Original Research

Comparing perceptions and expressions of policy uncertainty

Waldo Krugell, Raymond Parsons, Lodewalt Venter, Willem Fourie

Received: 03 June 2025; Accepted: 24 Apr. 2026; Published: 06 Aug. 2026

Copyright: © 2026. The Authors. Licensee: AOSIS.
This work is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license (https://creativecommons.org/licenses/by/4.0/).

Abstract

Orientation: Policy uncertainty imposes real economic costs. South African policymakers and business analysts now have access to two concurrently published indices: the perceptions-based North-West University Policy Uncertainty Index (PUI) and the Phronesis Analytics Expressed Policy Uncertainty Index (EPUI). Both claim to measure policy uncertainty, but they use different methods and information sources. No prior study has characterised when and why they diverge.

Research purpose: This article provides the first quantitative characterisation of co-movement and divergence between the PUI and EPUI over their overlap period (2015Q3–2021Q4).

Motivation for the study: The aim is to identify the conditions under which each index provides distinct information, enabling practitioners to make informed decisions about which index to monitor for specific purposes.

Research approach/design and method: We compare the two indices using z-score standardisation, cross-correlation functions, Granger causality tests and Ordinary Least Squares (OLS) regression.

Main findings: The two indices are weakly correlated (r = −0.18), confirming they capture different dimensions of policy uncertainty. The PUI shows a significant negative relationship with business confidence (r = −0.48, p < 0.05) and leads business confidence by one to two quarters. The EPUI does not show a comparable pattern. The 2016Q3 episode illustrates the practical stakes: the PUI and EPUI gave contradictory signals that quarter, and an analyst relying on either index alone would have received an incomplete picture.

Practical/managerial implications: Practitioners should monitor both indices simultaneously. When they diverge, the divergence itself carries information about the mismatch between expressed and perceived uncertainty.

Contribution/value-add: This article provides the first quantitative comparison of the PUI and EPUI, characterises their divergence patterns and enables informed index selection.

Keywords: policy uncertainty; South Africa; PUI; EPUI; business confidence; gross fixed capital formation; Granger causality; cross-correlation.

Introduction

Orientation

Policy uncertainty indices are monitoring tools. When two indices measuring the same phenomenon diverge, the analyst faces a signal conflict: both readings are plausible, but they imply different conclusions. Resolving the conflict requires knowing what each index measures, why they might disagree and what that disagreement reveals about the underlying uncertainty environment.

South African analysts now face exactly this situation. The North-West University (NWU) Business School publishes the quarterly Policy Uncertainty Index (PUI), which aggregates perceptions of policy uncertainty from news coverage, economist surveys and business surveys (Parsons & Krugell 2022). Phronesis analytics publishes the Expressed Policy Uncertainty Index (EPUI), derived from natural language processing of parliamentary speeches and policy documents (Fourie 2023). Both indices claim to measure policy uncertainty, and as we show, they often disagree.

The international literature documents this problem at the global level. Ozturk and Sheng (2018) show that news-based, forecast-disagreement and financial-market measures of uncertainty give different signals about the same underlying environment. Baker, Bloom and Davis (2016) note explicitly that their EPUI captures only one dimension of policy uncertainty. What this implies for practitioners who must choose between conflicting index readings remains unanswered in the literature.

In the South African context, the problem is concrete. In 2016Q3, the PUI fell sharply while the EPUI remained elevated. An analyst monitoring only the PUI that quarter would have concluded that policy uncertainty was declining. An analyst monitoring only the EPUI would have concluded the opposite. Both readings were internally consistent. Neither was complete. No prior study has characterised this divergence or its implications for index users.

Research objective

This article provides the first quantitative comparison of the PUI and EPUI, characterising their co-movement and divergence over their overlap period (2015Q3–2021Q4). The specific contribution is to identify when the indices agree, when they do not and what conditions drive divergence, enabling practitioners to make more informed decisions about which index to monitor for which purpose.

The two indices measure different stages of a common transmission process. Political discourse shapes the information environment; business agents interpret that environment through news, data releases and expectations. The EPUI captures the upstream stage: what policymakers express in speeches and parliamentary records. The PUI captures the downstream stage: how businesses and economists perceive the resulting uncertainty. Under this construct, contemporaneous divergence between the two indices is expected, not anomalous. The 2016Q3 episode is consistent with this. Positive economic data releases suppressed perceived uncertainty (PUI) without reducing political ambiguity in parliamentary discourse and policy documents (EPUI).

