Estimating current values of sustainable development goal indicators using an integrated assessment modeling platform: “Nowcasting” with International Futures

1.Introduction

The 17 Sustainable Development Goals (SDGs) with 169 targets and 230 indicators constitute a framework and guide for advancing human development and protecting environmental health. The ability to assess where we are on the path toward the SDGs helps reveal the most glaring global, regional, and national development deficiencies with implications for how policies should prioritize and allocate effort and resources. As countries strive to achieve the goals, it is critically important to understand our current situation. Although the United Nations Department of Economic and Social Affairs maintains a large database across SDGs that draws on many more specialized data sources, data are often spotty and out of date.

Yet, as emphasized also by van Soest et al. [1], most efforts to build a current-status foundation for analysis of potential progress have been limited to selected goals and indicators [2, 3, 4, 5, 6, 7, 8, 9, 10]. Comprehensive human, social, and biophysical development data are often not available and building series across countries, time, and issue areas is complicated by infrequent observations, inconsistency, and relative inattention to some countries. Many series are difficult or costly to obtain. However, such data are important for informing policy priorities and decision-making for development. Modeling tools, including integrated assessment models (IAMs) are increasingly used to forecast alternative scenarios of progress toward the SDGs, and the projects underlying them need to nowcast current data values as a foundation for doing so.

International Futures (IFs) is a long-term, multi-issue model system with variables that extend across the range of the SDGs. IFs models 186 countries annually through 2100 and forecasts more than 700 variables, including approximately 100 related to the SDG indicators across the 17 goals. Such a large and integrated model requires a data processing tool for hole-filling and reconciling of its initial conditions. In IFs, that data tool is called the “preprocessor”. Its function is preparing (nowcasting) the initial conditions for the suite of hard-linked models from data with significant inadequacies. A key value of the IFs preprocessor is its ability for near instantaneous change of the model base year. As new or updated data for any given model base year become available, the IFs model base year can be rebuilt to fill remaining data holes and enhance data consistency. Also, the preprocessor can prepare any historical model base year since 1960, facilitating comparison of historical model runs against subsequent data.

Although IFs is most commonly used as a long-term tool for scenario development and analysis, the preprocessor essentially acts as a distinct nowcasting component within the broader model. The IFs preprocessor provides a uniquely extensive and highly integrated approach to SDG-focused nowcasting. It supports understanding national, regional, and global development patterns.

We proceed by introducing the IFs system. We then elaborate three elements of preprocessor activity required to prepare a complete and internally consistent variable initialization for the integrated models of IFs: hole filling; ensuring consistency within issue areas including population, education, health, agriculture, energy, infrastructure, and governance; and providing coherence across such issue areas. We conclude by demonstrating the nowcasting capabilities of the International Futures system across various SDG target variables.

2.The International Futures model system: Context for the preprocessor

Development of International Futures (IFs) by Barry Hughes began in the late 1970s, when it was primarily an educational tool for helping think about longer-term global futures. Development has been continuous across the intervening period [11]. Early in this century the system began to move beyond academia and be used heavily by and in support of a variety of policy-oriented institutions: international organizations including the European Union [12, 13], the World Bank [14], the African Union [15], and the World Health Organization [16]; governments including the United States [17] and Peru [18, 19]; and nongovernmental organizations including the Institute of Peace [20] and the SENS Research Foundation [21]; as well as corporations and think tanks. While now available under General Public License, continued development is concentrated within a team at the Frederick S. Pardee Center for International Futures at the University of Denver’s Joseph Korbel School of International Studies.

IFs integrates twelve modules within a system that broadly represents human, social, and biophysical development (see Fig. 1). These models include agriculture, demographics, economics, education, energy, environment, gender, governance, health, infrastructure, international politics, and technology. Technology and gender variables exist across these modules. This extensive and highly integrated issue coverage corresponds fairly closely to that of the Sustainable Development Goals, supporting increasingly extensive use in their analysis [6, 7, 10].

Figure 1.

The models of the International Futures (IFs) system. Note: Blue indicates models in IFs primarily focused on human development; green represents socioeconomic development; black shows models especially important to bio-physically sustainable development. Source: Authors.

