Free digital data may add trillions to China’s GDP, model estimates

A new economic model argues that China’s official gross domestic product leaves out a category of output that no statistical ledger has ever attempted to record: the data that consumers hand over to firms as a by-product of every digital transaction, at a zero monetary price. Applying a recursive data-economy framework to a bottom-up measure of digital investment built from the asset disclosures of up to 5,120 Chinese listed firms, the study estimates that counting this bartered data would have raised measured Chinese GDP in 2024 by 1.60 percent, or 2.16 trillion RMB, under a broad measurement boundary that treats telecommunications network assets as data infrastructure. Under a narrower boundary that excludes those assets, the adjustment falls to 0.60 percent, or 0.82 trillion RMB.

The authors are careful about what these numbers are. The two figures are alternative measurement boundaries, not statistical bounds, because the model does not identify a unique asset perimeter. On the listed-firm universe alone, before any extrapolation to unlisted companies, the broad-boundary figure is 0.91 percent of GDP. Roughly two-fifths of the headline adjustment is therefore extrapolation resting on an assumption about the digital intensity of unlisted firms that the authors cannot verify. The broad-boundary central case also carries a calibrated scenario range of 0.77 to 4.07 percent of GDP for 2024, which reflects alternative calibration assumptions rather than sampling uncertainty. Every figure is conditional on a normalization, denoted chi, that maps measured digital investment onto the value of usable data; the headline results impose chi equal to one, a calibration choice for which no Chinese evidence exists.

The mechanism behind the exercise comes from the data-economy framework of Farboodi and Veldkamp, in which data is a state variable that firms accumulate to forecast a moving economic environment. Transactions generate data as a free by-product; firms use it to sharpen predictions and improve product quality; and because national accounts capitalize nothing that changes hands without a price, this entire flow sits outside the ledger. The direction of the omission follows from the framework rather than from independent evidence. The paper prices one thing only, the bartered data flow, and explicitly does not assess the accuracy of what the accounts do record or estimate the net bias in Chinese GDP, a contested question with a substantial forensic literature of its own.

Converting one year’s data flow into an asset value requires an amplification factor, which the paper derives in closed form. If a year’s data raises the firm’s knowledge stock and that increment depreciates geometrically, the present value of the resulting profit stream relative to its first-year value is the reciprocal of one minus the discount factor times one minus the depreciation rate. The depreciation rate itself is derived from Bayes’ law rather than assumed from accounting convention: because firms forecast an AR(1) state that drifts, yesterday’s precision is worth less today, and the rate varies with the size of the knowledge stock. Under a conventional fifteen-year amortization schedule the amplification factor would be 6.0; under the model-implied rate it is 2.8. Applying the accounting convention would therefore overstate the value of Chinese data assets by roughly a factor of two.

The empirical scaffolding is built from Chinese listed-firm disclosures. Because no Chinese statistical agency publishes an analogue of the U.S. Bureau of Economic Analysis estimate of own-account data investment, the authors classify itemized fixed-asset and intangible-asset line items by keyword matching, isolating digital hardware and digital intangibles from 289,505 fixed-asset rows and 200,020 intangible rows. Two corrections proved essential. The disclosed figures are period-end balances, stocks rather than flows, and treating them as investment would overstate the key moment by a factor of four to five; the authors convert stock to flow using an accounting depreciation rate varied between 0.15 and 0.33. Separately, aggregating interim filings and counting the accumulated stock of newly listed firms as investment had manufactured a fake pandemic-era boom, showing the wedge peaking at 3.3 percent of GDP in 2020; the corrected series shows 0.9 percent. The resulting investment flow averages 0.219 percent of GDP over 2008 to 2024.

