The Real Cause of Inflation — Money Supply, Expectations, or Supply Chains?

Prices go up. Then we instinctively look for the culprit. Did someone print too much money? Is it because wages went up? Because of oil prices? Or is it simply that people raised prices in advance thinking, "They're going to go up anyway"?

The phenomenon of prices rising across the board is called inflation. Even economists do not answer with one voice regarding its causes. Some say, "Ultimately, it is because too much money was printed." Some say, "Prices and unemployment are a trade-off." Others say, "If people believe prices will rise, they actually do." Since 2020, the explanation of "pandemics and energy shocks" has been added.

This article spreads out four conflicting explanations on a single map. They are the quantity theory of money (the amount of money determines prices), the Phillips curve (the relationship between unemployment and prices), expectations (people's forecasts about future prices), and supply shocks (sudden cost shocks arising from raw materials, energy, and logistics). They are different maps drawn by four schools of thought. We will follow what each map explains well and where it loses its way. Then we will see why inflation cannot be neatly summarized by a single cause.

Almost all papers cited here are preprints (arXiv) released before peer review. Therefore, not a single line can be concluded as a final verdict. This uncertainty itself is part of this story.

First Map — "Prices Rise When Too Much Money Is Printed"

This is the oldest and most intuitive explanation. When the money in circulation increases, more money chases the same goods. That is why prices go up. The total amount of money circulating in the economy is called the money supply. The explanation that the money supply is the root of inflation is the quantity theory of money.

The data partially supports this intuition. Data from the US and the eurozone were examined to see how different brackets of money growth rates affect the distribution of inflation. The bracket with higher growth rates (the upper quantile) had a statistically clear impact on the inflation distribution. The impact of the lower and middle brackets was negligible (arXiv:2308.05486). However, the size of this impact changed over time. This can be read as an asymmetric picture where nothing much happens when money is printed moderately, but the connection comes alive when it is printed excessively.

However, this map has blurred over time. Tracking Ghana from the 1970s onward, the degree to which inflation responds to money growth (elasticity) continues to fall. By the final period, it becomes statistically insignificant. Based on periods that showed negative elasticity, this paper argues that the introduction of inflation targeting in 2002—a method where a central bank sets a price target and operates policy accordingly—was justified. It recommends maintaining and strengthening targeting, arguing that it contributed greatly to anchoring expectations (arXiv:1805.11562).

There is also a view that goes a step further. When a central bank operates interest rates and inflation targets together, the money supply becomes the result rather than the starting point of policy. This state, where the money supply is determined within the economy, is called endogenous. At this point, an analysis suggests that if the central bank can be completely trusted, the price theory only needs the Fisher equation—the relation that the nominal interest rate is the sum of the real interest rate and expected inflation. The same study adds a caveat. If trust is imperfect or expectations are not entirely rational, more than that is needed. Also, if policy is viewed as lasting only for a finite period, the money supply will eventually become a variable determined from the outside (exogenous) once again. Then people have an incentive to look at money supply statistics starting today to form inflation forecasts (arXiv:2301.11207). This is a signal that the money supply is only one facet of inflation, not the whole story.

The Second Map — "The Trade-off Between Unemployment and Inflation"

The second map looks at the labor market. When the economy overheats and unemployment falls, wages and prices rise. When a recession arrives, price pressures subside. The curve that plots this inverse relationship between unemployment and inflation is the Phillips curve. In long-term data from Japan, a statistically reliable Phillips curve was confirmed (R²=0.68, arXiv:1002.0277). This is a model that treats both inflation and unemployment as functions of the rate of change in the labor force. In this model, when unemployment increased, inflation decreased, and the two moved together at the same time (a negative simultaneous relationship).

