Technology investment and innovation incentives can enhance international competitiveness and export performance by improving production efficiency, facilitating product differentiation, and upgrading quality. Empirical studies commonly report that R&D expenditure, equipment renewal, investment subsidies, and tax incentives are positively associated with export growth, export intensity, or value-added upgrading. In particular, evidence regarding the role of digital intensity in raising export quality has underpinned technology-upgrading policies, suggesting that digital-intensive industries may exhibit stronger policy transmission efficiency [3], [12], [13].

However, technology upgrading typically entails adjustment costs and learning dynamics. Some studies note that policy-induced technology investment may initially trigger resource reallocation and organizational restructuring, leading to delayed export responses and, in some instances, heightened short-run volatility [14]. Despite this, much of the literature evaluates policy performance through aggregate or industry-average effects, implicitly assuming broadly similar response mechanisms across sectors. For small open economies with highly heterogeneous industrial structures, this average-effect perspective may mask asymmetric responses, leading to a systematic underestimation of volatility risks in high-tech industries or adjustment costs in traditional manufacturing.

Industry heterogeneity is a key source of differential policy effects. High-tech and traditional manufacturing industries differ structurally in R&D intensity, human capital, and accumulated knowledge, implying distinct levels of absorptive capacity. The ability of firms or industries to identify, assimilate, and apply external knowledge determines the speed and extent to which technology investment translates into productivity gains and market competitiveness [6], [15].

High-tech industries typically convert technology investment into process efficiency and product upgrading more rapidly, leveraging participation in global value chains to adjust export strategies. In contrast, upgrading in traditional manufacturing more often takes an incremental form, and policy responses depend on organizational adaptation to new technological structures and technology readiness [5].

Technology intensity also shapes the nature of short-run volatility risk. In technology-intensive or globally fragmented industries, upgrading is often accompanied by significant structural reorganization and market reallocation, making volatility and uncertainty more likely to rise during the early transition [14]. Accordingly, technology intensity is not merely a background characteristic to be controlled for; it operates as a key moderating condition for the effects of technology-investment policy [16].

Export performance is inherently multidimensional. While export growth rates are widely used as the primary indicator, a single metric may not adequately reflect structural change and risk costs during transformation [17]. Consequently, the literature increasingly argues for jointly considering export orientation and export stability. Export orientation reflects reliance on external markets and the direction of structural adjustment, whereas Export Stability corresponds to exposure to short-horizon fluctuations under external shocks and captures short-horizon volatility exposure, which is conceptually distinct from long-run resilience [9].

Industrial upgrading and export-structure adjustment often involve resource reallocation, making higher short-run volatility a common feature during transition [4]. In more technology-intensive industries, faster and larger-scale restructuring can amplify the trade-off between "improved export orientation" and "increased short-run instability." This trade-off reflects the reality that global value chain participation may generate export gains while simultaneously introducing more complex volatility dynamics [10].

Observing both export orientation and export stability alongside export growth can therefore help identify heterogeneous adjustment paths under policy stimulus and reduce potential bias from inferring policy performance solely from growth indicators [9], [17], [18].

Existing research suggests that technology-investment policy is generally associated with improved export performance, but effects may be delayed and accompanied by short-run volatility risk [3], [4]. Differences in technology intensity and absorptive capacity may determine whether policy-induced investment is translated into export competitiveness and how volatility risk evolves during transformation. Although recent work has begun to consider firm size as a moderating factor in policy responses [11], there remains limited evidence from a unified empirical framework that systematically compares high-tech and traditional manufacturing in terms of differential responses in export growth, export orientation, and export stability. Moreover, many studies continue to emphasize average-effect tests [8].

Against this background, the present study groups industries by technology intensity and employs fixed-effects and interaction-term designs to examine heterogeneity in policy effects. This approach complements the average-effect perspective and provides an empirical basis for differentiated industrial policy design.

To examine whether technology-investment policy effects exhibit systematic heterogeneity, this study employs an industry-level panel dataset covering eight major manufacturing industries in Taiwan from 2016 to 2023 (N = 8, T = 8). These industries constitute a substantial proportion of Taiwan’s total manufacturing output and exports; thus, the sample is highly representative of the industrial structure and effectively reflects aggregate policy-response patterns. The dataset is compiled from official statistics and constructed as an industry–year panel.

