- Can Artificial Intelligence Tell Ghana Where Inflation is Heading Before Official Data?
Ghana could soon know where inflation is heading weeks before the official figures are published, after Bank of Ghana researchers developed an artificial intelligence-powered system that tracks hundreds of thousands of online prices and produces near-real-time estimates of changes in the cost of living.
The proposed system, called e-Inflation, uses automated software to collect prices from websites, clean and classify the information, and calculate headline, food and non-food inflation using the same basket and expenditure weights applied to Ghana’s official Consumer Price Index.
Its early results are striking. Between May 2021 and December 2023, the model’s headline inflation estimates recorded a correlation of 0.9941 with the official Ghana Statistical Service series. The average difference was only 0.3684 percentage points.
Food inflation produced a correlation of 0.9836 and an average deviation of 0.6909 percentage points, while non-food inflation achieved a correlation of 0.9944 and an average difference of just 0.1375 percentage points.
In straightforward terms, the experimental system followed Ghana’s official inflation numbers extremely closely, even as the economy moved from relatively stable prices into the inflation crisis of 2022 and then through the disinflation that followed.
That performance raises an important policy question: if technology can show the direction of prices earlier, should the Bank of Ghana, Ministry of Finance and other public institutions continue waiting until the middle of the following month to understand what happened during the preceding reference period?
The answer offered by the working paper is not to discard the existing statistical system. It is to use Big Data and artificial intelligence to provide policymakers with a faster signal while the official process continues to supply the national benchmark.
Prepared by Francis Mawuli Abude, Collins Kwesi Appiah and Julius B. Dasah, the October 2026 Bank of Ghana working paper describes e-Inflation as a timely, cost-efficient and robust framework capable of supporting monetary-policy formulation and broader economic decision-making.
The paper is research in progress, published to encourage discussion, and its conclusions do not necessarily represent the position of the Bank’s Board or Management. That distinction matters because the model remains an experimental policy tool rather than Ghana’s new official measure of inflation.
Why Two Weeks Can Matter
Under the existing system, prices are collected during the first week of the month, but the inflation rate for that reference month is generally released around the middle of the next month. The delay is normal in official statistics because data must be gathered, verified, processed and assessed before publication.
But a lag that is statistically necessary can be economically costly. When the cedi is depreciating sharply, global fuel prices are rising or food-supply disruptions are spreading, conditions may change faster than the official calendar.
A central bank deciding whether to raise, hold or reduce its policy rate needs the most current information available. Businesses negotiating contracts or setting prices face the same problem. By the time conventional data confirm a shift, expectations may already have changed and economic actors may have adjusted their behaviour.
The e-Inflation proposal attempts to close that information gap. Its automated tools can collect price information every week and calculate inflation almost immediately, rather than waiting for enumerators, forms and multiple stages of manual processing.
That does not make official data obsolete. It creates a leading indicator—a high-frequency estimate that can warn policymakers about emerging pressure before the definitive statistics arrive.
How The Digital System Works
At the centre of e-Inflation are software robots programmed to visit selected websites and extract information such as the retailer, product name, identification number, brand, description, package size, regular price and discounted price.
The paper identifies 120 websites and other online sources, including retailers, universities, utility providers, transport organisations and public institutions. These digital sources are treated as the equivalent of market centres within the conventional approach.
The system collects approximately 500,000 price observations, compared with 47,877 observations under the official methodology. It is associated with about 20,000 online outlets or product links, while the conventional CPI draws information from 8,337 physical outlets across 57 market centres.
More information does not automatically mean better information. The value lies in how the data are cleaned, matched and weighted. Online listings can be duplicated, incorrectly labelled, unavailable, discounted temporarily or left unchanged long after the actual price has moved.
To address those risks, the system stores raw information in a Hadoop Distributed File System before using Python or R to clean, format, standardise, validate and, where necessary, limit extreme observations.
A Naïve Bayes machine-learning classifier then sorts products into the appropriate consumption categories. These categories follow the internationally recognised Classification of Individual Consumption According to Purpose used by the GSS.
The architecture is designed as an automated pipeline. Data move from online sources through collection, storage and cleaning to classification, CPI computation, inflation analysis and a report that could be submitted to management and the Monetary Policy Committee.
