Price Pass-Through of Austria’s Single-Use Plastics Producer Charges: Evidence from Retail Offer Spells

Abstract

Single-use plastics (SUPs) impose substantial environmental costs. Following Directive (EU) 2019/904, Austria introduced producer charges and mandatory participation in collection and recycling systems. This paper exploits a monthly aggregated and disaggregated panel of retail offer spells drawn from a price-comparison platform to estimate the extent to which compliance costs pass through to posted online prices in Austria. The treated sample comprises keyword-matched SUP products—balloons, to-go cups, wet wipes, plastic bags, food containers, tobacco-filter items, beverage bottles, and plastic wraps—observed alongside a control group of non-SUP listings over 2020–2024. A two-way fixed-effects (TWFE) specification places the average post-treatment price increase at approximately 4.1 percent. A sequential TWFE model that disaggregates the administrative reporting phase (from March 2023) from the payment-due phase (from March 2024) reveals that the larger adjustment occurs during the earlier reporting stage, with a reporting-only effect of approximately 8.1 percent and an incremental payment-phase effect of 5.6 percent. For balloons—a category subject to pronounced regulatory fee exposure—event-study estimates exceed 50 percent in the months immediately following the initial payment date and remain elevated throughout most of the post-treatment window. Taken together, these findings indicate that Austrian online retailers began adjusting prices in advance of fee-payment deadlines, a pattern consistent with anticipatory pass-through of expected compliance costs rather than a discrete response to realized payments. As the data contain price observations but not quantity data, the analysis speaks to price incidence and does not extend to consumption or environmental outcomes.

1. Introduction

Single-use plastics (SUPs) litter beaches, waterways, and public spaces extensively. On EU beaches alone, plastics account for around 80–85 percent of collected litter items, about half of which are single-use products [1]. Street cleaning and waste collection tied to plastic debris cost municipalities billions of euros annually—costs that, absent regulation, are largely absorbed by the public purse. Environmental policy redirects part of that burden back to producers and consumers through charges, reporting obligations, and extended producer responsibility (EPR), pushing market prices closer to social costs [2].

Whether these upstream charges actually reach the shelf is an open empirical question. Under imperfect competition, pass-through depends on demand curvature, cost shares, and market conduct [3]. Existing evidence on sales and excise taxes suggests pass-through to retail prices is often substantial [4,5], and even modest levies on plastic bags have produced large consumption responses [6,7]. Deposit–refund systems form a related instrument aimed at curbing litter and improving collection rates [8]. What remains less understood is how compliance-based EPR regimes—combining reporting obligations, system-participation requirements, and per-quantity fee settlement—affect the prices of regulated products in online retail markets.

This study addresses that question for Austria. Following Directive (EU) 2019/904, Austria introduced producer charges tied to reported quantities, with fees of €225 per ton for food packaging, beverage cups, and plastic wraps, and €450 per ton for tobacco products with plastic filters [9]. Producers report volumes through the ARA Online platform; the compliance sequence requires reporting of prior-year quantities by 15 March and fee settlement through participation in licensed collection and recycling systems. March 2024 is therefore the treatment date used throughout this work, marking the first instance at which a reporting year translates into an actual payment obligation.

The empirical strategy uses a disaggregated and duration-weighted monthly panel of retail offer spells from a price-comparison platform covering both stand-alone e-tailers and third-party marketplace sellers, including those operating through Amazon and eBay. The observation window runs from January 2020 through end-2024, with a small number of spells extending into early 2025. The treated sample is defined by keyword matching to SUP-relevant product titles—balloons, to-go cups, wet wipes, plastic bags, food containers, tobacco-filter items, beverage bottles, and plastic wraps. The control group consists of non-SUP items belonging to the same categories in the product tree and, alternatively, non-related graphics card listings as a stress-test benchmark, all of which were active at the start of the sample window. Two baseline specifications are estimated: a standard two-way fixed-effects (TWFE) model and a dynamic event study, both supplemented by a sequential TWFE that separates the reporting phase from the payment phase.

The main findings are threefold. First, the pooled TWFE estimate places the post-payment price increase at approximately 4.1 percent—economically modest but statistically precise. Second, the sequential model shows that this average understates the timing pattern: the report-only phase coefficient is approximately 8.1 percent, exceeding the 5.6 percent incremental effect of the payment phase, which suggests that sellers began repricing before any fee was actually due, consistent with anticipatory pass-through of expected compliance costs. Third, in the balloon category—a product group with high regulatory exposure either through input material composition or the applicable SUP fee schedule—event-study coefficients exceed 50 percent in the months immediately following the first payment date and remain elevated through most of the post-treatment window, pointing to large and persistent pass-through in highly exposed segments. Negative pre-treatment coefficients in this category likely reflect the influence of lagged energy-price shocks rather than genuine pre-trend divergence.

These results contribute to several strands from the literature. They add to the growing body of evidence on the incidence of environmental charges and EPR systems [2,3,6,7,8], using high-frequency retail data rather than aggregate price indices such as consumer price indices (CPIs). They also speak to the behavioral economics of regulatory compliance: the finding that price adjustment precedes the first payment date is consistent with anticipatory firm behavior in response to legal and administrative onset, rather than realized costs alone. Finally, they provide early evidence on the Austrian SUP regime, complementing the related analysis in Reichel [10], who studies price responses in both Austria and Germany using a constructed price-index approach and a broader product control group, and finds pooled Austrian difference-in-differences contrasts of approximately 13 index points over a twelve-month post-policy window.

Several limitations warrant acknowledgment at the outset. The graphics card control group is conceptually distant from the treated SUP products, which means the parallel-trend assumption rests on the capacity of unit and time fixed effects to absorb differential trends and on the absence of broad substitution effects between the two groups. The data contain prices but not quantities, so the analysis speaks to price incidence and not to consumption, substitution, or environmental outcomes. The 2022–2023 European energy-price shock likely affected SUP-intensive and packaging-intensive supply chains differently, introducing a source of confounding that cannot be fully addressed through lagged input-price controls owing to collinearity with the treatment timing.

The remainder of the paper is organized as follows. Section 2 reviews the relevant literature. Section 3 describes the EU and Austrian regulatory background. Section 4 presents the data. Section 5 sets out the empirical strategy and reports the results using raw offer spells. Section 6 presents additional analysis from an aggregated panel, and Section 7 concludes.

2. Literature Review

2.1. Pass-Through of Taxes and Environmental Charges

The incidence and pass-through of commodity taxes has a long theoretical and empirical tradition. Under competitive markets with constant marginal costs, a per-unit tax is fully passed forward to consumers. Under imperfect competition, the pass-through rate depends on the curvature of demand and the conduct of firms, and can exceed 100 percent or fall well below it [3]. Empirical estimates from fuel-tax and sales-tax settings confirm that pass-through to retail prices is often substantial, though it varies across market structures, product types, and tax designs [4,5].

