This systematic review design follows PRISMA 2020 guidelines ([30]; [51]).

Due to the hyper-fragility of the Burundian economy, we allowed both quantitative and qualitative empirical research, as is standard for meta-analyses of Sub-Saharan African literature ([4]; [6]; [9]). Our a priori protocol focused on the review question and the narrow definition of “determinant” articulated in Section 1.

Our database search took place over six months, June - December 2025, utilizing the following platforms: Web of Science, Springer Link, Wiley Online Library, Taylor & Francis Online and Google Scholar. The literature search publication window was from January 2005 to December 2024, to cover the post-conflict (2005-2014), crisis (2015), and post-crisis (2015-2024, including the post-COVID-19 period) phases. Contrary to an earlier mistaken statement, we were not focused on “a targeted 2020-2025 corpus”. The period we searched was January 2005 - December 2024 (the emphasis on the 2020-2024 corpus was a priority in our analysis but not an eligibility constraint), and no paper was included or excluded on the basis of year alone.

Below we set out, by database, date, and search string, the parameters used and the number of items we retrieved in that database. We found a total of 113 items, and after we eliminated duplicates, we screened 88; of the full texts, we examined 28 and included 12 in the initial corpus (Table 1).

Table 1.Reproducibility supplement: source-specific search parameters.

Note: The records were imported into a reference manager and duplicate records were removed via a two-step process. After an automatic de-duplication performed using the DOI and title information, near-duplicate records were checked manually and confirmed by two reviewers.

Specifies information as to how relevant literature was searched for: lists the five major academic databases which were searched and the Boolean syntax used, as well as date coverage of 2005-2024 used to identify a comprehensive number of studies (Table 1).

The selected studies must include at least one time-series, panel, or mixed methodological specification based on peer-reviewed empirical published literature ([7]; [50]). Detailed inclusion criteria and exclusion criteria are given in Table 2. A total of two variations are considered in this empirical specification of variables. Firstly, a study is included in the synthesis corpus only if the outcome measure is real GDP or real GDP growth or a directly estimated determinant of GDP growth; other development outcome measures such as food security, aid effectiveness, and financial-sector assessment are carried over to the contextual background and not included in the corpus counts.

Secondly, a falsifiable rule is employed concerning regional studies: cross-country studies qualify as primary evidence only if a) Burundi comprises not less than 10% of the observations in the relevant dataset (panel-based), and b) a Burundi-specific result (country dummy, country subsample, or time series estimate) is reported.

Studies that meet a) but not b) are to remain as comparative background and referenced as such throughout the synthesis process. Studies that satisfy neither a) nor b) are to be excluded.

Table 2. Inclusion and exclusion criteria.

Extracted data included bibliographic information, study design, time period, variables, estimation method, and the magnitude and statistical significance of effects, as indicated in the source. Data extraction was performed independently by two researchers, with any disparities settled through consensus. The quality appraisal was carried out across four weighted dimensions adapted from [38]: methodological rigour (33.3%), data quality (25%), analytical rigour (25%), and policy relevance (16.7%).

The current iteration provides a complete explication of the scoring process.

Each dimension is assessed using a 1 to 4 anchor scale: 1) represents significant concerns (e.g., inappropriate or undocumented design); 2) moderate concerns (i.e., appropriate design but limited diagnostics); 3) minor concerns (i.e., appropriate design with routine diagnostics); and 4) strong (i.e., appropriate design with a comprehensive set of diagnostics and robustness checks, and translucent data). The comprehensive quality score is the sum of these weighted scores, ranging from 1 to 4. Studies with an index lower than 1.75 (average score below “moderate”) are omitted from primary synthesis.

Sensitivity of primary results was assessed by reapplying the qualitative synthesis to data restricted solely to studies achieving a score of 3.

High inter-rater reliability was confirmed with an intra-class correlation coefficient (ICC) of 0.89.

Due to variation in methods, evidence synthesis was executed as a narrative rather than formal meta-analysis. The synthesis presents evidence in two streams: i) Burundi-specific studies, comprising the key findings for the review question, and ii) Regional studies, which offer contextual and comparative evidence only. For studies that do not utilize designs permitting causal inference (e.g., confirmed Granger-causality), findings will be reported as associations and correlations rather than causal effects. Unless effects were quantified by the source study, any claims about their magnitude will reflect stated directional and statistical significance.

A systematic search generated 113 records; this left 88 records after duplicate removal; 60 records were excluded at title or abstract screening (38 not being Burundi-specific, 15 non-empirical, and 7 outside the date limit); 28 records were read in their entirety, of which 11 were rejected (8 not specific enough on GDP/growth outcomes and 3 being duplicates), making it so that the body corpus is composed of 16 articles, all assessed under the GDP criterion.5 Studies published from 2020-2024 are the most frequent, due to the current post-pandemic and post-2015 research interests in the role of mobile money and digitization (these themes are considered as part of the context when not used as dependent variables, cf. Table 2). The included empirical studies are summarized in Table 3.

