Balassa-Samuelson Effect

The puzzle the effect solves

If you walk into a barbershop in Zurich and ask for a haircut, expect to pay roughly 60 Swiss francs — about 65 US dollars. The same haircut in Manila costs about 200 Philippine pesos, roughly 3.50 US dollars. A factor of nearly twenty. The barbers are doing essentially the same job. Why the gap?

Purchasing power parity, the textbook starting point for thinking about exchange rates, would say that two identical goods or services should cost the same across countries after converting at the market exchange rate. The Economist's Big Mac Index makes this explicit for one tradable benchmark. But PPP fails systematically and predictably for services — and not in random directions. Richer countries are systematically more expensive in services. Poorer countries are systematically cheaper. The pattern is so robust that it has a name: the Penn effect, after the Penn World Tables data set that first documented it cleanly.

Bela Balassa and Paul Samuelson, in two independent papers published in 1964, supplied the canonical explanation. Their model is a two-sector economy with a single labour market, and from it falls one of the most reliable predictions in international economics: richer countries should be more expensive, with the gap roughly proportional to their lead in tradable-sector productivity. The model has been refined for sixty years, but the core mechanism remains the same.

The dual-sector mechanism

Think of every economy as having two sectors:

• Tradable sector (T). Manufactured goods, commodities, electronics, vehicles. These goods can be shipped across borders relatively cheaply. International arbitrage — the threat that buyers will source from cheaper suppliers — pins prices of tradables to a roughly common world level P_T.
• Non-tradable sector (N). Haircuts, restaurant meals, doctor visits, taxi rides, housing, schooling, dry cleaning. Produced locally, consumed locally. No international arbitrage. Prices P_N are determined by domestic supply and demand.

In a competitive labour market within a country, workers can move between the two sectors, so wages must be (approximately) equalised across them. Tradable workers earn W = A_T · P_T (real wage equals labour productivity in tradables times the tradable price, assuming labour is the dominant input). Non-tradable workers earn the same nominal wage W. The price of non-tradables must therefore satisfy

P_N  =  W / A_N  =  (A_T / A_N) · P_T

where A_N is non-tradable productivity. The general price level is some weighted average of P_T and P_N:

P  =  P_T^α  ·  P_N^(1-α)  =  P_T  ·  (A_T / A_N)^(1-α)

where α is the share of tradables in consumption. The key prediction follows immediately: a country with high tradable productivity A_T relative to non-tradable productivity A_N has a high domestic price level P. Since tradable prices are roughly equalised internationally and non-tradable productivity is similar across countries (a Swiss barber is not much faster than a Filipino barber), the prediction reduces to: richer countries — those with high tradable productivity — have higher price levels.

Worked example: Switzerland vs the Philippines

Make the model concrete. Assume both countries have non-tradable productivity normalised to 1 — a haircut is a haircut everywhere. Tradable productivity differs: A_T^CH = 5, A_T^PH = 1 (Swiss tradable workers are five times more productive). Tradable prices are equalised at the world level P_T = 1. Set the consumption share of tradables α = 0.4.

Variable	Switzerland	Philippines	Comment
Tradable productivity AT	5	1	Manufacturing productivity gap
Non-tradable productivity AN	1	1	Barbers cut hair the same
Tradable price PT	1	1	World market arbitrage
Wage W = AT·PT	5	1	Set by tradable sector
Non-tradable price PN = W/AN	5	1	Haircut costs five times more
Overall price level P = PT0.4·PN0.6	2.63	1	Switzerland 2.6× more expensive
Real exchange rate (P_CH / P_PH)	2.63		Predicted, before demand-side effects

Real-world numbers compress the gap somewhat — Switzerland is about 1.65× the US price level, not 2.6× — but the qualitative pattern is robust. The richer the country in tradable productivity, the higher its real exchange rate. The 35 percentage-point slope of the Penn-effect regression of log price level on log income per capita is consistent with α ≈ 0.4 and a productivity-gap structure broadly like Balassa-Samuelson's.

The Penn effect — the empirical bedrock

The most influential evidence for Balassa-Samuelson is the Penn effect: a robust cross-country regression of price level on real income per capita. Using IMF World Economic Outlook data, the slope of log price level on log GDP per capita (PPP) typically runs 0.30 to 0.40 across post-war samples. A country with twice the US real GDP per capita is predicted to have a price level roughly 30-40 percent higher than the US; one with half is predicted to be roughly 22 percent cheaper.

Penn-effect data for 2023:

Country	GDP per capita (PPP, USD)	Price level (US = 100)	Real exchange rate vs US
Switzerland	87,000	165	+65% (expensive)
Norway	82,000	130	+30%
United States	76,000	100	numeraire
Germany	63,000	95	−5%
South Korea	50,000	83	−17%
China	22,000	60	−40%
Vietnam	13,000	40	−60%
India	10,000	40	−60%
Ethiopia	2,800	30	−70%

The fit is not perfect — outliers reflect taxation differences, housing-market frictions, demand-side preferences and quality variation — but the systematic pattern is unmistakable. Cross-sectional R² in such regressions typically runs 0.5-0.7.