For environmental policy instruments specifically, the evidence points in the same direction. Convery et al. [6] study the Irish plastic bag levy—at the time a rare example of a direct charge on a single-use item at point of sale—and document a sharp reduction in consumption following its introduction, implying meaningful price transmission and a corresponding demand response. Homonoff [7] exploits variation in bag-fee adoption across Washington, D.C. neighborhoods to show that even modest charges of five cents substantially reduce disposable bag use, consistent with a combination of price sensitivity and salience effects. Walls [8] surveys the literature on deposit–refund systems, which combine a charge at purchase with a rebate at return, and finds that such instruments improve collection rates and reduce litter when the deposit is sufficiently salient. Taken together, this body of evidence establishes that consumer-facing charges on plastic and disposable products can generate large behavioral responses, but leaves open the question of how producer-side EPR charges—which are not directly visible at point of sale—transmit to retail prices.

2.2. Extended Producer Responsibility and EPR Incidence

EPR systems shift end-of-life costs for packaging and other products to producers, who typically discharge their obligations through financial participation in licensed collection and recycling schemes [1]. Unlike a simple per-unit tax, EPR charges are levied annually on reported quantities, entail administrative and compliance costs beyond the statutory fee, and may create incentives to reduce packaging intensity over time. The incidence of EPR charges is therefore not mechanically determined by the fee rate alone; it depends on how firms incorporate compliance costs into pricing decisions, on the degree of competition in affected markets, and on the transparency of the charge to downstream buyers.

Empirical evidence on EPR incidence at the retail level remains relatively sparse. Most studies focus on aggregate recycling rates, collection system performance, or per-ton cost estimates rather than on retail price outcomes [8]. The closest antecedent to the present study is Reichel [10], who estimates price responses to Austrian and German SUP producer charges using a constructed centered price-index event-study design with product and retailer fixed effects. That study reports a pooled Austrian difference-in-differences contrast of approximately 13 index points over a twelve-month post-policy window and approximately 20 index points over the full sample, with balloons exhibiting particularly strong and persistent effects. The present study builds on that evidence using a TWFE panel design with disaggregated and aggregated monthly log prices, a sequential treatment structure that separates the reporting phase from the payment phase, and a tighter focus on the Austrian compliance timeline. It also employs a more narrowly defined control group, matching treated SUP listings to non-SUP products within the same product-tree categories.

2.3. Behavioral Responses to Compliance Obligations

A growing literature in behavioral economics and public finance examines how firms respond to regulatory obligations beyond the direct cost of compliance. The present study contributes to this literature by documenting that price adjustment in the Austrian data appears to begin during the administrative reporting phase—before any fee becomes financially due—consistent with anticipatory repricing in response to expected future liabilities. This pattern is related to the broader literature on tax anticipation effects, which demonstrates that firms and households frequently adjust behavior in advance of statutory changes when future obligations are foreseeable [11]. It also connects to the economics of regulatory salience: once producers are required to track and report quantities, compliance costs become salient prior to financial settlement, which may be sufficient to trigger upward price adjustment even in the absence of realized payments.

3. Background

3.1. The EU SUP Directive

Directive (EU) 2019/904 on the reduction of the impact of certain plastic products on the environment entered into force in June 2019 and required member states to transpose it into national law by July 2021 [1]. The Directive targets ten categories of single-use plastic items that account for a disproportionate share of marine litter, including food containers, beverage cups, cutlery, plates, straws, cotton bud sticks, balloons, and cigarette filters with plastic components. Its instruments include outright bans on certain items, mandatory labeling requirements, consumption reduction targets, and EPR obligations requiring producers to finance litter clean-up, awareness campaigns, and waste collection infrastructure. Member states retained latitude over the precise design of their EPR systems; Austria and Germany both adopted producer-charge mechanisms linked to reported quantities.

3.2. Austria: Producer Charges and the ARA System

Austria’s transposition of the Directive took effect through the packaging ordinance and the single-use plastics regime administered by the Bundesministerium für Klimaschutz [9]. The compliance architecture is as follows. From 1 January 2023, producers and importers of SUP items covered by the Directive are required to participate in a licensed collection and recycling system—in practice, primarily through Altstoff Recycling Austria AG (ARA), the dominant licensing body [12]. Participation entails annual reporting of quantities placed on the Austrian market and payment of fees to the system operator, which finances collection, sorting, and reporting infrastructure.

The fee structure is product-specific. Food packaging, beverage cups, and plastic wraps carry a charge of €225 per ton, while tobacco products containing plastic filters face a higher charge of €450 per ton, reflecting greater associated clean-up costs [9]. Micro-enterprises falling below defined quantity thresholds may opt for a flat payment of €13. Quantities placed on the market in a given calendar year must be reported to the system by 15 March of the following year, with fee settlement proceeding from that report within the compliance calendar. March 2024 therefore constitutes the first date on which quantities placed on the market in 2023 generate a concrete financial obligation, and serves as the primary treatment date throughout this work. Reporting obligations applied from March 2023 for the 2022 reference year, but the corresponding payment was smaller or transitional for many firms still completing system registration; March 2024 represents the first full compliance cycle [13,14].

The digital reporting infrastructure is managed through the ARA Online platform, through which producers register, submit quantity data, and receive fee assessments [12]. The existence of a structured reporting portal rendered compliance costs administratively salient from the first reporting period, prior to the first full payment obligation falling due. This institutional feature is consequential for interpreting the sequential TWFE results: the report-only phase from March 2023 through February 2024 captures a period during which producers were tracking and declaring quantities without yet bearing the full payment obligation for the 2023 reference year.

3.3. The Macroeconomic Context

The observation window from 2020 to 2024 encompasses several overlapping macroeconomic shocks of relevance to identification. The COVID-19 pandemic disrupted supply chains and suppressed consumer demand throughout 2020 and 2021, producing broad price movements across retail categories. From early 2022, the war in Ukraine generated a sharp energy and commodity price shock that raised production costs for plastics, packaging materials, and logistics across Europe [15,16]. These cost pressures likely bore more heavily on SUP-intensive and packaging-intensive retail segments than on less input-sensitive product groups, constituting a potential source of confounding: a divergence in prices between treated and control products that predates the payment date and may persist into the post-treatment period. The empirical strategy addresses this partially through a restricted sample window and through specifications that incorporate lagged Brent crude prices, the German energy import index, and Austrian gas prices as reduced-form cost shifters. These controls are nonetheless collinear with the regulatory timeline, so the absorption of energy-price confounding remains imperfect in the absence of granular product-level data on SUP input shares and fee exposure.

A further potential source of confounding is the growth in public awareness of the environmental costs of single-use plastics and, more recently, of per- and polyfluoroalkyl substances (PFAS) associated with certain plastic packaging materials. Heightened media attention to these issues over the observation window may have shifted consumer demand for affected product categories independently of the regulatory charge, a confound that cannot be separated from the compliance effect using the available offer-spell data.