Table 3.Included empirical studies in the review.

Continued

Note: The quality index = weighted average of the 4 dimensions of assessment (anchors of Table 4). All indices are above 1.75-the threshold for inclusion-and therefore, no paper was discarded due to quality issues. All the regional papers that did not have a confirmed Burundi-based estimation are designated “contextual” and will not be used to answer the review question.

Screening and selection were in accordance with the standard PRISMA process. All identified articles were collected, duplicates were removed, and finally, records were screened for eligibility at the title and abstract level by two independent reviewers. Full texts of potentially suitable studies were obtained and evaluated against inclusion criteria. Any discrepancies were resolved through discussion and/or referral to the third reviewer. This was documented on the PRISMA flow diagram (Figure 1).

Table 4 reports the per-study appraisal scores and overall quality index. The corpus is of moderate-to-good quality (mean index 2.83, range 2.4 - 3.4). No Burundi-specific study exceeded 3.4, reflecting limited robustness testing in single-country work; the sensitivity check using only studies scoring ≥ 3 did not alter any qualitative conclusion of the synthesis, indicating that the findings are not driven by lower-quality studies.

Figure 1.PRISMA flow diagram of study selection.

Table 4.Per-study quality appraisal (1 - 4 anchor scale; weights: rigor 33.3%, data 25%, depth 25%, and policy 16.7%).

Inflation is the single most ubiquitous macroeconomic predictor of growth in the Burundi-specific literature, with [17] reporting a significant negative effect of consumer-price variability on GDP growth in Burundi. In the regional literature, a non-linear threshold is reported whereby moderate inflation (~5% - 7%) can coexist with growth, but inflation above 10% is damaging ([9]; [4]). Qualifier: This is based on studies from Tunisia and other non-Burundian countries, and has not been estimated for Burundi; the 10% figure should therefore be interpreted as an outside benchmark rather than a Burundi estimate. Because a fragile country like Burundi may be vulnerable even to moderate shocks, the true threshold is likely to be lower, an issue for future research.

Political fragility, conflict, and political instability are the most consistent anti-growth contributors in the Burundi literature: [45] emphasize conflict and instability as the key causes of Burundi’s sluggish long-term growth, while [38] quantify a loss of around 4.7 percentage points of GDP per annum, through deterioration of the rule of law and reallocation of expenditure to security. These are correlations/estimates from a particular econometric technique, not causal claims: the only definite link of causality between FDI, trade and growth in the Burundi-specific evidence ([37]).

In Burundi, export/import contribution to growth is weak ([11]), and dependency on imports neutralizes potential gains; the long-run association (trade openness-growth) is positive ([43]), as well as positive for exports and negative for imports ([62]). Regional contextual studies suggest more uniformly positive FDI-growth and openness-growth relationships ([3]; [23]; [29]), with [7] showing that trade moderates FDI-growth. The difference is explained by the mediating effect of institutions: in fragile countries like Burundi, FDI and openness fail to translate into measured growth (GDP) in the country. Because the regional estimates do not contain a verified Burundi element (Table 3), they are used here as contextual background to the review, not directly to answer the research question.

There is no evidence that digitalization and financial inclusion predict growth in Burundi, while [44] report a negative association between unemployment and growth in Burundi. This is consistent with what is reported regionally (Okun’s-law-type relationships: [18]; [15]) but is muted, perhaps because underemployment in the large informal sector biases estimated unemployment figures.

The synthesis identified three research gaps. i) Post-2019/2024 caveat: no eligible studies cover the post-COVID-19 recovery phase. ii) Priority methods: Burundi-specific research continues to rest on time series and traditional ARDL approaches; research based on threshold and non-linear dynamics, and the new generation of analysis tools (Markov-switching, PVAR, non-linear ARDL), has yet to be applied at the national level. iii) Priority sectors or areas: informal sector, environment and climate vulnerability, and effects of digitalization on growth are not estimated for Burundi.

Figure 2. Methodology and key variables analysis.

Looking at the distribution of variables according to the Data Extraction Framework (Figure 2), we can see the emphasis placed on macro and external determinants. The most commonly studied variables across the 17 studies are the ones of FDI and GDP Growth and Trade (Exports/Imports). Other variables like Conflict and Inflation are only explicitly featured in a reduced number of studies that were written about the “specialized” literature as presented in Figure 3.

The studies included in the review were then weighted based on the following criteria: Methodological Rigor (33.3%), Data Quality (25%), Depth of Analysis (25%) and Policy Relevance (16.7%) to confirm the conclusions presented are derived from high-quality evidence. The 4 dimensions of the framework used to assess the methodological rigor of the studies included in the review are described in Table 5.

Figure 3.Frequency analysis and methodological gaps.

Table 5. The 4 dimensions of the framework used to assess the methodological rigor of the studies included in the review.

Source: Framework adapted from [38].