However, this curve does not hold everywhere and at all times. Looking across 41 countries from 1980 to 2016, consistent cracks appear. The subject of verification is the expectations-augmented Phillips curve, which incorporates people's price expectations (expectations) into the mix. During calm periods, it held true in most developed countries, but it did not work well in emerging and frontier markets (markets less developed than emerging economies). And during recessions, the relationship completely broke down even in developed countries. That space was largely and distinctly filled by a component following past inflation (backward-looking element) and a component following future expectations (forward-looking element) (arXiv:2511.22785, arXiv:2511.22786, arXiv:2512.01697). The paper interprets this as a sign that markets become more sensitive to inflation during recessions. The Phillips curve is a conditional map that mainly operates in developed countries during calm periods.

The shape of the curve itself is also a matter of debate. Has the structural slope of the curve (the "skeleton") changed? Or has the property of the curve bending during extreme labor market overheating (non-linearity, the "shape") been turned on? There are attempts to distinguish between these using U.S. pre- and post-pandemic data (arXiv:2506.14030).

Apart from the debate over the skeleton and shape, studies also differ in their answers as to which variables drive inflation. One must account for the fact that countries and time periods differ. In Indian data, the curve was strongly non-linear, with prominent thresholds and interactions between variables. What most strongly drove headline inflation (based on overall consumer prices) was expectations. Next came past inflation and the output gap (the difference between the output the economy actually produced and the output it can sustainably generate). Supply shocks were negligible except for rainfall (arXiv:2504.05350). In contrast, a neural network Phillips curve showed that the 2021 surge was due to a large positive output gap starting in late 2020 (arXiv:2202.04146).

Why has the curve flattened? Flattening means that even when unemployment falls, prices do not rise as much as they used to. A dual labor market model (a model that views the labor market as divided into stable and unstable jobs) points to three causes. They are workers' bargaining power, the labor supply elasticity of wages in the secondary labor market, and changes in labor force composition (arXiv:2103.06482). There are also attempts to redraw the curve. One theoretical model proposes writing the curve using labor market tightness (how few unemployed people there are relative to the number of job openings) instead of unemployment. Then, even if the Beveridge curve (the curve plotting the relationship between unemployment and job openings) shifts, the Phillips curve is not affected. This model is called the 'Beveridge-Phillips curve' (arXiv:2401.12475). Predictive research that reformulates the curve using unobservable factors (latent variables) is also one such attempt (arXiv:2601.11601).

The Third Map — "If You Believe It Will Rise, It Rises"

The third map looks inside people's heads. Expectations are the forecasts of businesses and households about how much prices will rise in the future. When businesses and households believe that "prices will rise in the future," they raise prices and demand wages to match that belief. Therefore, expectations themselves create inflation. This is the same context in which expectations were the top driver in the Indian data we saw earlier (arXiv:2504.05350). A machine learning analysis of 20 advanced economies also pointed out that the role of expectations stands out. However, that importance appears to have weakened somewhat over the past 10 years (arXiv:2208.14653).

One of the most fascinating findings dealing with expectations is the threshold of attention. Attention is how much people pay attention to inflation news. This attention is estimated to change abruptly around an inflation rate of about 4%, and double once it crosses the threshold. According to this estimate, when attention is high, adverse supply shocks lead to greater inflation. An increase in people's attention in 2021 is analyzed to explain half of the subsequent supply-driven inflation (arXiv:2308.09480). In this study's model, a shock that would usually end quickly turns into stubborn inflation when it catches people's attention. Conversely, when attention is low, managing expectations (guiding people's price forecasts to a desired level) becomes difficult. Then, trapped at a lower bound where interest rates cannot be lowered any further while prices also remain low, one can fall into an "attention trap" (arXiv:2105.05297).

Then, what are expectations made of? An analysis looking at U.S. household monthly data from 1978 to 2016 reported that how much people are exposed to inflation and monetary policy news helps explain expectations. Since 1983, an asymmetry also exists. News that "prices are rising" and news that "monetary policy is easing" work more strongly than news in the opposite direction (arXiv:2009.11557). There was also an experiment mimicking British household surveys using a large language model (GPT) that had not learned about inflation surges. In the short term, GPT followed the overall average forecast of the survey and official statistics. It also reproduced regularities in perceptions by income, housing type, and class. Like humans, it was particularly sensitive to food inflation information, but it did not have a consistent consumer price model (arXiv:2512.14306).