A validation audit of the keyword classifier uncovered a measurement defect the authors say is not specific to China. The audit hand-coded every distinct asset description with material frequency or value, a census rather than a sample. The classifier proved highly specific, rarely flagging non-digital items, and count-weighted recall was near perfect at 0.998. Yet on value it captured only 0.208 of coded digital fixed assets. The entire shortfall traced to a single cause: Chinese has two orthographic variants for the word communication, and the keyword lists in circulation contained one but not the other, so communication equipment and telecommunications lines were never flagged. A classifier scoring an F1 of 0.945 on counts was thus missing four-fifths of the value at stake, a hazard that count-weighted validation alone would have concealed. The authors include the missed network equipment, which at 1,272.6 billion RMB of 2024 stock is roughly twice the corrected total for everything else and is concentrated in the three major telecommunications carriers, which hold 71 percent of broad-boundary digital capital.

In levels, the estimate implies that measured Chinese nominal GDP of 134.8 trillion RMB in 2024 would rise to 137.0 trillion once the data consumers transferred to firms is counted, with roughly 18 trillion RMB accumulated since 2008. Averaged over 2008 to 2024, the broad-boundary central case is 1.53 percent of GDP. The wedge is large in absolute terms while remaining a small share of a fast-growing economy. The year-to-year path is treated cautiously: differencing a stock amplifies measurement error in any single year, and the coverage adjustment moves mechanically with the listed share of GDP, so the level is the more informative object.

Several quantities the model requires cannot be identified from Chinese data, and the paper is unusually explicit about the consequences. The knowledge stock at which firms operate is not identified without a forecast-error series of the kind constructed for the United States from analyst guidance, and no adequate Chinese counterpart exists. The persistence of Chinese productivity is estimated at 0.786 after detrending, but across alternative filters and sample windows the estimate ranges from 0.27 to 0.93, a range the authors carry through to the headline. The comparison with the United States, whose own-account data investment flow was 0.25 percent of GDP against China’s 0.32 percent, is likewise not identified: one U.S. persistence estimate puts American amplification above China’s, another puts it below, and the ranking flips with the choice of filter. The paper reports no cross-country ranking.

Two independent checks corroborate the accumulation the model describes. A text-based measure of data-factor intensity in annual reports rises 3.4-fold between 2010 and 2023 within a balanced panel of firms, confirming genuine within-firm intensification rather than a composition effect from new listings, and it correlates with the independently constructed digital investment series at 0.97 in aggregate. The 2024 accounting reform, which for the first time allows firms to recognize data resources on the balance sheet, produced disclosures from 109 firms totalling roughly 2.08 billion RMB, three orders of magnitude below the model’s estimate. The authors explain the gap: historical-cost recognition captures data firms bought or deliberately built, not the barter flow generated as a by-product of transactions, which has no separable acquisition cost. The two measures are complements rather than rivals.

The authors frame the contribution as bounded. The estimates quantify one positive omission channel, not the net bias in official Chinese GDP, and positivity itself follows mechanically from positive measured investment and an amplification factor above one rather than from an empirical finding about sign. What the analysis speaks to is magnitude and boundary. The dominant uncertainties, the mapping chi, the coverage assumption and the knowledge-stock multiple, are not identified by Chinese data at all and are set by normalization or analogy, so the uncertainty that dominates the exercise is not sampling error in the firm disclosures but the absence of data on quantities the model requires. The paper’s proposed remedy is a credible Chinese forecast-error series, which would replace the range over the knowledge stock with a point estimate and settle the cross-country comparison in one step.

Subject of Research: Model-based estimation of the unrecorded value of bartered consumer data in China's GDP using listed-firm digital investment data

Article Title: Is China’s GDP undervalued? A digital economy model

Article References: Wang, W. (2026). Is China’s GDP undervalued? A digital economy model. Digital Economy and Sustainable Development, 4(1), Article 14. https://doi.org/10.1007/s44265-026-00087-8

Image Credits: AI Generated

DOI: 10.1007/s44265-026-00087-8

Keywords: China, GDP, digital economy, data as an asset, national accounts, data barter, economic modelling, measurement, listed firms, data depreciation, mis-measurement, calibration