4. Data

The main analysis draws on high-frequency online offer spells from a price-comparison platform covering the Austrian e-commerce market. Each offer spell records a continuous listing of a product by a specific retailer, together with a start and end timestamp, an observed price, and a set of retailer and product attributes. The raw data are irregular in time because listings enter and exit the platform at different moments and may remain active for varying durations. For the additional analysis specifications in Section 6, these raw spells are converted into a duration-weighted monthly panel in which the unit of observation is a retailer–product pair observed in a given calendar month. The estimation window covers January 2020 through December 2024.

4.1. Treated Sample Construction

The treated sample is defined by keyword matching on product titles. Matching uses case-insensitive string patterns designed to capture product groups plausibly exposed to the Austrian SUP compliance regime.

Table A1. Keyword categories used to construct the treated sample.

SUP Category	Translation	Keyword Patterns
tabak	Tobacco-related products and filters	tabakfilter; zigarettenfilter; zigarette; rauchwaren; nikotinprodukt; tabakprodukt; e-zigarette; rauchgerät; tabakröhre; filterzigarette
becher	Single-use cups and to-go drink cups	to-go becher; einwegbecher; kaffeebecher; kunststoffbecher; trinkbecher; coffee to go; takeaway becher; getränkebecher; wegwerfbecher; plastikbecher
lebensmittelbehaelter	Single-use food containers	lebensmittelbehälter; takeaway box; to-go behälter; einwegbox; essensbox; menüschale; mittagsschale; essensbehälter; food container; kunststoffschale
tueten_folien	Plastic wraps, films, and small packaging	folienverpackung; verpackungsfolie; plastikfolie; säckchen; tütchen; verpackungseinheit; folie verpackung; beutelverpackung; kunststoffverpackung; kleinverpackung
flaschen_ohne_pfand	Non-deposit beverage bottles, especially disposable plastic bottles	getränkeflasche; pet flasche; einwegflasche; kunststoffflasche; saftflasche; wasserflasche; getränkebehälter; softdrinkflasche; limonadenflasche; to-go flasche
flaschen_mit_pfand	Deposit and reusable beverage bottles	mehrwegflasche; pfandflasche; getränkeflasche pfand; getränkebehälter pfand; rückgabeflasche; flasche mit pfand; getränkeflasche mehrweg
plastiktueten	Plastic carrier bags and shopping bags	kunststofftragetasche; plastiktüte; einkaufstüte; leichte tragetasche; dünne plastiktüte; tragetasche einweg; einwegtragetasche; tüte plastik; kleine plastiktasche
feuchttuecher	Wet wipes and disposable cleaning or hygiene wipes	feuchttuch; reinigungstuch; hygienetuch; babyfeuchttuch; pflegetuch; intimtuch; kosmetiktuch; einwegtuch; nassreinigungstuch; desinfektionstuch
luftballons	Balloons and balloon decoration products	luftballon; ballon latex; partyballon; heliumballon; einwegballon; deko ballon; ballonset; kinderballon; ballon dekoration

lists the nine keyword categories, their German labels, and the underlying matching patterns. The treated categories are balloons (luftballons), to-go cups (becher), food containers (lebensmittelbehaelter), plastic wraps and films (tueten_folien), non-deposit beverage bottles (flaschen_ohne_pfand), deposit and reusable bottles (flaschen_mit_pfand), plastic carrier bags (plastiktueten), wet wipes (feuchttuecher), and tobacco-filter products (tabak). A product enters the treated sample if its title matches at least one pattern in any of these categories. The composition of the treated sample is depicted in Figure 1.

4.2. Control Sample Construction and the Baseline-Survivor Restriction

• Intended design.

The ideal control group for this design consists of products that (i) are sold on the same platform by the same retailers, (ii) face similar upstream cost dynamics and demand seasonality, and (iii) fall outside the scope of the SUP regime. Let $M_i^{C,0}\in \{0,1\}$ denote membership in this originally intended counterfactual. The guiding principle was to draw the comparison group from products at the same level of the platform’s category tree as the treated items—direct substitutes or functional complements subject to substitution effects once the SUP charge raises the relative price of plastic alternatives—made from reusable, natural, or otherwise non-regulated materials. This class includes reusable versions of the treated products (stainless-steel bottles, glass food containers, textile carrier bags), products from natural or non-plastic materials (paper cups, cardboard food trays, wooden cutlery), and non-SUP items whose demand is plausibly linked to treated products through substitution effects.

Table A2. Intended keyword categories for the counterfactual sample.

Category	Translation	Keyword Patterns
Tabakprodukte—rare substitutes	Tobacco alternatives and non-standard smoking substitutes	%kräuterzigarette%; %pfeife%; %zigarre%; %snus%; %verdampfer ohne nikotin%; %rauchfreies nikotin%
To-go cups—reusable or natural materials	Reusable or natural-material drink cups	%keramikbecher%; %emaillebecher%; %kupferbecher%; %kokusnussbecher%; %mehrweg goblet%; %glas tumbler%
Food containers—non-plastic materials	Food containers of glass, ceramic, wood, or similar materials	%tiffin box%; %glasdose%; %keramikbehälter%; %holzbox%; %wachstuchbox%
Bags and wraps—natural or biodegradable materials	Natural-material bags, wraps, and biodegradable packaging	%wachstuch%; %stoffverpackung%; %juteSäckchen%; %leinenbeutel%; %reispapierverpackung%
Bottles without deposit—alternative materials	Non-deposit bottles made from alternative materials	%glasflasche klein%; %keramikflasche%; %steinzeugflasche%; %kupferflasche%; %emailleflasche%; %bambusflasche%
Bottles with deposit—reusable alternatives	Reusable and refillable bottle alternatives	%milchflasche glas%; %bierflasche glas%; %nachfüllflasche%; %tee flasche glas%; %flasche aus holz%
Carrier bags—textile-based alternatives	Textile and reusable carrier bags	%baumwolltasche handgefertigt%; %jute beutel bio%; %papiertüte deluxe%; %leinentasche%; %upcycling tasche%; %netztasche%
Wet-wipe alternatives—textile products	Reusable cloth-based wipe alternatives	%baumwolltuch%; %waschlappen bio%; %leinen tuch%; %textiltuch%; %nachhaltiges pflegetuch%
Balloon substitutes—decorative alternatives	Non-balloon decorative substitutes	%stoffgirlande%; %papierrosette%; %wimpelkette%; %papierlaterne%; %naturdeko%; %holzdeko%; %stoffblume%

lists the nine intended counterfactual keyword categories, their translations, and the underlying matching patterns. Had this design been implemented in full, the control group would have held fixed at least part of the retail environment in which treated products are sold.