A state where expectations are tied near the central bank's inflation target and do not waver easily is called anchoring. A theoretical analysis showed in a textbook New Keynesian Phillips curve model that a minimum level of trust—where the probability of the central bank breaking its promise is not 100 percent—is a necessary condition for anchoring. If there is no trust at all (if promises are invariably broken), the pattern in which inflation moves (dynamics) becomes locally unstable (arXiv:2012.02662).

Fourth Map — "Pushed Up by the Pandemic and Energy"

The fourth map has received the most attention since 2020. Pandemic lockdowns, supply chain collapses, and the energy price surges triggered by Russia's invasion of Ukraine. This explains how such supply-side shocks, or supply shocks, pushed up costs and drove up prices.

Researchers analyzed the years 2020 to 2023 as a scenario inside Mark-0, an agent-based model (a simulation where households and firms operate according to behavioral rules). They plugged in pandemic-driven inflation by dividing it into three parts: production and consumption shocks, supply chain shocks, and energy price shocks. Various explanations such as demand-pull (overheated demand), cost-push, and profit-driven inflation (firms widening their profit margins) were also examined together (arXiv:2306.01284).

There are results from applying a time-varying distributional regression model to United States inflation to split the risks. The risk of prices falling sharply (the downside) was found to stem mainly from weakening demand and inflation persistence (the property where prices once raised do not easily come down). The risk of prices rising sharply (the upside) stemmed mainly from supply shocks, especially energy prices. However, the effect of energy shocks varied significantly across time periods and was asymmetric (arXiv:2403.12456). There is also an analysis that United States inflation risks during the recent inflation surge period were driven primarily by economic recovery and soaring raw material prices (2024). Monetary policy adjustments and credit spreads (the interest rate gap between risky bonds and safe bonds) partially mitigated this (arXiv:2405.17237). A 2020 analysis modeling the Federal Reserve's perspective placed long-term expectation trends, the Phillips curve, and energy price fluctuations as the three components of inflation. When estimated, energy prices acted on prices not only through the Phillips curve but also through an independent expectations channel, and the energy price cycle overpowered the Phillips curve, explaining the inflation puzzle of the past decade (arXiv:2006.14110).

Yet how deep and long-lasting are supply shocks? Here the maps diverge again. In the euro area, demand factors are shown to account for at least 30% of the post-pandemic rise in core inflation (price increases excluding volatile items like food and energy). This suggests it is not purely a supply story (arXiv:2505.05536). In Israel, lockdowns changed the composition of consumption, which could have caused bias (a tilted error in one direction) in official price indices. This is because official price indices are calculated using a fixed basket of goods. However, that bias was surprisingly small and temporary (arXiv:2506.09875). A supply shock is clearly one pillar, but it is not a sole culprit untangled from demand and expectations. Furthermore, shocks do not fall equally on everyone. An analysis suggests that supply-driven inflation disproportionately increases the unemployment tail risk (the possibility of rare but severely bad outcomes) for workers vulnerable to economic conditions (arXiv:2505.05757).

So, Who Is the Culprit?

When you overlap four maps, one thing becomes clear: there is no single cause. The money supply shows a live signal when it grows excessively (arXiv:2308.05486). The Phillips curve operates mostly in advanced economies during calm periods (arXiv:2511.22785).

Expectations amplify shocks when they cross the threshold of attention (arXiv:2308.09480). Supply shocks raise upward risk, but they are intertwined with demand (arXiv:2403.12456, arXiv:2505.05536). Recent inflation is closer to a complex phenomenon where all four operated simultaneously.

Can we catch it with policy, then? We need to be honest here as well. Monetary policy is a useful tool, but the economic control power reported by primary studies had been systematically overestimated. This is the conclusion of a study synthesizing meta-analyses (methods that combine and re-analyze the results of multiple studies). Once you strip away the bias (publication bias) where statistically significant results are more easily published, that power is more moderate, uncertain, and subject to long time lags (arXiv:2509.19591).