For non-specialist readers, the process resembles building a digital price enumerator that never needs to travel from market to market. It continuously scans published prices, organises them according to the official consumption basket and compares them with earlier observations.
Keeping The Official Basket
The strength of the experiment is not merely the volume of data. The researchers deliberately retain the same 307 items and expenditure weights used in the official CPI, making it possible to compare the digital results with the GSS benchmark.
Using identical weights is crucial. If the digital model gave greater importance to electronics and less to food simply because more online prices were available for electronics, its inflation rate could look statistically sophisticated while failing to represent how Ghanaian households actually spend their money.
Even so, representation remains the model’s central challenge. A price displayed on an Accra-based online platform may not reflect what a household pays in Bolgatanga, Wa, Hohoe or a rural community where purchases occur mainly in informal markets.
Ghana’s economy remains heavily cash-based and informal despite rapid growth in mobile money, digital payments and e-commerce. Many foodstuffs, local services and household necessities do not have continuously updated internet prices.
The paper acknowledges that prices for public toilet services, resold tap water, charcoal and firewood are difficult to obtain online. Those items must be estimated or collected from physical markets.
This is why e-Inflation should initially complement rather than replace field surveys. A hybrid system could use digital data where coverage is strong while retaining physical collection for informal, rural and locally traded goods and services.
Surviving Ghana’s Inflation Shock
The most persuasive evidence in the paper is the model’s performance across sharply different economic conditions.
Official headline inflation increased from 7.7 per cent in May 2021 to 12.6 per cent in December of that year. It later surged from 37.2 per cent in September 2022 to 54.1 per cent in December 2022, before falling to 23.2 per cent by December 2023.
The e-Inflation estimates moved closely with the official series throughout these changes. That is important because some models perform well when inflation is stable but become unreliable when exchange-rate depreciation, fuel adjustments and supply disruptions produce abrupt price movements.
The results suggest that online prices were responsive enough to capture both the acceleration and subsequent slowdown. For monetary policymakers, direction can be as important as the precise rate. Recognising that inflation has begun to turn may affect the timing and communication of interest-rate decisions.
However, the reported accuracy should be interpreted carefully. A near-perfect correlation shows that the two series generally moved together; it does not prove that their levels were identical or that the digital model will perform equally well in the future.
Inflation series that share the same basket and weights and experience the same national shock are expected to trend together. Correlation alone cannot determine whether e-Inflation consistently anticipates turning points, performs equally across all divisions or remains stable when websites change their structures.
The average deviations are therefore important, but additional testing would strengthen the case. Policymakers would benefit from measures of absolute error, root-mean-square error, bias, performance during turning points and results from periods not used to develop the system.
Independent validation by the GSS, academic researchers or another statistical institution would also increase confidence and reduce the risk that a promising internal experiment is adopted before its weaknesses are fully understood.
What It Could Change at the Bank of Ghana
For the Bank of Ghana, the immediate value of e-Inflation lies in monetary-policy surveillance. The Monetary Policy Committee does not make decisions from one indicator, but inflation is central to its mandate of maintaining price stability.
A weekly estimate could reveal whether food pressure is broadening, whether transport costs are passing through to other prices or whether a period of cedi appreciation is beginning to reduce imported inflation.
That information could improve staff forecasts and allow the MPC to test different scenarios before official numbers are released. It could also help the Bank explain why it is tightening, holding or easing policy when businesses and households perceive conditions differently from the most recently published data.
There is nevertheless a communication risk. Publishing several inflation estimates could confuse the public, particularly if the e-Inflation reading diverges from the official rate. Ghana would need a clear hierarchy: the GSS CPI remains the official statistic, while e-Inflation is identified as an experimental nowcast or early-warning indicator.
Recommendations From the Researchers
The authors recommend that central banks, national statistical offices and research institutions adopt the framework for real-time inflation-related decision-making because it has shown itself to be timely, accurate, robust and cost-efficient.
They call on the International Monetary Fund to provide technical assistance and training to ministries, central banks and statistical agencies in developing countries. Such support should cover artificial intelligence, data science and the predictive techniques needed to implement similar projects.
Development partners interested in strengthening Africa’s statistical capacity and digital systems are urged to provide resources for adoption across countries participating in the Strategy for the Harmonisation of Statistics in Africa.