Development of the preprocessor began near the turn of the century with impetus from heavy use of IFs, the need to update the base year of the system regularly, the need to continually add updates of data from a very large set of sources across the issue areas even for any given base year, and the nearly overwhelming burden of any attempt to create manual nowcasts of all of the model’s initial conditions for any identified base year. Work on the preprocessor began at about the same time as the development of a web-based version of IFs to complement that available for download use (both versions freely available at pardee.du.edu).

In that same period, the sophistication of the graphical user interface (GUI) of the IFs system that integrates access both to the preprocessor and to forecasting increased significantly. In work with IFs, forecasts are understood to be possible futures or scenarios, not predictions. Also understood is that the model proper computes those forecasts with fully integrated, dynamically recursive use of the component models, not through any form of univariate or small variable-set extrapolation.

The GUI provides users four major options and extensive sets of sub-options. Primarily used on the forecasting side, the first option is extensive graphical, tabular, and map-based display of all variables from the Base Case scenario and any other scenario run in the model with variables saved across countries and time. Displays can consist only of forecasts or can combine historical data series with forecasts to help the model user gauge reasonableness of future values. The second main menu option allows the user to create, save, run, and refine scenario interventions consisting of initial condition changes for variables (not often done because the preprocessor provides those), parameters (the primary element shaping scenarios), and selected functional forms (allowing the user some additional control over model structure).

The third major GUI option bridges data/preprocessor and model forecasting capabilities. It provides longitudinal, cross-sectional, and panel access to all data supporting IFs. Further it allows the saving of functions (with flexible functional forms and supporting estimation statistics) from bivariate and multivariate data analysis that can then be used in the preprocessor or the model. One significant advantage of saving multiple functions for the same dependent variable is facilitating analysis across those as the database and even appropriate functional form changes. The final GUI option is Extended Features and contains many data handling sub-options, including calling up the preprocessor to completely (re)compute (nowcast) initial model values for any selected base year.

3.The preprocessor and hole-filling for individual variables

Preprocessing – building the data load for the base year of the model in IFs – consists of multiple processes occurring interactively that this and the next two sections discuss sequentially: hole filling; addressing consistency within issue areas; ensuring consistency across the issue areas. A foundation for the first process of ensuring completeness of country coverage in the model’s base year involves data preparation.

4.Model-specific nowcasting in the IFs preprocessor

The procedures involved in further data processing for each model vary based on the data available for each variable and on our understandings of the accounting frameworks and causal dynamics within each area and across them. The set of 15 preprocessor modules corresponds closely but not completely to the models in IFs. The sequence of their processing owes much to the needs of some later preprocessing modules for the results of earlier ones. The list and dominant sequence are: population, agriculture, energy, education, infrastructure, government consumption, economy, water, sociopolitical, international conflict, state failure, values, health, environment, and technology. Sometimes, a module or a part of it is accessed more than once. The discussion here necessarily simplifies the scope of preprocessor attention to and across these areas before focusing in the next section on integration/reconciliation across them.

As a general rule, while individual variable nowcasting involves statistical estimation, processes across variables within an IFs model requires algorithmic specification. Many of the algorithms involve reconciling incompatible values within accounting systems. Many others reconcile physical and monetary values. Still others involve initializing multivariate indices.

5.Integration and consistency: The social accounting matrix

The preceding section did not discuss two modules in the preprocessor’s initialization of data: economics and government consumption (which we conceptualize as a component of a broader governance module). These systems require bringing together and working across most of the other modules discussed above. For instance, spending on education, health, and infrastructure are part of total government consumption and therefore of government finance more generally. Production, consumption, and trade of energy and agricultural production are important parts of the larger economies, especially in the lower-income countries of most interest with respect to the human and social development variables within the SDGs. In general, analyses of synergies and trade-offs across pursuit of SDGs is rooted in accounting system relationships as well as in understanding causal dynamics (like the importance that access to safe sanitation has for avoiding diarrhea among children and thereby limiting undernutrition). The IFs forecasting system uses both the accounting and causal elements. The preprocessor needs to provide initial conditions for both, and to assure that accounting systems are balanced in the base year.

Figure 2.

Flows represented in a social accounting matrix. Source: Authors.