We test the implied lead-lag structure using cross-correlation function (CCF) and Granger causality tests and assess the relationship of each index to real-sector indicators using OLS regression. Given the 26-observation sample, all econometric results should be treated as indicative rather than confirmatory.

Literature overview

The international literature establishes that different measurement approaches to policy uncertainty capture different phenomena.

Baker et al. (2016) introduced a news-based index for the United States and showed that elevated uncertainty precedes declines in investment, employment and output. The EPU index now covers more than 26 countries. Ozturk and Sheng (2018) compared news-based, forecast-disagreement and financial-market measures for G7 economies and found that news-based measures lead the others by one to two quarters. Indices differ not just in level but in timing. Ahir, Bloom and Furceri (2022) extended this with the World Uncertainty Index, based on Economist Intelligence Unit country reports, and showed it predicts investment and economic growth across 143 countries with a one-to-two quarter lag. The consistent finding across this body of work is that uncertainty measures are not interchangeable: they capture different information, respond at different speeds and correlate differently with real-sector outcomes.

South African contributions have developed a range of indices, each studied in isolation. Redl (2018) combined professional forecaster disagreement, news-based counts and Reserve Bank Monetary Policy Review mentions of uncertainty into a composite index and found it predicts recessions and co-moves with financial volatility. Hlatshwayo and Saxegaard (2016) built a Factiva-based EPU variant and showed that policy uncertainty dampens the responsiveness of exports to the exchange rate. Kotze (2017) and Aye (2019) focused on fiscal uncertainty using Generalized Autoregressive Conditional Heteroskedasticity (GARCH)-based volatility measures and found that shocks reduce output and investment. Binge and Boshoff (2020) constructed a composite measure from Bureau of Economic Research (BER) micro-surveys, text mining and financial indicators and found it significantly associated with lower real economic growth from 1992 to 2017. Kirsten (2020) showed that news-based uncertainty shocks reduce industrial production and depreciate the exchange rate. Additional contributions that use financial volatility as an uncertainty proxy include Balcilar, Gupta and Jooste (2017), Leballo (2020), Lesame (2021) and Ekeocha et al. (2023).

Two gaps remain in this literature. Firstly, every South African index has been assessed in isolation. No study has placed two concurrently available South African (SA) indices side by side and asked whether they agree. Secondly, while the international literature has demonstrated that different indices capture different phenomena, this has not been tested for SA-specific indices. Practitioners using either the PUI or the EPUI have no empirical basis for understanding when the two diverge, what drives that divergence or what it implies for their monitoring decisions. This paper addresses both gaps.

Research design

Parsons and Krugell’s Policy Uncertainty Index

The PUI, compiled by the Business School at North-West University, is published quarterly. It is a composite measure combining a news-based uncertainty indicator, a survey of economists and a BER manufacturers survey indicator, each weighted equally.

The news-based measure counts Google News articles containing ‘policy uncertainty’ and ‘South Africa’ within 10 words of each other (AROUND(10) operator). The economist survey asks five questions covering perceived changes in uncertainty, economic effects and policy areas of concern. The BER survey asks business managers whether political constraints are affecting their business. The three components are averaged quarterly, with the series base period set at 2015Q3 = 50 (Parsons & Krugell 2022). In-sample values range from approximately 40 to 80.

Phronesis Analytics’s Expressed Policy Uncertainty Index

Fourie (2023) used natural language processing of 10 972 speeches by South African policymakers and 29 357 parliamentary discussion records to construct the EPUI. It captures three dimensions: policy change intention (frequency of change-indicating language using a predefined lexicon and topic modelling); policy ambiguity (presence of hedging and unclear references) and policy inconsistency (cosine similarity between speeches and the government’s medium-term policy priorities). Scores for each dimension are aggregated quarterly and standardised as z-scores.

Conceptual framework

The PUI and EPUI are theoretically distinct in their information sources, their aggregation logic and their expected relationship to real-sector outcomes. The PUI aggregates signals from news coverage, economist surveys and business surveys: all three reflect how agents interpret the policy environment at the time of measurement. The EPUI captures what policymakers express in speeches and parliamentary records. These are different stages of the same transmission process. Expressed uncertainty precedes perceived uncertainty, with a lag that depends on information processing, media coverage and whether countervailing economic signals are present.

