FORECASTING INFLATION EXPECTATIONS THROUGH BIG DATA ANALYSIS

In today's macroeconomic reality, inflation is not merely a price growth index, but the result of a complex redistribution of resources. Since the final abandonment of the gold standard in 1971, the global financial system has shifted to the use of fiat (decree) money, the stability of which relies exclusively on trust in the regulator. In a fractional-reserve banking system, the creation of the money supply occurs through credit expansion mechanisms, which forms a so-called "inflationary overhang"—an excess of liquidity that threatens the stability of the consumer market. [3, p. 9, 62]

Inflation has a pronounced redistributive character: the first to receive new money is a privileged "club" (large banks and state corporations) that has the opportunity to purchase assets at old prices, while the wave of emission reaches ordinary consumers in the form of already devalued currency. In such an environment, traditional methods of measuring inflation based on historical data begin to lag catastrophically, turning forecasting into the "art of managing expectations". [3, p. 39]

Classic inflation monitoring via the Consumer Price Index (CPI) suffers from significant time lags and the problem of "vintage" data. Official figures are published with a delay of several weeks, which, during periods of market volatility, makes them useless for operational decision-making. Research shows that using final revised data instead of preliminary "vintage" data leads to an artificial underestimation of forecast error by 8–17%. [4, p. 5, 14]

Big Data technologies allow these barriers to be overcome through:

• Web scraping: Daily collection of millions of prices from retailer websites in real-time. [4, p. 11]
• Vintage data analysis: Accounting for the actual moment of statistics release, which allows for the construction of more honest predictive models. [4, p. 10]
• Regional detail: The ability to account for the specifics of local markets, such as the Siberian macro-region, where standard models often fail. [2, p. 5]

The application of Machine Learning (ML) algorithms has become a standard for the Bank of Russia. Unlike rigid econometric models, ML methods such as Random Forest and Gradient Boosting are capable of effectively processing non-linear dependencies across hundreds of variables—from oil prices to real wage dynamics.

Table 1.

The comparative effectiveness of models in forecasting the Russian CPI

Gradient boosting demonstrates the best adaptability to the specifics of Russian data, especially when accounting for real delays in the publication of statistics. Furthermore, on forecasting horizons of more than a year, ML methods show stable superiority over traditional ARIMA models, better capturing structural shifts in the economy. [2, p. 12]

Conclusion

The integration of Big Data and machine learning into the inflation analysis process represents a qualitative transition from expert estimates to science based on high-frequency data. The conducted analysis allows for several fundamental conclusions regarding the future of macroeconomic forecasting.

First, the use of alternative data allows for solving the problem of "unanchoring" inflation expectations. In Russia, public expectations are extremely inertial and often do not align with the Bank of Russia's target indicators (4%). Big Data analysis enables the regulator to identify shifts in consumer sentiment even before they transform into actual price growth. This is critically important for preemptive intervention and maintaining trust in the national currency. Without such trust, as historical experience shows, inflation turns into a "vicious monster" that destroys any long-term financial plans. [3, p. 4]

Second, Big Data technologies allow for "surgical" adjustment of monetary policy. The traditional CPI is often viewed as the "average temperature in a hospital," a figure that frequently hides deep disproportions between sectors. Real-time analysis of millions of transactions makes it possible to identify the specific product categories where excess demand is forming. This allows for targeted responses that do not suppress overall economic activity with high interest rates. [3, p. 17]

Third, the expanded use of ML models in central banking practice contributes to increased economic transparency. The ability of gradient boosting and neural network models to account for data "vintage" and real publication delays makes forecasts more honest and reproducible. This reduces the risks of making erroneous decisions based on preliminary, incomplete statistical series. [4, p. 15]

Ultimately, fighting inflation in the conditions of a modern fiat monetary system is impossible through administrative methods alone. An accurate, scientifically grounded monitoring tool provided by big data analysis technologies is necessary. The future of macroeconomics lies at the intersection of classical money theory and advanced information processing methods, allowing for the anticipation of structural shifts before they become facts in statistical reporting. Only such synergy will allow for minimizing the redistributive costs of inflation and ensuring sustainable economic growth.