Following the OECD classification and considering Taiwan’s specific industrial context, the sample is divided into "High-tech industries" and "traditional manufacturing" [19]. This classification reflects structural differences in technological sophistication, R&D foundations, and global value chain integration, serving as the basis for subsequent comparative analysis. The grouping results are summarized in Table 1.

Rationale High R&D intensity and technological sophistication; deep integration into global value chains with superior absorptive and adjustment capacities.

Upgrading patterns primarily driven by process improvements and internal efficiency; focus on the optimization and deepening of existing industrial structures.

This study employs three industry-level dependent variables to capture export growth, orientation, and stability: 1) export growth rate, reflecting annual shifts in export scale, 2) export ratio, representing the degree of external market reliance, and 3) export stability, measuring short-term export volatility during transformation, operationalized as the three-year coefficient of variation of export values.

The key independent variable is investment penetration, which measures the intensity of technology-investment incentives adopted in each industry during the study period, and is used as a proxy for policy penetration and actual uptake of policy-induced technology investment. Technology Intensity is operationalized using a binary indicator (high-tech = 1; traditional = 0).

Control variables include import growth rate, import ratio, large firm sales ratio, and fx change (annual change in the Real Effective Exchange Rate (REER)) to account for international demand, industry concentration, and macroeconomic fluctuations. Detailed variable definitions are presented in Table 2.

Export ratio Ratio of total exports to total output % value, measuring the degree of external market reliance and export orientation.

Export stability Three-year coefficient of variation of - export values (standard deviation/mean), capturing short-term export volatility (risk). A higher CV indicates greater volatility and lower stability.

Grouping Technology intensity Binary indicator: High-tech industries = - variable 1; Traditional manufacturing = 0. Used for heterogeneity and interaction analyses.

Control Import growth rate Year-on-year import growth rate, % variable reflecting changes in international demand and input costs.

Import ratio Ratio of imports to total output value, % capturing dependence on external inputs.

Large firm sales ratio Share of large-firm sales in total industry % sales, used as a proxy for industrial concentration and size structure.

FX change Annual change in the REER, measured as - the natural log of the ratio of the current to lagged exchange rate.

To estimate the impact of technology-investment policy on export performance, this study adopts a fixed-effects panel regression model to control for time-invariant structural differences across industries and to identify the policy effect from within-industry variation over time. The baseline model is estimated separately for the high-tech and traditional manufacturing subsamples to compare policy responses across groups. The baseline specification is as in Eq. (1).

Yi,t = αi + β1InvPeni,t + β2Xi,t + εi,t, (1)

where Yi,t denotes the export performance indicator for industry i in year t (i.e., export growth rate, export ratio, and export stability); InvPeni,t represents Investment penetration, capturing the intensity of policy-induced technology investment; Xi,t is a vector of control variables; αi represents industry fixed effects; and εi,t is the error term.

Although the high-tech subsample is smaller after grouping, the data have a finite population character, and the coefficients are interpreted as reflecting structural relationships rather than relying on random-sample inference [20]. In addition, given industry heterogeneity, group-specific estimation helps mitigate potential aggregation bias that may arise when heterogeneous industries are forced into a single pooled specification, thereby supporting unbiased parameter estimation [21].

To strengthen statistical power and directly test heterogeneity, the study augments the baseline model by adding an interaction between the policy variable and the technology-intensity indicator (InvPeni,t × HTi), This yields an interaction-based heterogeneity model estimated using the pooled sample, as shown in Eq. (2).

Yi,t = αi + β1InvPeni,t + β2(InvPeni,t × HTi) + β3Xi,t + εi,t. (2)

In this model, HTi is a dummy variable equal to 1 for high-tech industries and 0 for traditional manufacturing. Consequently, β1 represents the marginal policy effect for traditional industries, while β2 captures the additional marginal effect (difference in slopes) for high-tech industries relative to traditional industries. The total policy effect for high-tech industries is given by β1+β2. Finally, to assess robustness by controlling for time-specific common shocks, Eq. (2) is extended to include year fixed effects (δt), forming a Two-Way Fixed Effects (TWFE) model as shown in Eq. (3).