Shown along with the results is Figure 2, a graphic of the distribution of techniques applied within the literature corpus. As is visible from the graph, the body of literature is highly focused on traditional time series or panel regressions, with only a small fraction (16.7%), employing non-linear or more complex techniques. The quality assessment framework and weighting criteria are presented in Figure 4.

Figure 4. Quality assessment framework and weighting criteria.

The findings suggest that macroeconomic, political and structural factors are significant drivers of GDP in Burundi. Macroeconomic Stability and Inflation: Inflation proved to be highly significant with respect to GDP growth ([17]), and it has been suggested to have a non-linear dynamic threshold relationship where growth becomes negative beyond an inflation threshold of 10%, depleting purchasing power and disrupting market mechanisms ([9]; [17]).

Political Fragility and Conflict: The principal finding is that the political factor shows the most significant negative impact on GDP growth ([38]); the quantitative analysis calculated the loss to the Burundian economy caused by the civil conflict as having an annual GDP growth rate drop of 4.7%.

Drivers from the External Sector: The analysis suggested that FDI and trade openness promote growth in an environment of institutional stability ([3]; [22]); however, contrasting predictions were derived when [11] determined that FDI levels in Burundi are often insufficient to offset the detrimental impact of import reliance. Labor Markets and Digitalization: The findings showed that unemployment negatively correlated with economic performance ([44]) and that digital financial services, such as mobile money, were found to encourage entrepreneurship and resource mobilization, despite lacking much empirical evidence from Burundi ([8]).

As seen across the time span of the study, there has been a decline in the availability of recent peer-reviewed empirical research in recent years. A major research gap exists across the beginning of the 2019-2024 time frame, as no peer-reviewed empirical study met the inclusion criteria in 2019 specifically. With an increase in available research for the period 2020-2024, there is a need for more high-frequency data for the overall picture of post-COVID-19 recovery beyond the initial work of [8]. Important research gaps remain concerning the impact of the informal sector, environmental sustainability, and the long-term effect of digitalization on economic growth in Burundi.

The synthesis comprised 8 Burundi-specific and 8 regional-contextual empirical articles on the determinants of GDP and growth in Burundi (2005-2024). Consistently, Burundi-specific evidence points to inflation, conflict, and unemployment as detrimental to GDP growth, with only one quantified estimate of a growth loss (civil conflict at 4.7pp annually; [38]). The impact of FDI and trade openness on growth is also sensitive to institutional stability, whereas no country-specific evidence is available in the case of Burundi for estimating the growth impact of digitalization, financial inclusion, the informal sector, or climate vulnerability. Despite aligning with governance-growth relationships for Sub-Saharan Africa (e.g., [7]; [3]; [20]), our study indicates that Sub-Saharan African averages may mask such a conditional relationship, which can be statistically negligible or even negatively associated with country growth.

In line with the literature, the effects of inflation and unemployment are negative ([17]; [9]; [15]; [44]; [18]), while the effects of “pulling” FDI and openness are positive ([3]; [22]; [29]). Divergences relate to the external sector; most regional panel studies support a positive FDI-growth association ([3]; [22]; [29]), in contrast to Burundi-specific evidence, which is scattered ([11]; [43]; [62]). This may imply scope and absorptive-capacity effects, since a panel averages the effects over all countries and a single country limits the industrial base. We caution that most designs do not claim causal effects.

This is consistent with the interaction model that macroeconomic variables influence growth through institutions and structural constraints, e.g., where conflict causes investment reductions, inflation reduces planning quality, weak institutions and infrastructure constrain FDI and openness effects, and poor ICT infrastructure constrains financial and digital-inclusion effects ([17]; [38]; [20]; [12]; [50]).

The evidence used a contextualized endogenous growth model, where human capital, knowledge diffusion, and institutions are not substitutes for macroeconomic policy but act as mediating channels that condition its effectiveness ([52]; [5]). Burundi’s experience shows that, in the absence of governance quality and human capital preconditions, external factors (openness, capital accumulation) will not promote growth.

Macroeconomic stability is a precondition for any growth push. Monetary authorities should aim at transparent, responsible single-digit inflation-targeting, as there is no reason to relax the 10% external benchmark to ignore; credibility requires transparent communication, predictable policy, fiscal consolidation, and deepening financial markets for transmission efficiency.

Lasting conflict settlement, inclusive, democratic institutions and rule of law, property rights, and security are essential to unleash investment ([38]). Investment promotion should be targeted (ICT, agro-processing, manufacturing), including conditional performance on technology transfer, jobs, SME integration, and procurement transparency to diminish policy risk ([22]; [10]).

Labor-absorptive growth and skills matching, linking Burundi to regional value chains to avoid dependence on raw commodities, reducing fragility through new digital infrastructure and a friendly fintech environment ([50]; [42]), combining climate-risk awareness and resilient, climate-sensitive investment plans, areas where further research and policy action are needed ([53]; [32]). These evidence-based policy recommendations are summarized in Table 6.

Table 6.Evidence-based policy recommendations.