One survey summarizes that there is no persuasive evidence in advanced economies that inflation targeting improves price performance. The benefits may have been larger in emerging economies. The effect of improving fiscal discipline has very limited evidence, but is supported. Taken together, it concludes that there are no macroeconomic benefits superior to other monetary strategies, or if any, they are negligible (arXiv:2305.17474). Another analysis notes that it has not found strong evidence that inflation is necessarily lower just because a central bank has been independent for a long time (arXiv:2003.02208).

The earlier Mark-0 model reached a more provocative conclusion. In this model, getting 2020s inflation under control was less about the direct effect of interest rate hikes and more about the anchoring of expectations. However, strong anchoring is harmful to consumption and unemployment. Thus, this model points out that the window for an "optimal" policy is very narrow (arXiv:2306.01284).

What about forecasting? Inflation is notoriously difficult to forecast. All sorts of techniques have been mobilized, including neural networks, transformers, and Bayesian nonparametrics. Some methods beat the comparison baseline (benchmark) for specific inflation types and horizons. Transformers showed improvements in 6 out of 16 experiments. Bayesian nonparametric models showed improvements overall and in the left tail (the low inflation side).

Yet it remains difficult to consistently beat a random walk (the simplest prediction that leaves the next value equal to the current value). The performance of Long Short-Term Memory (LSTM) networks is summarized as "competitive but not overwhelming" (arXiv:2303.15364, arXiv:2202.13793, arXiv:2104.03757).

The Bank of England's forecasts are evaluated as having been "fairly accurate compared to the benchmark" despite recent large errors (arXiv:2501.07386). According to one analysis, when divided by whether realized inflation at the time of forecasting is below the target, there is a bias in the Bank of Korea's forecasts. Simple correction methods generally increased accuracy (arXiv:2512.16068).

International factors also exist. Data from multiple countries point to the U.S. inflation rate and oil prices as the dominant drivers of inflation in each country (arXiv:2212.05841). In monthly data from a larger number of countries, international factors were relevant in explaining the high-inflation side (the right tail of the distribution) in many advanced economies. This was more so during high inflation than during average or low inflation. During the recent high-inflation period, the predictive power of international factors also grew. However, inflation in low-income developing countries was barely affected by international conditions, and middle-income countries showed mixed results (arXiv:2410.20628).

Conclusion

The story of inflation is fascinating because there is no single right answer. The quantity theory of money provides a strong intuition that "printing too much money" is the cause, but that connection grew weaker over time in Ghana. It also varied by period in the United States and the eurozone. The Phillips curve depicts the trade-off between unemployment and inflation, but it breaks down during recessions. Expectations are a hidden variable that amplifies shocks, but what creates expectations is still under study. Supply shocks are one pillar of recent inflation, but they cannot be separated from demand and expectations.

History shows that the real effects of inflation are also not simple. The hyperinflation in Germany from 1919 to 1923 significantly reduced the real burden and bankruptcy of nominal debt (debt whose amount is fixed in monetary digits). Companies that had more nominal debt at the start of inflation saw their interest burdens drop more and had higher employment. This result aligns with the debt-inflation channel that operates even when prices and wages are flexible (arXiv:2405.13296). A panel of four South Asian countries showed that higher political stability was associated with lower inflation (arXiv:2310.08415). There are even arguments that find the root of inflation in production inefficiencies (arXiv:2301.03063, an economic and philosophical consideration rather than an empirical study). Inflation is perhaps the very intersection where money, demand, supply, expectations, and institutions meet.

Therefore, the most honest answer to "Why do prices rise?" is not a definitive statement, but multiple perspectives. The maps drawn by various schools of thought point in different directions. Yet none of them shows the whole picture. Moreover, the fact that predictions and policy effects remain uncertain strips away any confidence that inflation has been completely tamed. The next time prices rise, unfolding all four maps together before pointing a finger at a single culprit will get us closer to the truth.