Contemporaneous divergence between the indices is therefore expected and informative. It signals a decoupling between the political discourse and the business perception environment, which is itself a condition worth monitoring. The 2016Q3 episode illustrates this. The PUI fell sharply, driven by better-than-expected gross domestic product (GDP) growth in 2016Q2, stabilising drought conditions, easing inflation and Finance Minister Gordhan’s fiscal restraint at the Medium-Term Budget Policy Statement (MTBPS). These signals reduced perceived uncertainty directly, through the news and survey components of the PUI. The EPUI, built from parliamentary speeches and policy documents, remained elevated. The divergence was not a measurement error in either index. It reflected a genuine decoupling between political discourse and business perceptions, and that decoupling was informative.

A testable implication follows: if expressed uncertainty leads perceived uncertainty, the EPUI should lead the PUI at some positive lag in the CCF. We test this in the Section ‘Quantitative analysis of co-movement and lead-lag relationships’.

Analytical approach

We analyse the overlap period 2015Q3–2021Q4 (n = 26 quarterly observations).

Z-score standardisation transforms both indices to a common scale for visual comparison, following the EPUI’s published construction. Four clarifications govern all exhibits: (1) The EPUI enters Table 1, and all correlation analyses as the published series, standardised over the full EPUI sample (2015Q1 onwards); the overlap-window moments in Table 1 are sub-sample descriptive statistics. (2) The PUI enters Table 1 in its raw composite-score form (mean approximately 51.83, SD approximately 7.95 over the overlap window). (3) For the figures, both series are re-standardised over the overlap window for visual comparison only; these re-standardised series are not used in regression or Granger tests. (4) Gross Fixed Capital Formation (GFCF) and BER Business Confidence Index (BCI) enter all analyses in their raw published form.

Contemporaneous correlation is assessed using Pearson correlation coefficients with two-tailed p-values. We also compute a rolling 8-quarter Pearson correlation to examine time-varying co-movement. Cross-correlation functions are computed for lags −4 to +4 quarters. A positive lag k means the uncertainty index at time t is correlated with the indicator at time t + k, indicating the uncertainty index leads. Both series were inspected for unit roots using the Augmented Dickey-Fuller test and treated as stationary over the overlap window. Pre-whitening was not applied; the reported cross-correlations may partially reflect autocorrelation within each series rather than genuine lead-lag structure. Cross-correlation function results should be treated as indicative of lead-lag patterns rather than as conclusive evidence.

The OLS specification regresses each indicator on a constant, the contemporaneous uncertainty index, and two lags, with Newey-West Heteroskedasticity and Autocorrelation Consistent (HAC) standard errors (2 lags). An Autoregressive Distributed Lag (ARDL) specification would be preferable with a longer sample; at n = 26, adding a lagged dependent variable would reduce degrees of freedom and risk near-multicollinearity, so the simpler specification is retained with HAC correction for residual serial correlation.

Granger causality F-tests are applied for PUI to GFCF, EPUI to GFCF, PUI to BER BCI and EPUI to BER BCI, with max_lag = 2. With 26 observations, power is limited. Results should be read as indicative patterns rather than as strong causal evidence.

Ethical considerations

Ethical clearance to conduct this study was obtained from the Stellenbosch University Social, Behavioural and Education Research Ethics Committee (Ref. No. 30793).

Results and discussion

Visual comparison of Policy Uncertainty Index and Expressed Policy Uncertainty Index

Figure 1 plots the two indices over the overlap period. The indices co-move in some quarters and diverge in others. Both patterns carry information.

FIGURE 1: North-West University Business School Policy Uncertainty Index versus Phronesis Analytics Expressed Policy Uncertainty Index: (a) Raw index values and (b) Z-score standardized overlay.

The 2016Q3 divergence illustrates the practical stakes. The PUI fell sharply while the EPUI remained elevated. Policy Uncertainty Index narrative reports point to the perception-shifting signals that quarter: better-than-expected GDP growth in 2016Q2, drought conditions stabilising, inflation settling and Finance Minister Gordhan’s fiscal restraint at the MTBPS. These signals reduced perceived uncertainty through the news and survey components of the PUI. The EPUI, built from parliamentary speeches and policy documents, measured a different reality: political discourse remained ambiguous. The divergence was not a measurement error in either index. It reflected a genuine decoupling between the political discourse environment and business perceptions at the time.

This is the article’s central practical finding: the two indices can give opposite signals during the same quarter, and both can be right. An analyst monitoring only the PUI in 2016Q3 would have concluded that the uncertainty environment was improving. An analyst monitoring only the EPUI would have concluded the opposite. Neither conclusion was complete.

Quantitative analysis of co-movement and lead-lag relationships

Table 1 presents descriptive statistics and contemporaneous correlations for all series over the overlap period.