• Implemented design.

In the platform data, the intended alternatives are sparsely listed and unevenly matched to treated products, leaving a control group too thin for panel fixed-effects estimation with adequate pre-period coverage. The implemented control sample is therefore drawn from all non-SUP products in the platform data that satisfy the life-cycle restrictions below and do not belong to the treated keyword categories. Let $M_i^C\in \{0,1\}$ denote the implemented counterfactual-match indicator. This broader control pool retains the category-tree logic of the intended design where the data permit: it includes the natural-material and reusable alternatives of Table A2 wherever spell coverage is sufficient, and fills the comparison group with the remaining non-SUP listings on the platform.

A separate stress-test specification restricts the control group to a single keyword, grafikkarte (graphics card), with $M_i^C=1$ if and only if the product title matches this rule and satisfies the life-cycle restrictions below. Graphics cards share the same platform infrastructure and price-reporting conventions as the treated products but are driven by entirely different cost factors—semiconductor cycles and, during part of the sample, cryptocurrency-related demand. If an effect is detectable against this economically distant counterfactual, it is unlikely to reflect correlated cost shocks between treated and control products. The main estimates employ the broader non-SUP pool as well as category-matched listings in the duration-weighted monthly aggregated panel analyzed in Section 6.

• Baseline-survivor restriction.

The control sample is further restricted by a baseline-survivor filter. Let $b_i$ denote the product birth timestamp, $d_i$ the death timestamp, and $t_0^{\mathrm{base}}=1,577,836,800$ the Unix time for 1 January 2020 00:00:00 UTC. The implemented control set is

$$\mathcal{C}=\left\{i:M_i^C=1,b_i\le t_0^{\mathrm{base}},(d_i=⌀\mathrm{or}{d}_i>t_0^{\mathrm{base}})\right\},$$

(1)

so that only products already active at the opening of the sample window are retained. All post-2020 entrants and units that exited before January 2020 are excluded by construction.

• Implications for panel composition and identification.

Because $\mathcal{C}$ is conditioned on survival at $t_0^{\mathrm{base}}$, the retained control sample mechanically over-represents older and more persistent products while under-representing short-lived listings, later entrants, and products with intermittent market presence. Formally, the observed control is not drawn from the full untreated population $\mathcal{U}$ but from the selected subset $\mathcal{C}\subset \mathcal{U}$ satisfying the survival condition at baseline. Three consequences follow: the panel is tilted toward incumbent products; if turnover patterns differ between treated and untreated goods, the restriction induces differential sample selection unrelated to the regulation; and the final panel represents a survivor cohort defined at baseline rather than the full market over 2020–2024. The final sample offer spell length distributions and observational unit counts are depicted in Figure 2.

This is not random attrition but systematic conditioning on survival at the start of the window. If survival correlates with price levels, retailer persistence, assortment quality, or product type, treatment–control comparability may be affected. The design remains valid, but the estimand is narrower: the treatment effect should be interpreted relative to a specific untreated survivor cohort rather than the full population of available untreated goods. The parallel-trend diagnostics in Section 4.5 should be read with both the composition of the control pool and this survival conditioning in mind.

4.3. Summary Statistics and Price Distributions

Given the economic distance between treated SUP products and the graphics card control documented in Section 4.2, cross-group-level comparisons carry no claim of cross-group comparability; they serve only to characterize each group’s price distribution before and after policy onset.

Table A6. Descriptive statistics: offer-spell panel by treatment status.

	Treated (SUP Products)			Control (Non-SUP Products)
	Full	Pre-Policy	Post-Payment	Full	Pre-Policy	Post-Payment
Obs.	1,738,870	870,500	468,450	1,039,833	690,900	163,051
Units	20,721	—	—	7054	—	—
Average monthly price (EUR)
Mean	24.99	23.95	26.86	24.38	24.65	24.71
SD	373.42	526.32	38.49	70.75	83.67	23.92
Median	18.95	18.09	19.29	15.64	14.99	16.95
p10	8.90	9.36	8.51	6.13	5.96	7.05
p90	38.19	36.38	40.94	53.90	53.90	59.47
Log average monthly price
Mean	2.94	2.92	2.97	2.81	2.79	2.89
SD	0.61	0.54	0.70	0.80	0.82	0.76

Notes: Post-payment refers to months from March 2024 onward. Price statistics are computed on spell-level average prices.

reports the main figures. The treated products have a mean monthly price of EUR 24.99 and a median of EUR 18.95; the control group averages EUR 24.38 with a median of EUR 15.64. Average treated prices rise from EUR 23.95 in the pre-policy period to EUR 26.86 in the post-payment period, while control prices shift only from EUR 24.65 to EUR 24.71. Dispersion is large in both groups. Figure 3 shows the log-price densities: the two distributions overlap substantially but the treated density is shifted modestly rightward. Within the treated sample, categories differ visibly in both central tendency and dispersion, so pooled regression estimates average over products with different baseline price levels and potentially different regulatory incidence.

4.4. Category-Level Descriptive Dynamics

Figure 4 plots relative log-price paths by treated category in event time, normalized to zero at $t=-1$, with the control path included for reference. The trajectories are heterogeneous: some categories remain close to baseline around both policy thresholds, while others exhibit larger deviations in the post-payment period. A modest average treatment effect may therefore coexist with large category-specific responses if strongly affected categories are offset by weakly affected ones. This motivates the sequential specifications that separate the reporting phase from the payment phase and the category-level analyses reported in subsequent sections.

4.5. Parallel-Trend Diagnostics

Figure 5 provides a category-by-category pre-trend diagnostic using a duration-weighted and aggregated panel of raw offer spells to remove the influence of extreme price outliers and provide some smoothing, indexing treated and matched-control series to 100 at $t=-7$. Support for parallel trends is uneven: in some categories the two series track closely over the pre-period; in others, divergence appears well before payment onset. These diagnostics do not invalidate the design, but they imply that credibility varies across product groups and that pooled estimates should be interpreted as reduced-form averages over categories with different degrees of pre-period comparability.

4.6. Covariates and Energy Input-Price Controls

The seller-characteristic controls include market location indicators, sales channel indicators, payment method indicators, and shipping destination indicators, all coded as binary variables from the raw platform fields documented in Appendix C.

Three monthly energy price series enter extended specifications as auxiliary controls: Brent crude oil prices (USD per barrel, FRED St. Louis), the German energy import index (2021 = 100, Destatis), and Austrian natural gas spot prices (EUR per MWh, OEGPI). Their correlation structure is reported in Appendix E.1 Figure A1.