This article is for informational purposes regarding the economy, and most of the cited evidence consists of preprints (arXiv) that have not been peer-reviewed. There is significant uncertainty to use them as a basis for investment or policy decisions, and the details may be updated by subsequent research in the future.

Source Papers (arXiv)

Quantity Theory of Money · Money Supply

• Top distribution of money growth rates affects inflation (quantile sensitivity) (arXiv:2308.05486)
• Weakening inflationary impact of money growth (Ghana) (arXiv:1805.11562)
• Endogeneity of money supply in inflation and interest rate targeting (arXiv:2301.11207)
• U.S. inflation and oil prices as dominant drivers of country-specific inflation (arXiv:2212.05841)

Phillips Curve

• Instability of expectation-augmented Phillips curves across 41 countries (calm and recession periods) (arXiv:2511.22785, arXiv:2511.22786, arXiv:2512.01697)
• Skeleton and shape of the U.S. Phillips curve before and after the pandemic (arXiv:2506.14030)
• India's non-linear Phillips curve (machine learning): expectations are the top driver (arXiv:2504.05350)
• Neural network Phillips curve and deep output gap (explaining the 2021 surge) (arXiv:2202.04146)
• Dual labor markets and the flat Phillips curve (arXiv:2103.06482)
• Beveridge Phillips curve (arXiv:2401.12475)
• Forecasting with the latent variable Phillips curve (arXiv:2601.11601)
• Long-term estimation of Japan's Phillips curve (arXiv:1002.0277)
• Mittag-Leffler Phillips curve fit (arXiv:1604.00369)

Expectations · Attention · Credibility

• Inflation attention threshold (~4%) and supply shock amplification (arXiv:2308.09480)
• Low attention and the monetary policy and attention trap (arXiv:2105.05297)
• News exposure and household inflation expectations (1978–2016, asymmetry since 1983) (arXiv:2009.11557)
• Inflation drivers through the lens of machine learning: the role of expectations (arXiv:2208.14653)
• Large language models' attitudes toward inflation (food sensitivity) (arXiv:2512.14306)
• Minimum credibility is a necessary condition for anchoring, zero credibility causes local instability (New Keynesian model) (arXiv:2012.02662)

Supply Shocks · Pandemic · Energy

• Post-COVID inflation ABM (Mark-0) (arXiv:2306.01284)
• Demand and supply decomposition of inflation risk (energy upside) (arXiv:2403.12456)
• Drivers of U.S. inflation risk during the recent surge (economic activity and commodities, partially mitigated by monetary policy and credit spreads) (arXiv:2405.17237)
• Eurozone: demand explains ≥30% of core inflation rise (arXiv:2505.05536)
• Fed's inflation perspective model (estimation: independent expectation pathway for energy) (arXiv:2006.14110)
• COVID-era inflation weighting bias (Israel) (arXiv:2506.09875)
• Supply-driven inflation and unemployment tail risk (arXiv:2505.05757)

Policy Effects · Forecasting · International Factors

• Meta-analysis of monetary policy effects (overestimation) (arXiv:2509.19591)
• Empirical survey of inflation targeting effects (arXiv:2305.17474)
• Central bank independence and inflation (arXiv:2003.02208)
• Forecasting for monetary policy (Bank of England evaluation) (arXiv:2501.07386)
• Target bias in Bank of Korea inflation forecasts (arXiv:2512.16068)
• International vulnerability of inflation (arXiv:2410.20628)
• Transformer neural network inflation forecasting (arXiv:2303.15364)
• Recurrent neural network (LSTM) inflation forecasting (arXiv:2104.03757)
• Bayesian nonparametric U.S. inflation forecasting (arXiv:2202.13793)

History · Institutions · Other

• Debt-inflation pathway of German (hyper)inflation (arXiv:2405.13296)
• Political stability and inflation (South Asia) (arXiv:2310.08415)
• Inflation and production efficiency (a philosophical inquiry) (arXiv:2301.03063)