The contemporaneous correlation between PUI and EPUI is r = −0.177 (p = 0.387). This is consistent with the transmission framework in Section ‘Conceptual framework’: the two indices measure different stages of the uncertainty process, so a weak or negative contemporaneous correlation is expected. The composite PUI-EPUI correlation is slightly more negative than any individual sub-index correlation (range: −0.163 to 0.275), which reflects the equal weighting of PUI sub-components. The substantive point is that the two indices are not redundant: monitoring only one gives an incomplete picture of the uncertainty environment.

The rolling 8-quarter correlation varies from −0.77 to 0.23, with a mean of −0.26. Each window contains only n = 8 observations, so point estimates carry wide confidence bands. These figures should be treated as exploratory illustrations of time variation rather than as directional evidence. They nonetheless suggest that co-movement between expressed and perceived uncertainty shifts with major structural events, with the 2016 political episode and the coronavirus disease 2019 (COVID-19) period as the likely drivers.

Both indices show negligible contemporaneous correlation with GFCF: r(PUI, GFCF) = −0.011 and r(EPUI, GFCF) = −0.068, neither significant. Fixed investment does not co-move with either uncertainty measure at quarterly frequency in this sample, consistent with the theoretical expectation that investment responds to uncertainty with a lag.

Figure 2 and Figure 3 show z-scores for the PUI and EPUI mapped to the BER’s Business Confidence Index and SACCI’s Business Confidence Index respectively. The PUI shows a significant negative contemporaneous correlation with BER BCI: r(PUI, BER BCI) = −0.480 (p = 0.013). This is the strongest bivariate result in the dataset and is consistent with the transmission channel: higher perceived policy uncertainty suppresses business confidence. The EPUI is positively but insignificantly correlated with BER BCI (r = 0.208, p = 0.308), consistent with the EPUI capturing an upstream stage of the transmission process that does not map directly onto contemporaneous business confidence.

FIGURE 2: Policy uncertainty versus Gross Fixed Capital Formation (Seasonally Adjusted and Annualised Rate [SAAR]) – Z-scores, 2015Q3–2021Q4.

FIGURE 3: Policy uncertainty versus (a) BER Business Confidence Index and (b) SACCI Business Confidence Index – Z-scores, 2015Q3–2021Q4.

Given the number of pairwise comparisons in this framework, the PUI-BER BCI correlation and the marginal Granger result below are the strongest signals in an exploratory design. They should be read as directional evidence rather than confirmatory tests.

Figure 4 presents the CCF for each uncertainty-indicator pair at lags −4 to +4 quarters. For PUI versus BER BCI, the largest negative correlation occurs at lag +1, suggesting PUI leads business confidence by approximately one quarter. This is consistent with the transmission channel: elevated perceived uncertainty feeds into lower business confidence in the following quarter. The EPUI CCF shows no dominant lag structure against BCI. Against GFCF, neither index shows a clear CCF pattern, reinforcing the conclusion that fixed investment is not strongly driven by uncertainty at short horizons in this sample. As noted, pre-whitening was not applied; the CCF patterns should be treated as indicative of lead-lag structure rather than as evidence of predictive relationships.

FIGURE 4: Cross-correlation functions: (a and c) Policy Uncertainty Index or (b and d) Expressed Policy Uncertainty Index vs Gross Fixed Capital Formation and BER BCI (lags −4 to +4).

Table 2 presents Granger causality results. At lag 1, the null that PUI does not Granger-cause BER BCI is marginally rejected (p < 0.10). Given 26 observations, this test has limited power; the result is consistent with a predictive relationship between perceived uncertainty and business confidence but does not confirm one. Neither index significantly Granger-causes GFCF. These findings complement the CCF evidence: perceptions-based uncertainty appears to precede movements in business confidence, while the link to fixed investment is weaker and likely operates over longer horizons.

OLS regressions (Table 3) confirm these patterns. Gross Fixed Capital Formation regressed on PUI, contemporaneous and two lags, yields no significant coefficients, consistent with near-zero correlations. BER BCI regressed on PUI shows a negative contemporaneous coefficient with marginal significance under HAC standard errors. Expressed Policy Uncertainty Index regressions return insignificant coefficients for both GFCF and BCI, consistent with the EPUI capturing an upstream stage of uncertainty transmission that does not map directly onto short-run business cycle fluctuations.

