Coriolis Technologies provides datasets of global trade flows covering the time period 1996 to 2020, with annual and monthly data from January 1996 to the current month. The data cover import and export flows for 200 countries by partner, sector and partner-sector trade routes. The goods dataset uses OECD mirroring techniques to ensure that bilateral trade flows are identical. For example, exports of oil from Saudi Arabia to Germany will have the same value as imports of oil into Germany from Saudi Arabia.
Data refinement methodology
Coriolis Technologies uses the UN Comtrade statistics as the main source of comprehensive and detailed information on international trade. The UN Comtrade’s database provides information as it is reported by country authorities and does not apply any data integrity verification or missed data filling. Consequently, the raw UN Comtrade dataset presents the following relevant issues:
- Absent or sparsely reported data for many countries and territories, especially in Africa, Asia and Latin America Large asymmetries in bilateral trade values reported by a pair of countries
- Partly disclosed information due to statistical confidentiality (relevant especially to the arms trade and sanctions regimes)
- Classification reporting and conversion issues (UN Comtrade does the conversion between different classifications, based on the most detailed layer of the classification, while some information can only be reported on an aggregated level)
- Delayed reporting and past information amendments by some countries
In order to build consistent international trade data and increase its coverage on the bilateral level to the extent possible, Coriolis Technologies shares the trade mirroring approach used, in particular, by the OECD.
This approach uses the information reported by partner countries to resemble and verify the reporter’s data as well as to fix the asymmetries. Mirror statistics are mostly helpful for small and medium-sized trading countries, where there is no reported data available or the data is very sparse. However, in many cases they help to reveal some systematic trade underestimation for some well-established reporters, or information which was not disclosed.
Coriolis Technologies implemented the trade mirroring algorithm for the UN Comtrade Harmonised System (HS) classification, revision 1996, as follows:
- Trade mirroring applied for each trade flows between all the reporting countries
- Simple trade mirroring is done for cases when the trade is reported only from one of the sides.
- When a trade is reported by both sides and there is an asymmetry, the common trade value is calculated as a weighted average of the two reported values. The weights are defined in accordance with the country reporting reliability index, developed by Coriolis Technologies.
- If the asymmetry is extremely high, the lower value is assumed to be underreported, and the resulting trade value is set to the higher declared figure.
- The resulting trade values on 2 and 4-digit levels, calculated for the HS 1996 classification are compared with those for HS classification used when reporting (normally country statistic offices report in the most recent classification for the year, and this information is converted to previous versions of HS classifications with certain losses, related to detailed data disclosure). If the difference between the trade value in HS 1996 and HS ‘as reported’ is higher than 40%, the former is assumed to be wrong and the HS ‘as reported’ value is taken as the best estimate.
- When the whole dataset is processed for a single year, country total imports and exports are calculated for the refined trade values for each commodity group, and the total cumulative import and export values are obtained for each country and
The resulting ‘refined’ dataset appears to be the most complete of the available source data, and features the following attributes:
- Provides best available estimates of aggregated trade statistics, where they were not reported or significantly distorted
- Gives an indication of hidden trade flows and the true state of trade relationships, when the ‘refined’ values diverge significantly from the ‘raw’ figures. The amount of hidden trade can be estimated as the difference between the two on a commodity or aggregate levels.
- Is import/export consistent: each country import value is equal to the counterparty’s export value for every commodity group/classification code, and the global import total converges with the export total.
Trade forecast methodology
Coriolis Technologies employs a momentum-based forecasting algorithm in order to produce trade value forecasts. The method below is generic as applied to a longer time period. However, for this report we have limited our forecasts to the end of 2017.
The forecasting method features the following concepts:
- Analyses dynamics of differentiated time series with 10-year historical horizon, applies outliers filtering and absent data filling.
- A double momentum forecast: the first taking a 10-year moving window and the second a three-year moving window. The former provides long-term stability to the five, 10 and 15-year forecast, while the later impacts the one to three-year dynamics of the forecasted time-series.
- A constant growth rate of 1.0% (continuous compounding) is applied to extremely volatile and/or highly fragmented time series.
- Shares the Gaussian approach in order to provide long-term forecast, given the growing degree of uncertainty.
The forecasting algorithm, if extreme volatility or incomplete data identified, takes the average of the last three years (2012-2014) as the starting level, and then calculates the forecast based on the 1% annual growth rate on the logarithmic scale (1% log return).
This is based on the assumption that if the data is not reliable, the last three-year estimate is a far more legitimate predictor of the 2015 ground level.
The model does not rely on any assumptions about the economic cycle or supply chains and is built purely on the drivers of trade in the data (identified through Principle Component Analysis) and the data’s momentum. This approach is chosen in preference to bilateral trade modelling such as gravity modelling or partial equilibria for two reasons: first, it does not require normative assumptions about trade costs or geography and second, it does not require assumptions about GDP growth to be made. The relative simplicity of the model provides an important balance between flexibility and reliability of the forecasting algorithm, especially in the short term.take me back