Yi,t = αi + δt + β1InvPeni,t + β2(InvPeni,t × HTi) + β3Xi,t + εi,t, (3)

where δt captures time-varying factors common across industries such as global business cycle conditions, exchange-rate movements, and external demand shocks. This specification allows the study to evaluate whether the estimated policy effects remain stable after controlling for common time trends. For inference, all models employ HC1 cluster-robust standard errors clustered at the industry level to address heteroskedasticity and within-industry serial correlation over time [22].

We first examine the statistical properties of the variables. Table 3 reports group-wise descriptive statistics for high-tech and traditional manufacturing industries over the sample period, together with difference tests. These difference tests are reported for descriptive purposes only, to summarize sample characteristics. The statistics reveal clear divergences in export patterns and structural features across the two groups:

I. Dependent variables: High-tech industries exhibit marked advantages in export performance. Both Export Ratio (0.790) and Export Growth Rate (0.097) are significantly higher than those of traditional industries (0.349 and 0.016, respectively; Diff = 0.442*** and 0.081**), indicating stronger export orientation and short-term export expansion in high-tech industries. For Export Stability (CV), the high-tech value (0.108) is slightly higher than that of traditional industries (0.081), but the difference is not statistically significant (Diff = 0.028). At the descriptive level, this suggests no clear group difference in short-horizon export volatility.

II. Independent variable: The mean Investment Penetration in high-tech industries is 0.131, which is about 1.7 times that of traditional industries (0.078). Although this difference is not statistically significant (Diff = 0.052), the lack of significance may reflect the very high within-group variation in high-tech industries (SD = 0.155), indicating heterogeneous adoption intensity across sub-industries.

III. Control variables: Traditional manufacturing shows a significantly higher Import Ratio (0.694) than high-tech industries (0.273) (Diff = −0.421***), implying greater dependence on imported inputs. In addition, the Large Firm Sales Ratio is significantly higher in high-tech industries (0.887) than in traditional industries (0.623) (Diff = 0.264***), suggesting a more concentrated market structure in high-tech sectors.

Overall, the descriptive evidence highlights substantial cross-industry heterogeneity in growth momentum, export orientation, and market structure. Given the “high mean–high variance” pattern of policy penetration, the subsequent fixed-effects models further control for time-invariant industry characteristics to identify policy effects more precisely.

parentheses. Difference statistics are reported for descriptive comparison only and do not constitute causal inference.

Significance: *p <0.05, **p <0.01, ***p <0.001.

To assess whether the policy effect differs by industry technology intensity, we estimate fixed-effects models separately for the high-tech and traditional manufacturing subsamples. Table 4 reports the results.

In high-tech industries, Investment Penetration is positively and significantly associated with Export Ratio (β =0.447, p < 0.01), suggesting a significant relationship between policy-induced investment and increased export orientation in high-tech sectors [23].

In traditional manufacturing, the coefficient is significantly negative (β=−0.124, p < 0.05). This pattern aligns with the view that, during the early stages of transformation, traditional industries may prioritize internal process upgrading and efficiency improvements over immediate outward expansion, reflecting an "internalization" tendency [10].

For Export Stability, the policy coefficient in high-tech industries is positive and significant (β = 0.299, p < 0.05), implying that while technology investment may strengthen outward orientation, it can also be associated with heightened export volatility and risk [14]. In traditional manufacturing, the corresponding coefficient is not statistically significant (0.011), indicating that export volatility appears less responsive to policy-induced investment in this group.

Regarding export growth rate, the high-tech coefficient is positive (β = 0.247, p ≈ 0.08) but only marginally significant, while the traditional coefficient is negative and not significant (β = −0.112), suggesting that the direct stimulus of the policy on export growth is limited.

Regarding controls, Import Growth Rate is positively significant for Export Growth Rate in both groups, consistent with a co-movement between import expansion and export momentum. For the high-tech Export Ratio, Import Growth Rate shows a significant negative coefficient (β = −0.145, p < 0.05), which may reflect that imports of intermediate inputs can mechanically dilute the export share, while also indicating strong complementarity between imports and exports within global value chain participation, leading to parallel development of the production–technology structure [23], [24].