Implications for policy and analysis

1. PPP-adjusted GDP differs from market-rate GDP — by a lot

Cross-country GDP comparisons that use market exchange rates systematically understate the size of poorer economies. China's 2023 GDP at market rates was about USD 17 trillion; at PPP rates roughly USD 33 trillion — close to US-sized. The mechanism is exactly Balassa-Samuelson: Chinese goods and services are cheaper when measured in dollars at the market rate because Chinese non-tradable prices are lower. The IMF reports both for this reason.

2. Real exchange-rate appreciation is normal in catch-up growth

As an emerging economy raises its tradable productivity toward the frontier, its real exchange rate should appreciate. This is not a sign of policy failure, currency manipulation, or 'Dutch disease' — it is a structural feature of successful catch-up growth. South Korea's real exchange rate appreciated roughly 50 percent against major currencies between the 1970s and the 1990s as its tradable productivity caught up. China's real exchange rate appreciated roughly 30 percent between 2005 and 2015. Both are consistent with the Balassa-Samuelson mechanism operating cleanly.

3. Eurozone convergence and the periphery problem

Joining a currency union eliminates nominal exchange-rate adjustment but does nothing to suppress Balassa-Samuelson dynamics. Lower-productivity Eurozone entrants — Greece, Portugal, Spain — saw rising non-tradable prices as their tradable sectors caught up, while their nominal exchange rate was fixed to Germany's. The result was a real appreciation that hurt competitiveness, contributing to the 2010-12 crisis. Inside a currency union, Balassa-Samuelson convergence shows up as inflation differentials, not exchange-rate moves.

4. Welfare comparisons need PPP weights

If a Swiss worker earns 5,000 dollars a month and an Indian worker earns 1,000 dollars, the welfare gap is not 5×. Living costs differ. PPP-adjusted incomes correct for the Balassa-Samuelson price-level gap; raw market rates do not. International poverty lines (World Bank: 2.15 USD/day PPP, 2017 ICP) are constructed in PPP precisely because Balassa-Samuelson means that nominal dollars are not comparable.

Variants and extensions

• Bhagwati-Kravis-Lipsey (1984). Generalisation distinguishing capital intensity rather than just productivity as the driver. Richer countries are richer in capital and labour productivity, both of which raise non-tradable prices.
• Bergin, Glick, Taylor (2006). Show that Balassa-Samuelson holds clearly in panel data with multi-decade averaging but less so in pure cross-section, suggesting short-run real-exchange-rate variation is dominated by other forces.
• Demand-side Balassa-Samuelson. Engel curves are non-homothetic: richer households consume disproportionately more services. This raises demand for non-tradables in rich countries, further driving up P_N. Estimated to add 30-50 percent on top of the supply-side mechanism.
• Quality bias. What looks like 'the same' restaurant meal or haircut may differ in quality across countries. Some of the measured price gap is actually a quality gap. The OECD's Eurostat-OECD PPP programme tries to control for this through detailed item-by-item price collection.
• Trade-cost adjustments. Tradables are not freely tradable — they face shipping, tariff and non-tariff costs typically averaging 10-30 percent. This weakens the law-of-one-price assumption and adds noise to the Penn-effect regressions.
• Housing as the dominant non-tradable. Modern work (Cheng-Burstein-Eichenbaum 2018) emphasises that housing dominates non-tradable prices in advanced economies and that supply restrictions in major cities amplify the price-level gap beyond pure Balassa-Samuelson.

Common pitfalls

• "PPP must hold therefore Balassa-Samuelson is wrong." Absolute PPP does not hold across countries; relative PPP holds approximately over very long horizons for tradables only. Balassa-Samuelson is the structural explanation for why absolute PPP fails.
• "The model predicts haircuts will eventually equalise." Only if productivity gaps close completely. As long as tradable productivity differs, non-tradable prices differ — equilibrium, not transition.
• "It applies only to manufacturing-vs-services." No — the relevant distinction is tradability, not sector. Some manufactured goods are non-tradable (heavy concrete blocks); some services are tradable (software, finance). The category matters.
• "It explains real exchange rates fully." It accounts for 30-60 percent of cross-country variation. The rest — government spending share, monetary regime, capital controls, demand-side preferences — adds noise that the basic Balassa-Samuelson model does not capture.
• "China's real exchange rate undervaluation refutes Balassa-Samuelson." No — the model predicts that China should appreciate as it catches up. The persistent undervaluation in the 2000s reflected explicit policy choices (reserve accumulation, capital controls) layered on top of the structural mechanism. Once those eased post-2014, the real exchange rate strengthened broadly as predicted.
• "The effect operates only for advanced economies." The effect runs continuously across the income distribution; both the rich-vs-middle and middle-vs-poor margins are well-documented. The biggest absolute price-level gaps are in fact between the richest and poorest economies.