Plastic packaging depends on petrochemical feedstocks—primarily naphtha, ethylene, and propylene—so its input costs track broad fossil-energy prices closely. The 2022–2023 European energy shock raised costs across plastics, packaging, and logistics [15,16]. If these shocks passed through more strongly to treated supply chains than to the graphics card control, a treated–control price divergence could arise before the SUP regime became binding and persist into the policy window. Unit and month fixed effects capture time-invariant unit differences and common monthly shocks but not differential trends arising from heterogeneous input-price exposure. The lagged energy variables are intended to absorb part of this cost channel; in the absence of product-level input shares they remain partial controls rather than a structural solution.

4.7. Outcome Variable

The main outcome variable is the natural logarithm of the average price within each retailer–product–month; descriptive statistics for this variable appear in Appendix F.1. Log prices reduce the influence of extreme values similar to the monthly duration-weighted aggregated panel analysis in Section 6 and allow regression coefficients to be interpreted approximately as percentage changes for modest effect sizes. Baseline tables therefore use log average monthly prices as the primary outcome variable.

5. Empirical Approach and Results

The analysis estimates the effect of the Austrian SUP regulatory regime on posted e-commerce prices using high-frequency retail offer spells drawn from a price-comparison platform covering stand-alone e-tailers and third-party marketplace sellers, including those operating through Amazon and eBay. The observation window runs from January 2020 through end-2024 (till early 2025 for some spells). Two baseline specifications are estimated on the Austria-only sample, both centered on March 2024 as the compliance-payment date. This timing follows directly from the Austrian packaging and SUP institutional sequence: firms report quantities for the prior calendar year by 15 March and settle fees through participation in collection and recycling systems, so March 2024 is the first date at which the 2023 reporting year translates into an actual payment obligation (for the institutional background, including annual reporting by 15 March, system-participation obligations from 1 January 2023, and the compliance architecture of Austrian collection and recycling systems, see European Union [1], European Parliament and Council [9], Altstoff Recycling Austria AG [12], Wirtschaftskammer Österreich [13], Umweltbundesamt (UBA) [14]).

5.1. Panel Structure and Notation

Units $i=1,\dots ,N$ are unique retailer–product pairs; $t=1,\dots ,T$ indexes calendar months. The outcome $Y_{it}$ is the log of the duration-weighted average monthly price for unit i in month t (nominal price levels and median monthly prices are used in robustness checks; the main tables use log average monthly price to ease interpretation and reduce the influence of extreme values [17]). Treatment status is $D_i\in \{0,1\}$ and $Pos{t}_t=1\{t\ge 2024:03\}$. Relative event time is $r_{it}=t-t_0$, where $t_0$ is March 2024; $r=-1$ as the omitted reference period. Unit fixed effects are ${\alpha }_i$ and calendar-month fixed effects are ${\lambda }_t$. Identification follows the standard difference-in-differences logic: within-unit variation over time eliminates permanent heterogeneity, while month effects absorb common calendar shocks [11,18].

5.2. Covariates

The control vector ${\mathbf{X}}_{it}$ absorbs time-varying differences in seller characteristics. Market location indicators cover Austria, Germany, the United Kingdom, the Netherlands, and Poland—the five countries that account for the large majority of cross-border seller activity on the platform and that registered the highest non-Austrian offer-spell volumes in the raw data. Sales channel indicators distinguish online-only offers, collection, and in-store pickup. Payment method indicators cover Mastercard, Visa, American Express, and Diners Club. Shipping destination indicators cover Austria, Germany, the United Kingdom, Poland, the Netherlands, and Ireland. Additional binary seller-attribute controls include country-specific sales and availability markers.

Extended specifications add monthly input-price series—Brent crude from FRED, the German energy import index (2021 = 100) from Destatis, and Austrian gas prices in €/MWh from OEGPI—plus their first three lags. These series proxy petrochemical and packaging input costs. The 2022–2023 European energy shock provides the main motivation: it likely affected SUP-intensive retail segments more severely than the graphics card control, so a treated–control divergence may arise from differential cost exposure rather than the regulation [15,16]. Without product-level input shares these series remain reduced-form cost shifters rather than structural indices, and they do not fully resolve the differential trends concern—a limitation addressed further in Section 5.10.

5.3. Estimating Equations

• Event study.

The dynamic specification is

$$Y_{it}={\displaystyle \sum _{\begin{array}{c}r=-12\\ r\ne -1\end{array}}^{12}}{\beta }_r\left(1\{r_{it}=r\}\times D_i\right)+{\mathbf{X}}_{it}^{\prime }\gamma +{\alpha }_i+{\lambda }_t+{\epsilon }_{it}.$$

(2)

Pre-treatment coefficients ($r<0$) test for differential pre-trends; post-treatment coefficients ($r\ge 0$) trace price adjustment as the policy becomes binding. The design compares a single treated cohort against a never-treated control group, so the multi-cohort contamination problems of staggered settings do not apply [18,19]. Because all treated units share a common treatment date, switching to the Sun–Abraham or Callaway–Sant’Anna estimators would deliver the same comparison; those corrections address cross-cohort contamination that is absent here.

• Pooled TWFE.

Restricting (2) to a single post-period indicator gives the standard TWFE:

$$Y_{it}=\delta \left(D_i\times Pos{t}_t\right)+{\mathbf{X}}_{it}^{\prime }\gamma +{\alpha }_i+{\lambda }_t+u_{it}.$$

(3)

$\delta$ is the average post-March-2024 price gap between treated and control units relative to the pre-period. With a single treated cohort against a never-treated group, this coincides with the standard two-group difference-in-differences estimand and under parallel trends.

• Sequential TWFE with multiple periods.

The institutional timeline motivates separating two phases. Treated firms first face a reporting requirement in March 2023 (quantities for 2022 declared) and then a financially binding payment in March 2024. Define $Repor{t}_t=1\{t\ge 2023:03\}$ and $Fe{e}_t=1\{t\ge 2024:03\}$. The sequential model is

$$Y_{it}={\delta }_R\left(D_i\times Repor{t}_t\right)+{\delta }_F\left(D_i\times Fe{e}_t\right)+{\mathbf{X}}_{it}^{\prime }\gamma +{\alpha }_i+{\lambda }_t+{\xi }_{it}.$$

(4)

Because $Fe{e}_t$ becomes active only after $Repor{t}_t$ is already in effect, ${\delta }_R$ captures the reporting-only effect (March 2023 through February 2024) relative to the pre-policy baseline, and ${\delta }_F$ captures the incremental effect of the payment phase. The cumulative post-March-2024 effect is ${\delta }_R+{\delta }_F$. This is a multi-phase TWFE, not a staggered-adoption design: all treated units move through both phases on the same dates.

Category-level estimates run (3) on sub-samples restricted to treated units in category c against the full control group, yielding a category-specific coefficient ${\delta }_c$ for each $c=1,\dots ,C$.

5.4. Inference

Standard errors are reported under two clustering schemes: by unit (retailer–product pair), which allows arbitrary serial correlation within listings; and two-ways, by unit and calendar month, which additionally allows cross-sectional dependence arising from platform-wide pricing movements [20,21]. The choice between these schemes bears on inference in high-frequency retail panels; both are reported as a robustness check.