Policy uncertainty and investment: Evidence from capital projects

The theoretical link between policy uncertainty and investment is well established: uncertainty acts as a tax on investment by inducing firms to delay irreversible capital expenditure (Baker et al. 2016). The quarterly regressions in Section ‘Quantitative analysis of co-movement and lead-lag relationships’ return null results for aggregate GFCF, consistent with investment responding over medium-term horizons rather than within a quarter and with the limited power of tests at n = 26.

Figure 5 shows total Nedbank capital projects declining from R246 billion in 2016 to R179 bn in 2019 and R159 bn in 2020, before recovering to R187 bn in 2021. The sustained decline from 2016 to 2020 coincides with elevated PUI readings and declining BER BCI and is dominated by falls in Transport and Communications and Finance and Real Estate. Mining projects also trended downward, consistent with persistent mining policy uncertainty in the PUI narrative reports for this period.

FIGURE 5: Nedbank Capital Projects Register: (a) Sector breakdown and (b) total (R billions, 2015–2021).

This pattern is descriptively consistent with policy uncertainty suppressing private investment over the medium term. The annual frequency, absence of a counterfactual and null results from the quarterly regressions preclude a causal interpretation. The Nedbank data are presented as descriptive context and motivation for future investigation, not as corroboration of the econometric findings.

Conclusion

Three main conclusions follow from this analysis.

Firstly, the PUI and EPUI are weakly and negatively correlated (r = −0.18) over the 2015Q3–2021Q4 overlap period. This is consistent with the transmission framework: the two indices measure different stages of the uncertainty process, expressed uncertainty upstream and perceived uncertainty downstream, and contemporaneous divergence is therefore expected. Neither index is redundant. They capture different information, and monitoring only one gives an incomplete picture of the uncertainty environment.

Secondly, the PUI has a significant negative relationship with the BER Business Confidence Index (r = −0.48, p < 0.05) and leads business confidence by approximately one quarter in the CCF. This is consistent with perceived policy uncertainty suppressing business confidence through a one-quarter transmission lag. The EPUI does not show a comparable pattern, consistent with it capturing uncertainty at an earlier point in the transmission chain, before it has registered in business perceptions.

Thirdly, neither index shows a strong contemporaneous relationship with aggregate GFCF at quarterly frequency. The Nedbank Capital Projects Register shows a sustained medium-term decline in private investment from 2016 to 2020 that coincides with the period of elevated uncertainty and depressed confidence, but the null quarterly results and data limitations preclude a causal interpretation.

These findings have a concrete implication for practitioners. The 2016Q3 episode shows what single-index monitoring costs: the PUI and EPUI gave contradictory signals, and an analyst relying on either alone received an incomplete picture. The recommendation is not simply to monitor both indices but to treat divergence itself as a signal. When the PUI and EPUI disagree, the gap between expressed and perceived uncertainty is informative about the transmission environment: it indicates that political discourse and business perceptions are responding to different drivers, and that neither reading alone is sufficient. Policymakers facing conflicting signals should ask whether economic data are temporarily suppressing perceptions, as in 2016Q3, or whether expressed uncertainty has genuinely declined. Business analysts using the indices for investment timing will find that perceived uncertainty (PUI) tracks near-term confidence signals more closely, while expressed uncertainty (EPUI) may serve as a leading indicator of potential perception shifts.

The overlap period ends in 2021Q4. Extending the EPUI beyond this point would allow estimation of longer-horizon investment effects and a more powerful test of the lead-lag structure between expressed and perceived uncertainty. Future work might also develop a formal diagnostic for detecting economically significant divergence episodes, enabling practitioners to identify in real time when the indices are providing materially different signals.

Acknowledgements

The authors would like to thank the anonymous reviewers for their comments.

Competing interest

The authors declare that they have no financial or personal relationships that may have inappropriately influenced them in writing this article.

CRediT authorship contribution

Waldo Krugell: Methodology, Writing – original draft, Writing – review & editing. Raymond Parsons: Conceptualisation, Writing – original draft. Lodewalt Venter: Writing – original draft. Willem Fourie: Conceptualisation, Data curation, Formal analysis, Methodology, Writing – review & editing. All authors reviewed the article, contributed to the discussion of results, approved the final version for submission and publication and take responsibility for the integrity of its findings.

Funding information

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Data availability

The data that support the findings of this study are available from the corresponding author, Waldo Krugell, upon reasonable request.

Disclaimer

The views and opinions expressed in this article are those of the authors and are the product of professional research. They do not necessarily reflect the official policy or position of any affiliated institution, funder, agency or that of the publisher. The authors are responsible for this article’s results, findings and content.

References