Balancing accounting system data values in a nowcast across these issue areas requires special attention to two accounting frameworks (elaborated in Hughes [11]). The first involves supply and demand of all goods and services in a general equilibrium framework. In monetary terms IFs represents supply, demand and trade in 6 sectors including food and energy, where values are taken primarily from the respective partial equilibrium models (and data modules) that produce physical values for conversion to monetary terms. In economic modeling this requires attention to accounting fidelity in an extended input-output model like the stylized representation of Table 2.

The second accounting framework encompasses the first and is the broader social accounting matrix (SAM). It represents the pattern of financial flows among households, firms, and government within and across countries, as represented stylistically (and therefore very incompletely) in Fig. 2 (see also Fig. 1 in Li [34]).

Although the values in these two accounting systems do not immediately relate to any of the SDGs, the values in any society directly affect ability to pursue them. Therefore, a full nowcasting foundation for analysis of SDG pursuit benefits from a full country-by-country specification of the extended input-output table in the context of the encompassing SAM. On a global basis such accounting information exists most fully within the database of the Global Trade Analysis Project (GTAP), which draws on a global network of researchers. Even that database has, however, significant limitations with respect to constituting a nowcasting foundation for analysis of SDG pursuit. In particular, it covers only 121 countries and omits many of the least developed countries where that pursuit is most challenging; contributions by researchers to the database are intermittent and sometimes only represent a single year that can be some considerable time back; and the dataset does not include many variables related to government and public finance that are related closely to SDG targets such as spending on education, health, and infrastructure.

To build the SAM the IFs economic preprocessor module thus draws very heavily on GTAP data, and augments it with that from the IMF, World Bank, OECD, and other sources. The preprocessor uses generic IO matrices at different levels of economic development, based on those from GTAP, to fill country data holes (they are also used in dynamic forecasting). The government consumption module reads existing data for spending on the military, education, health, infrastructure, and R&D, as well as data on social welfare and pension transfer payments, reconciling all of these to total government expenditure. The economic preprocessor module uses those values and data on government revenue sources including household, firm, and indirect taxes, along with data or estimates of household sectoral consumption, value added and trade by sector, and much more to create reconciled SAMs for each country.

Among the reconciliation processes in the economic preprocessor are some important internally to countries, including:

Many other processes work across countries:

6.Illustrative results: Nowcast values for SDG target variables

The Sustainable Development Goals cover a broad range of human development and environmental variables. In fact, at least one target variable is available within IFs for each of the 17 goals. The IFs preprocessor and broader model system are especially useful in assessing SDG progress with respect to goals related to human and social system development. That is in part because the quantitative targets for those goals are better defined. The higher-numbered goals include several that have no associated quantification. An example is Goal 12 to “Ensure sustainable consumption and production patterns.” Or quantification is more easily possible, but like Goal 14 to “Conserve and sustainably use the oceans, seas, and marine resources for sustainable development,” the focus in on global public goods and target values for specific countries often do not exist.

Table 3 identifies selected goals and targets for which there are clear numerical target values [35] in general use that can be evaluated at the country level, even when quantification may not have been specified in the original goal/target/indicator statements by the United Nations [36]. It has become very common, for example, to use the World Bank’s [37] quantification of ending or eradicating extreme poverty to mean reducing those suffering from it to fewer than 3% of the population. The IFs project and Table 3 uses that same target level for other goals that specify ending conditions such as hunger and uses the same logic of reaching 97% for goals related to universality such as with secondary education (electricity arbitrarily at 95% because of off-grid preference by some). Table 3 shows values for the same variables (providing associated variable names in IFs) that were included in formulation specifications of Table 2. Table 3 simplifies the official target descriptions. It shows values for the four World Bank country income groupings. To see values for each country, access IFs at pardee.du.edu.

The first line of numerical values in each cell was produced by use of the preprocessor to build values for initializing IFs in 2015, the current base year of the model. Extending the approach to nowcasting of the current year, the second lines show year 2020 values that a base rebuild for 2019 combined with a model run through 2020 produces. The reasons that the year 2020 values do not come directly from the preprocessor but instead from the combined 2019 rebuild and 2-year model run are that (1) very few SDG indicator variable estimates for the year 2020 are yet available in the foundational dataset (in fact, many values for 2019 are obviously nowcast from relatively sparse data) and (2) the year 2020 is especially challenging for nowcasting given the unknown but certainly broad impacts of COVID-19.