The paper also places the proposal within West Africa’s long-running monetary-integration agenda. It recommends collaboration among the West African Monetary Institute, West African Monetary Agency, West African Institute for Financial and Economic Management and ECOWAS Commission to replicate the methodology across member states.
The researchers further recommend that central banks and statistical offices build expertise in data science, artificial intelligence and Big Data analytics rather than relying entirely on external consultants.
That institutional point is fundamental. An inflation system is national infrastructure. The algorithms must be explainable, the data must be governed securely and the process must remain functional when commercial software changes or technical staff leave.
Governments are also encouraged to accelerate digital transformation, strengthen internet infrastructure and promote e-commerce, digital payments and wider participation in the digital economy. The more representative online economic activity becomes, the more reliable digital price measurement can be.
The Partnerships Ghana Will Need
The paper identifies outdated online prices and restrictions on web scraping as significant operational risks. A retailer may update its physical shelf price without changing its website, while another may block automated access for cybersecurity or commercial reasons.
The proposed response is greater collaboration with e-commerce companies and institutions that publish prices. Instead of relying solely on robots reading public webpages, the GSS and Bank of Ghana could negotiate secure data-sharing arrangements or application programming interfaces.
Such partnerships could provide cleaner and more current information, but they would need firm governance rules. Commercial businesses should not be able to influence the inflation measure by choosing which prices to submit, while public institutions must protect confidential data and prevent unauthorised use.
Regular comparisons between online and physical prices would remain essential. If a supermarket advertises a promotional price available only to selected customers, or delivery costs materially change the final amount paid, the model must determine which price represents household consumption.
A Practical Path Towards Adoption
Ghana should resist two extremes: dismissing the model because it is experimental, or treating its impressive initial correlations as sufficient reason to replace the official CPI.
A more credible path would begin with a formal pilot jointly governed by the Bank of Ghana and Ghana Statistical Service. The institutions could publish an experimental weekly index internally, compare it with field data and document its performance over several additional years.
The pilot should report results by region and consumption division, examine rural and urban gaps, and test the model during exchange-rate shocks, harvest cycles, utility adjustments and changes in petroleum prices.
An independent technical committee could review the data sources, algorithms, security arrangements and statistical methods. Selected code and methodological documentation should be published where possible to allow expert scrutiny without exposing sensitive systems.
The e-Inflation series could then enter the Bank’s internal forecasting framework as one of several indicators. If its reliability persists, a public experimental release could follow, clearly labelled and accompanied by confidence ranges and explanations of revisions.
Only after sustained validation should Ghana consider giving digital data a formal role in producing the official CPI. Even then, the likely destination is not a completely online index but a blended system combining automated collection with targeted fieldwork.
Beyond Inflation
The paper proposes extending Big Data nowcasting to gross domestic product and unemployment, two indicators that also arrive with delays and are difficult to measure in economies with large informal sectors.
It also recommends using machine-learning algorithms to move beyond nowcasting towards forecasting price movements over the medium and long term.
That ambition could strengthen economic management, but prediction is more difficult than measurement. A system may accurately describe current online prices without correctly anticipating future exchange rates, harvests, taxes, fuel costs or global commodity shocks.
Machine learning should therefore enhance rather than replace economic reasoning. Algorithms can detect patterns across enormous datasets, while economists assess causation, policy transmission and events that have little historical precedent.
A Modern Tool, Not A Statistical Shortcut
The e-Inflation study offers Ghana an important glimpse of how artificial intelligence can improve public policy without turning policymaking over to machines.
Its central achievement is practical: it shows that prices collected digitally can reproduce official inflation trends with remarkable accuracy and at greater speed. In a country where inflation shocks can rapidly weaken incomes, savings and business confidence, earlier evidence has genuine economic value.
But speed must not come at the expense of representation, transparency or public trust. The official CPI is more than a calculation. It is a national reference used in contracts, policy decisions and public debate. Any technology incorporated into it must withstand scrutiny across regions, income groups and market structures.
The most sensible conclusion is therefore neither replacement nor rejection. Ghana should preserve the GSS CPI as the official benchmark while developing e-Inflation into a high-frequency companion capable of alerting policymakers to changing conditions.
If properly validated, governed and communicated, the system could give Ghana something its economic managers have long needed: not a different inflation reality, but a clearer view of that reality before it becomes old news.
