithmic (translog) expenditure function.

Because both Fisher and Törnqvist indices allow consumer expenditure weights to adapt smoothly to price changes, they virtually eliminate the substitution bias that plagues the fixed-basket Laspeyres Consumer Price Index.

Hedonic Quality Adjustment and Scanner Data Innovations

In modern digital economies, goods undergo rapid technological improvement: a modern personal computer or smartphone may cost the same nominal price as a model sold five years ago, but it features ten times the processing speed, memory, and photographic resolution. If statistical agencies simply record the nominal price, they drastically overstate true inflation because they ignore the colossal expansion in product quality.

To resolve this challenge, Zvi Griliches (1961, 1971) formalized Hedonic Price Regression:

where z_{ik} denotes the k-th physical characteristic of good i (e.g., processor speed, RAM, screen resolution, vehicle horsepower). The regression coefficient beta_k represents the marginal implicit shadow price of characteristic k. Statistical agencies use hedonic regressions to decompose observed price changes into a pure quality change component and a residual pure inflation component, ensuring that consumer price indices track true utility costs rather than technological upgrading.

While classical index number theory focused predominantly on fixed-basket aggregative formulations (Laspeyres and Paasche), twentieth-century mathematical economics established rigorous microeconomic and axiomatic foundations through the work of François Divisia (1925), Erwin Diewert (1976), and Leo Törnqvist (1936).

1. The Continuous Divisia Index

In 1925, the French mathematical economist François Divisia proposed a continuous-time formulation of price and quantity indices. Let $p_i(t)$ and $q_i(t)$ represent the continuous price and quantity trajectories of commodity $i$ at time $t$. The aggregate expenditure at time $t$ is:

Differentiating both sides with respect to continuous time $t$:

Dividing by total expenditure $E(t)$:

Defining $w_i(t) = \frac{p_i(t) q_i(t)}{E(t)}$ as the instantaneous expenditure share of commodity $i$, and noting that $\frac{1}{x} \frac{dx}{dt} = \frac{d \ln(x)}{dt}$, the proportional growth rate of nominal expenditure decomposes into the sum of the instantaneous growth rates of the aggregate price index $P(t)$ and quantity index $Q(t)$:

where the continuous growth rate of the Divisia Price Index is defined as:

Integrating over a discrete calendar interval $[0, 1]$:

The Divisia index possesses remarkable theoretical properties: it satisfies both the Time Reversal Test and the Factor Reversal Test continuously. However, because economic data are collected at discrete intervals rather than continuously, the line integral cannot be evaluated without discrete approximation.

2. The Törnqvist Discrete Approximation

To approximate the continuous Divisia line integral using discrete data, Finnish statistician Leo Törnqvist (1936) replaced the continuous expenditure shares with the arithmetic mean of the base and current period expenditure shares:

Taking the exponential of the weighted sum of log price relatives yields the Törnqvist Price Index:

The Törnqvist index is symmetric with respect to time, treats both periods symmetrically, and accommodates smooth consumer substitution between goods.

3. Diewert's Theory of Superlative Index Numbers

In a landmark 1976 paper, Canadian econometrician W. Erwin Diewert introduced the concept of Superlative Index Numbers. A price index formula is defined as superlative if it is exact for a flexible unit cost function—meaning a cost function that provides a second-order Taylor series approximation to an arbitrary, twice continuously differentiable aggregator function.

Diewert established two foundational theorems:

1. Fisher's Ideal Index is Superlative: Fisher's index $P_F = \sqrt{P_L \times P_P}$ is exact for the flexible Quadratic Mean of Order $r = 2$ unit cost function:

2. The Törnqvist Index is Superlative: The Törnqvist index $P_T$ is exact for the flexible Translog (transcendental logarithmic) unit cost function:

Because both Fisher's Ideal Index and the Törnqvist Index are superlative, they accommodate flexible substitution behavior among consumers as relative prices shift, eliminating upper-level substitution bias. In practical empirical applications, numerical values of $P_F$ and $P_T$ rarely differ by more than a fraction of a percent.

Before statistical agencies can compute weighted aggregative indices (like Laspeyres or Fisher), they must first compile thousands of price quotations for individual items within detailed product categories where quantity weights are unavailable. These initial aggregations are termed Elementary Price Indices.

1. Classical Elementary Formulations

Let $p_{i,0}$ and $p_{i,t}$ denote the price of item $i$ ($i = 1, \dots, k$) in the base and current period within an unweighted elementary stratum.

• The Dutot Index (Ratio of Arithmetic Averages):

The Dutot index is sensitive to units of measurement unless all items within the stratum are homogenous in physical specification.

• The Carli Index (Arithmetic Average of Price Relatives):

The Carli index fails the Time Reversal Test: by Jensen's inequality for convex functions, the arithmetic mean of relatives and the arithmetic mean of inverted relatives satisfy:

The Carli index is systematically biased upward and has been largely banished from international statistical guidelines.

• The Jevons Index (Geometric Mean of Price Relatives):

The Jevons index satisfies both the Unit Test and the Time Reversal Test. Furthermore, it implicitly assumes that consumers maintain constant expenditure shares within the item category (unitary elasticity of substitution, $\sigma = 1$). For these reasons, the Jevons index is recommended by the international Consumer Price Index Manual as the preferred elementary aggregator.

2. High-Frequency Scanner Data and Chain Drift

In modern digital statistical systems, statistical offices increasingly ingest massive scanner barcode datasets and web-scraped daily transactions. To maintain timely product coverage, analysts frequently chain high-frequency monthly or weekly indices.

However, chaining indices under high-frequency sales promotions and seasonal restocking produces the phenomenon of Chain Drift. If retail prices fluctuate cyclically due to weekend promotions (dropping by 20% on sale and rebounding by 25% on Monday), while consumers purchase heavily at the discount price and abstain at regular price, chaining standard formulas generates a cumulative index that drifts systematically away from the true fixed-base price level. Resolving chain drift requires deploying multilateral index number methods, such as the GEKS (Gini-Eltetö-Köves-Szulc) or Rolling Year GEKS algorithms.