5.5. Pooled Baseline Estimates

Table 1. Baseline TWFE estimates of SUP treatment effects on monthly prices.

	Dependent Variable: ln(Average Monthly Price)
	(1) Standard TWFE	(2) TWFE with Multiple Periods
Treated × Post-payment	0.0398 ***
	(0.0044)
Treated × $1\left\{\mathrm{Report}-\mathrm{only}\mathrm{period}\right\}$		0.0782 ***
		(0.0082)
Treated × $1\left\{\mathrm{Payment}-\mathrm{due}\mathrm{period}\right\}$		0.0548 ***
		(0.0052)
Unit fixed effects	✓	✓
Month fixed effects	✓	✓
Standard errors	Clustered by unit	Clustered by unit
Observations	103,074	103,074
Number of units	3213	3213
$R^2$	0.9088	0.9097
Within $R^2$	0.0023	0.0120
RMSE	0.2382	0.2370

Notes: Column (1) presents the standard TWFE specification (3) with a single post-payment interaction. Column (2) presents the sequential TWFE specification (4), distinguishing the report-only phase (March 2023–February 2024) from the payment-due phase (from March 2024). All specifications include unit and month fixed effects. Standard errors in parentheses are clustered at the unit (retailer–product pair) level; see Section 5.4 and Table A9 for retailer-level clustering. Significance: *** $p<0.01$.

reports the central pooled results. Column (1) presents the standard TWFE (3) with a single post-payment indicator; column (2) presents the sequential TWFE (4), separating the report-only period from the payment-due period. Additional estimates using aggregated panel data appear in Appendix F.2,

Table A9. Multiperiod TWFE estimates: sensitivity to clustering level.

	Dependent Variable: ln(Average Monthly Price)
	(1) Multiperiod TWFE Clustered by Unit	(2) Multiperiod TWFE Clustered by Retailer
Treated × $1\left\{\mathrm{Legal}-\mathrm{to}-\mathrm{payment}\mathrm{period}\right\}$	0.1377 ***	0.1377
	(0.0240)	(0.0945)
Treated × $1\left\{\mathrm{Post}-\mathrm{payment}\mathrm{period}\right\}$	0.0450	0.0450
	(0.0291)	(0.0756)
Unit fixed effects	Yes	Yes
Month fixed effects	Yes	Yes
Standard errors	Unit	Retailer
Observations	51,537	51,537
$R^2$	0.9197	0.9197
Within $R^2$	0.0075	0.0075

Notes: The omitted category is the pre-legal period; both coefficients are interpreted relative to that baseline. Column (1) clusters at the unit (retailer–product pair) level. Column (2) clusters at the retailer level. All specifications include unit and month fixed effects. Standard errors in parentheses. *** $p<0.01$.

.

The column (1) estimate of 0.0398 log points corresponds to an average post-treatment price increase of approximately 4.1 percent (a treated product priced at €10 before the policy rises to approximately €10.41; one priced at €25 reaches approximately €26.03). The sequential TWFE in column (2) reveals a timing pattern obscured by the pooled average. The report-only coefficient of 0.0782 log points (≈8.1 percent) exceeds the payment-due coefficient of 0.0548 log points (≈5.6 percent). The larger coefficient in the reporting phase relative to the payment phase is consistent with sellers adjusting prices in anticipation of compliance obligations: a seller aware from March 2023 that quantities must be tracked and ultimately settled has an incentive to adjust prices in advance rather than absorb costs and reprice upon payment. The smaller payment-due coefficient indicates that the price response does not constitute a discrete jump at the payment date but emerges earlier and persists at a somewhat lower level once the system enters the payment phase.

• Comparison to prior evidence.

A pooled effect of 4–8 percent is broadly consistent with the existing literature on environmental-charge incidence. Convery et al. [6] find near-full pass-through of the Irish plastic bag levy, which at €0.15 per bag implied a retail price increase of approximately 4–6 percent depending on the product. Fuel-tax pass-through estimates in Marion and Muehlegger [4] cluster around 80–100 percent of the statutory rate in competitive retail markets, translating into price responses of comparable magnitude. For Austrian and German SUP segments, Reichel [10] reports a pooled Austrian difference-in-differences contrast of approximately 13 index points over a twelve-month post-policy window using a price-index event-study design; a larger figure, attributable in part to the broader post-payment window and the difference in normalization. The present 4 percent pooled estimate covers a shorter window and is partly driven by lower-exposure categories; the balloon sub-sample analyzed in Section 5.7 demonstrates that high-exposure products can exhibit effects exceeding 50 percent, consistent with the upper range in Reichel [10]. The Homonoff [7] estimates for US bag charges are smaller (1–3 percent) and reflect lower statutory rates; Walls [8] notes that pass-through tends to rise with market concentration and product specificity, both of which vary across the treated categories here. Taken together, the order of magnitude of the pooled estimate is plausible, though the phase pattern—larger in the reporting phase than the payment phase—is less directly comparable to the existing evidence, which does not generally separate anticipatory from realized cost responses.

• Energy confound.

A potential threat to a causal interpretation is that the 2022–2023 European energy shock raised costs for plastic-intensive supply chains more than for the graphics card control. If that differential persists into the 2023–24 window, part of the estimated treatment effect may reflect lagged cost pass-through rather than regulatory incidence. The energy controls described in Section 5.2 do not fully resolve the differential trends concern, as the lagged covariates are collinear with the treatment timing. The concern therefore remains without product-level input shares, and the estimates should be read with this caveat in mind.

5.6. Sensitivity to the Level at Which Standard Errors Are Clustered

Appendix F.2 Table A9 re-estimates the sequential TWFE clustering standard errors at two levels: the unit (retailer–product pair) and the retailer. Clustering at the unit level, which allows arbitrary serial correlation within a listing, yields a report-only coefficient of 0.1377 (≈14.8 percent) and a payment-due coefficient of 0.0450 (≈ 4.6 percent). Clustering at the retailer level—which additionally allows dependence across all listings from the same seller—leaves point estimates unchanged but reduces precision substantially, particularly for the earlier phase, pushing the payment-due coefficient below conventional significance thresholds.

Two observations follow. The stability of point estimates across clustering levels is reassuring: the estimated effects are not an artifact of the variance estimator. The precision loss at the retailer level suggests that residual dependence operates partly through seller-wide pricing decisions, so inference should be interpreted cautiously where treatment variation is concentrated within retailers.

5.7. Balloon Prices: A High-Exposure Case Study

Pooled estimates average over products that differ in regulatory exposure, input-material intensity, and pass-through capacity. Balloons constitute a useful high-exposure case study: a narrowly defined category with direct and well-documented exposure to the SUP regime. Appendix F.2,

Table A10. Event-study estimates of Austrian balloon prices around the first payment date.