Given nowcast values from the preprocessor for the base year, the model can run forward in time fully endogenously. It is also possible, however, to exogenously override selected variables in IFs during the model run. Those include two that discussion of the preprocessor described as “essential series” in the database, namely population and GDP, but also include other important driver variables such as educational attainment of adults and income distribution. For the analysis of COVID impacts in 2020 and subsequent years, it was important to provide exogenous estimates of GDP impact in at least 2020 and 2021. The IFs project has taken evolving IMF [38, 39] estimates from its World Economic Outlook and added them to the extensive IFs database. When the preprocessor and model recognize them to be there beyond the base year, they override otherwise endogenous model calculations. The IFs project presumption is that specialized expertise like that of the IMF should provide sounder values for such variables in the face of systemic shocks than the IFs system would without explicit representation of dynamics generating the growth disruption. The selective turning on of exogenous overrides also facilitates use of the model for a variety of scenario analyses. For instance, the sets of projections of GDP, population, and education for the five Shared Socioeconomic Pathways project [40] are in the IFs data file structure and can be swapped in to override model computations as desired.

Table 3 provides some insights into changes that are occurring with respect to important SDG indicator variables and simultaneously demonstrates the value of the nowcasting. For instance, it shows that, in general, progress with respect to poverty in low-income countries has been less in very recent years than it has been in lower-middle and upper-middle income countries. Internal project analysis (not shown in Table 3) also suggests that even though the GDP growth damage of COVID in low-income countries between 2019 and 2020 may not be greater than that for middle-income countries, their setbacks in progress against poverty will be greater. In contrast, it is the lower-middle-income countries that have suffered most since 2015 with respect to air quality.

Another insight from Table 3 is that low-income country progress on health variables such as maternal mortality has been substantial even over a period of only five years and in the face of economic growth downturns in 2020. Little is more important to societies than health and advances in technology and medical care systems are significant.

The differential progress indicated with respect to key elements of infrastructure, notably the rapid improvements in electricity access on one side, and the slow improvement in access to safe water and sanitation on the other, raises a flag with respect to the formulations within the IFs preprocessor and/or the underlying databases. Extensive nowcasting for very recent and even current years remains a work in progress.

7.Discussion and conclusion

Most integrated assessment models such as IFs have maintained base years much earlier than the current year. One cost of that is producing projected values in early years that can be obviously incorrect. Developments in the IFs preprocessor across the last decade have been making that lag increasingly unnecessary. In fact, as indicated above, given that GDP at MER or GDP per capita at PPP estimates are routinely being made by sources such as the IMF for the current and near future years, it is possible to use the preprocessor even in advance of current year-end.

Such nowcasting is not only important for modeling and forecasting, it is important for understanding the recent rates of progress toward the Sustainable Development Goals, the objectives that have come to dominate national, regional, and global policy pushes toward sustainable development. Further, there can be little doubt that before or soon after the 2030 horizon for the SDGs is reached, successor sets that grow integrally from the current set will continue framing those efforts.

The preprocessor has been a work in progress within the IFs project for two decades. It grew out of the recognition that far too many person-years were being invested in the rebasing of large-scale global models when it was possible to code the algorithms that scientists and their students were using and re-using manually. Automating the tedious process can also reduce errors.

The effort is not complete. Many data series from many sources beyond the “essential series” are desired to feed the preprocessor. Although many semi-automated procedures for pulling those into a common database have been created within IFs, it requires subject knowledge and careful attention to avoid egregious problems that can arise within the original source or data processing (not least issues such as changes in units reported, country names, and even conceptualization and original source data generation processes).

As we have made clear, use of the system for even current year rebasing is now possible, if not always completely satisfactory. The preprocessor will benefit from three efforts going forward: (1) more attention to careful extrapolation across the complete and wide range of data series accessed; (2) automation of the updating via re-estimation of the cross-sectional functions used to fill holes (including addressing how more automatic processes could deal with the sticky issue of best analytical forms of those functions changing across time); and (3) constant attention to improving all algorithms that fill holes, reconcile conflicts, and assure intra-data system accounting consistency. The importance of SDGs nowcasting and forecast-based analysis will drive the effort forward.