In terms of industrial structure, the large firm sales ratio significantly boosts export growth in high-tech sectors (β= 5.171, p < 0.01) but enhances stability (reduces volatility) in traditional manufacturing (β = −0.763, p < 0.001). The observed sectoral differences suggest that industrial concentration amplifies expansion in technology-intensive sectors while buffering external shocks in traditional sectors [25], [26]. FX Change is generally positive and significant in the high-tech sample, consistent with the expectation that currency depreciation strengthens export competitiveness. In the traditional sample, the coefficient for Export Growth Rate turns negative (β = −0.440), though not statistically significant, suggesting a possible offset between price competitiveness and higher import costs.

Overall, the group-specific estimates indicate that policy incentives operate through divergent transmission channels: fostering a "high-orientation, high-volatility" path in high-tech industries, while supporting internal adjustment in traditional sectors.

measured by CV (higher value = lower stability). Estimates for variables with limited within-group variation (e.g., large firm sales ratio in high-tech sample) should be interpreted as structural rather than statistical relationships.

To formally assess whether technology intensity moderates the policy effect, we pool high-tech and traditional industries into a single framework and estimate interaction models. Table 5 reports the estimates controlling for industry fixed effects. The interaction coefficient (β₂) is positive and statistically significant across all three export outcomes.

The positive and statistically significant interaction effects indicate that high-tech industries respond significantly more strongly to policy-induced investment than traditional industries. Wald tests further indicate that the total policy effect for high-tech industries (β₁ + β₂) is statistically different from zero in all models (p < 0.001).

Note: Robust standard errors clustered at the industry level are in parentheses. β₂ represents the coefficient of the interaction term (InvPen×HT). *** p<0.001; ** p<0.01; * p<0.05.

Table 6 presents the results of the TWFE models, which control for common year-specific shocks. The results remain robust. Notably, in the Export Stability model, the baseline effect for traditional industries becomes significantly negative (β₁ =−0.198, p < 0.05), while the interaction term remains positive (β₂ = 0.382, p < 0.01). The opposing signs of the baseline and interaction effects highlight a clear structural divergence: policy investment is associated with stabilization (reduced volatility) in traditional sectors, whereas it is linked to increased volatility in high-tech sectors.

Overall, the interaction analyses consistently indicate that technology intensity acts as a key moderator, distinguishing the high-growth/high-volatility path of high-tech industries from the stabilization path of traditional manufacturing.

Note: Models include both industry and year fixed effects to control for common aggregate shocks. Robust standard errors clustered at the industry level are in parentheses. *** p<0.001; ** p<0.01; * p<0.05.

The findings reveal a clear dual-track mechanism of policy transmission. First, high-tech industries exhibit a trade-off between growth and stability. Policy-induced investment significantly boosts export orientation, consistent with the absorptive capacity perspective where high-tech sectors rapidly transform resources into market expansion [6]. However, this outward shift is accompanied by heightened short-run volatility. These findings suggest that for technology-intensive sectors, deeper integration into global value chains brings growth opportunities but simultaneously introduces greater exposure to external shocks and structural instability [4], [14].

In contrast, traditional manufacturing follows an "internalization" trajectory. The limited or negative effects on export growth suggest that policy benefits are not immediately realized in external markets. Instead, this pattern reflects a technological trajectory focused on process innovation rather than immediate product expansion [27]. Traditional industries appear to allocate policy resources primarily toward internal capability building and equipment upgrading. Consequently, assessing effectiveness solely through short-term export metrics may understate the foundational improvements occurring within these sectors.

Theoretically, this study challenges the assumption that investment incentives yield uniform export outcomes, underscoring the necessity of incorporating industry technology intensity as a critical structural filter in empirical models.

For high-tech industries, while incentives catalyze export orientation, they should be complemented by measures to manage the associated volatility risk. Policymakers must recognize that rapid outward expansion often entails greater exposure to external shocks. Conversely, for traditional manufacturing, a stagnation in the export ratio should not be interpreted as ineffectiveness. Instead, it reflects a phase of process optimization where outward-oriented returns require a longer horizon. Accordingly, policy assessments must adopt differentiated designs: monitoring stability risks for high-tech sectors while valuing internal capability upgrading for traditional sectors.

This study relies on industry-level data, a limitation that precludes the observation of within-industry firm heterogeneity. Future research could extend this analysis by examining the impact of technology investment on productivity, quality upgrading, or import substitution, thereby providing a more comprehensive assessment of the policy’s multifaceted channels and outcomes.