Event Time	ln(${\mathbf{Average}\mathbf{Price}}_{\mathit{it}}$)		Event Time	ln(${\mathbf{Average}\mathbf{Price}}_{\mathit{it}}$)
	Coefficient	Standard Error		Coefficient	Standard Error
Payment month ($t=0$, 2024:03)	0.446 ***	(0.060)	$t+6$ (2024:09)	0.309 *	(0.148)
	(+55.9%)			(+34.7%)
$t+1$ (2024:04)	0.442 ***	(0.108)	$t+7$ (2024:10)	0.207 **	(0.093)
	(+54.8%)			(+22.5%)
$t+2$ (2024:05)	0.433 ***	(0.112)	$t+8$ (2024:11)	0.187 **	(0.067)
	(+53.2%)			(+20.3%)
$t+3$ (2024:06)	0.516 ***	(0.133)	$t+9$ (2024:12)	0.043	(0.075)
	(+66.2%)			(+4.1%)
$t+4$ (2024:07)	0.311 ***	(0.081)
	(+36.0%)
$t+5$ (2024:08)	0.248 ***	(0.079)
	(+27.8%)
Observations	178		Products included	15
Product fixed effects	Yes		Sample window	2020:01–2024:12
Month fixed effects	Yes		Treatment month	2024:03
Adjusted $R^2$	0.964		Final sample month	2024:12
RMSE	0.118		Standard errors	Clustered by product

Notes: Event time $t=0$ is March 2024; the omitted category is the pre-payment reference period. The latest observable post-treatment horizon is $t+9$ because the sample ends in December 2024. All specifications include product and month fixed effects. Italicized values report Kennedy corrected percentage effects: $100\times \left[\exp \left(\widehat{\beta }-\frac{1}{2}\widehat{\mathrm{Var}}\left(\widehat{\beta }\right)\right)-1\right]$, where $\widehat{\mathrm{Var}}\left(\widehat{\beta }\right)=\widehat{\mathrm{SE}}{\left(\widehat{\beta }\right)}^2$. *** $p<0.01$, ** $p<0.05$, * $p<0.10$.

reports the event study for Austrian balloon prices centered on March 2024 alongside the standard TWFE (3) and the sequential TWFE (4) estimated on the balloon sub-sample, with the full non-SUP control group retained in each case.

The TWFE estimate for balloons is 0.312 log points (≈37 percent), roughly eight times the pooled figure, and statistically robust across both clustering levels. The sequential TWFE decomposition reveals the by-now-familiar pattern: the report-only coefficient (0.271, ≈31 percent) exceeds the payment-due coefficient (0.198, ≈22 percent), and together they imply a cumulative post-March-2024 effect of 0.469 log points (≈60 percent). The event study makes the timing concrete: the payment-month coefficient is 0.446 log points (≈55.9 percent), and subsequent months remain strongly elevated—0.442 at $t+1$ (≈54.8 percent), 0.433 at $t+2$ (≈53.2 percent), 0.516 at $t+3$ (≈66.2 percent)—before gradually declining, with the coefficient falling to approximately 4 percent by $t+9$ (December 2024).

Three observations are noteworthy. First, for highly exposed products the regulatory impact is not merely detectable but substantial in magnitude. Second, the persistence over several months points to gradual pass-through or broad category repricing rather than a one-period spike. Third, the difference between a 55 percent balloon effect and the pooled 4 percent illustrates why aggregation matters: mixing strongly and weakly affected categories compresses the average. The balloon sub-sample comprises 178 observations across 15 products, so these estimates are indicative rather than definitive, but they motivate the disaggregated heterogeneity analysis in Section 5.8.

• Composition concern.

Part of the balloon price increase may reflect the exit of lower-priced listings rather than repricing by continuing sellers. Figure 2 shows that treated observation counts rise sharply from 2023 onward while the control series remains stable, consistent with compositional drift. A balanced-panel robustness check restricted to units surviving through both policy dates is reported in Section 6.2; point estimates are somewhat smaller but the qualitative pattern is preserved.

5.8. Heterogeneity Across Product Categories

Category-level TWFE models (3) estimated on category-c sub-samples confirm that the pooled average obscures substantial underlying heterogeneity. Appendix E.2 Figure A3 shows that ${\delta }_c$ varies in both sign and magnitude across categories. Figure 5 indicates that pre-treatment comparability with the control group is also uneven, so category-level estimates should be assessed jointly by point estimate, precision, and sample size rather than by coefficient magnitude alone.

A natural extension pools categories into exposure groups based on the fee tiers visible in Figure 1 and tests whether estimated effects rise monotonically with ex ante regulatory intensity. A monotone pattern across fee-linked groups would strengthen the causal interpretation of the pooled results.

• Demand-side shifts and PFAS salience.

Consumer awareness of plastic-related harms—driven partly by EU-level SUP communication and partly by media coverage of PFAS and microplastics—rose over the sample period. If demand for affected categories shifted independently of compliance costs, the estimated price response may partly reflect market contraction rather than cost pass-through. No clean instrument for separating regulatory-compliance salience from PFAS-driven consumer responses is available; the estimates should be interpreted as equilibrium price changes that may incorporate both cost and demand effects.

5.9. Heterogeneity Across Sellers

The panel combines stand-alone e-tailers and marketplace-based sellers, who differ in pricing technology, assortment breadth, and pass-through capacity [22]. Splitting the sample by seller type and re-estimating both (3) and (4) would clarify whether the observed price response is concentrated in one segment of the online market. Larger sellers may distribute fixed compliance costs across a broader product range, accelerating pass-through; smaller sellers may face proportionally heavier fixed burdens but weaker pricing power. These mechanisms cannot be separated within the pooled estimates.

This distinction matters both economically and econometrically. Because retailer-level clustering in Section 5.6 reduces precision noticeably, seller heterogeneity constitutes one source of within-group residual dependence. A seller-by-seller decomposition would assist in interpreting both the mechanism and the clustering structure.

5.10. Identification and Remaining Threats

Identification rests on treated and control prices following parallel trends absent the policy, conditional on unit effects, month effects, and controls. Pre-treatment event-study coefficients for $r<0$ provide a partial check. Pre-trend tests have well-documented power limitations, and conditioning inference on passing them can distort test size [23,24]; clean pre-trends are therefore treated as supportive evidence rather than conclusive validation.

• Unobserved confounding.

If treated products differ from controls along unobserved dimensions—brand positioning, consumer-segment exposure, or retailer strategic priorities—parallel trends may fail even after conditioning on fixed effects and controls.

• Anticipation.

Firms may have begun adjusting prices before March 2023, the formal reporting onset, through trade-association communications or interactions with licensing systems such as ARA. If so, the pre-treatment event-study coefficients may understate the true onset of the response and the reporting-phase coefficient may underestimate the full adjustment. Extending the pre-period beyond the current twelve-month window would help assess this possibility.

Because the control group is drawn from a separately sampled counterfactual product set, interpretation also depends on the sample-construction choices in Section 4.2—particularly the narrowing of the control group and the baseline-survivor restriction. These push the estimand toward units that remain observable throughout the window, so the estimates should be read as internal to the matched surviving sample rather than as representative of the full Austrian online retail market.

5.11. Conceptual Mechanism

Figure 6 summarizes the principal channels through which posted prices in the treated SUP sample may change over the observation window. The Austrian SUP regime can affect prices directly through expected and realized compliance costs, with statutory intensity differing across fee-schedule tiers (€225 and €450 per ton). The 2022–2023 energy shock and its lagged pass-through may affect treated categories differentially through heterogeneous input-mix exposure, particularly where petrochemical, packaging, and logistics inputs constitute a larger share of total costs. These cost channels may interact with price-adjustment costs, administrative salience, seller exit, and resulting market-share reallocation within the treated segment.

Observed price changes cannot be attributed exclusively to regulatory pass-through. Posted-price responses may combine regulation-induced cost changes, exposure-weighted input-cost shocks, seller-side repricing frictions, composition effects from exit and survival, and non-policy demand shifts. Because the data contain prices but not quantities, and because product-level input shares are not observed, these channels cannot be separated. The estimates are therefore reduced-form equilibrium price responses in the treated online segment rather than a structural estimate of pass-through of the statutory fee alone.

5.12. Summary of Findings

The evidence yields three principal conclusions. First, the Austrian SUP compliance regime is associated with a positive price response in the treated online segment: the pooled effect is economically modest but robust, on the order of 4–6 percent depending on specification, and falls within the range of prior pass-through estimates for comparable environmental charges [4,6,10].

Second, timing matters. Prices begin rising during the reporting phase, before any fee is due. The report-only coefficient of 0.0782 implies that a €15 item reaches approximately €16.22 before the first payment date—a larger adjustment than the €15.84 implied by the payment-due coefficient alone. Sellers appear to respond to the legal and administrative onset of the regime rather than exclusively to realized monetary obligations.

Third, aggregation conceals substantial heterogeneity. The balloon case demonstrates that high-exposure products can exhibit effects exceeding 50 percent in the immediate post-treatment period, sustained over several months before fading. A €5 balloon item rising to €7.80 in the payment month represents a qualitatively different phenomenon from the pooled 4 percent average. Disaggregated category analysis, seller-type splits, and composition checks are the natural next steps; the pooled estimate provides a useful benchmark but is unlikely to be the most informative object in a setting where regulatory exposure, packaging intensity, and pass-through capacity vary substantially across the sample.

6. Additional Analyses

The pooled TWFE results in Section 5 establish a positive average price response but leave four questions unresolved: whether effects differ across categories in a way that maps onto the SUP fee schedule; whether the baseline estimates reflect within-product repricing or compositional turnover; whether the response is concentrated in one segment of the seller population; and whether the price increase coincides with demand contraction or demand expansion. This section addresses each in turn using the aggregated panel of retailer–product–month offer spells described in Section 4.

6.1. Category-Level Heterogeneity and Fee-Tier Test

The nine treated keyword categories differ in their position in the Austrian fee schedule: balloons and tobacco filters face €450/ton, while the remaining seven categories face €225/ton. If the estimated price response reflects SUP pass-through, effects should be systematically larger in the higher-tier group.

Table 2. Category-level pass-through heterogeneity.

Tier	Category	${\widehat{\mathit{\beta }}}_{\mathit{c}}$	SE	p	N	Most Expensive Keyword-Based Selected Treated Sample Product
€450 per ton
	Balloons	$0.447$ ***	$\left(0.101\right)$	0.000	150	Konstsmide led motiv szenerie heissluftba… (€74)
	Tobacco filters	$0.342$ ***	$\left(0.102\right)$	$0.001$	122	Zebra ladegerät für rw 420, zigarettenanz… (€272)
€225 per ton
	Plastic bags	$2.696$ ***	$\left(0.103\right)$	$0.000$	4	Litepanels leichte tragetasche für astra… (€250)
	Food containers	$0.765$ ***	$\left(0.110\right)$	$0.000$	60	Blanco sitybox einhängbare kunststoffscha…(€43)
	Dep. bottles	$0.221$	$\left(0.244\right)$	$0.366$	2	Dennerle co2-adapter nano mehrwegflasche…(€23)
	To-go cups	$0.001$	$\left(0.202\right)$	$0.997$	1545	Villeroy & boch anmut platinum no. 1 kaffe…(€207)
	Non-dep. bottles	$-0.130$	$\left(0.271\right)$	$0.633$	6	Schott zwiesel basic bar selection wasser…(€41)
	Plastic wrap	$-0.372$ ***	$\left(0.116\right)$	$0.002$	60	Qeridoo fußsäckchen für fahrradanhänger… (€123)
	Wet wipes	$-0.435$ ***	$\left(0.130\right)$	$0.001$	104	B + w photo-clear 18 × 18 cm mikrofaser-reinig… (€44)
Tier pooled test ($H_0$: ${\beta }_{€450}={\beta }_{€225}$)
	€225/ton avg	$-0.023$	$\left(0.179\right)$	$0.896$	—	—
	€450/ton avg	$0.403$ ***	$\left(0.092\right)$	$0.000$	—	—
	Difference	$0.427$ **	$\left(0.201\right)$	$0.035$	—	—

Fixed effects: month (within-month demeaning). Standard errors: retailer-clustered (231 clusters), shown in parentheses. ${\widehat{\beta }}_c$: coefficient on $D_i^c\times 1\{\tau \ge -24\}$ from the joint model (5). Tier-test standard error is conservative. N: treated units in category. Most expensive product: highest mean price across all spells in the estimation window. Plastic bags ($N=4$) should be interpreted with caution. *** p < 0.01, ** p < 0.05.

reports category-specific coefficients ${\widehat{\beta }}_c$ from the joint interaction model

$$Y_{it}={\displaystyle \sum _{c=1}^C}{\beta }_c\left(D_i^c\times 1\{\tau \ge -24\}\right)+{\alpha }_i+{\lambda }_t+{\epsilon }_{it},$$

(5)

where $D_i^c=1\{D_i=1,C_{ic}=1\}$ and $\tau$ is months relative to the payment date. Standard errors are clustered at the retailer level (231  clusters).

The estimates are heterogeneous in both sign and magnitude. Figure 7 plots the coefficients with 95% confidence intervals and labels the most expensive product in each category.

Among the €450/ton categories, balloons show a coefficient of 0.447 log points (≈56 percent, $p<0.001$) and tobacco filters 0.342 log points (≈41 percent, $p=0.001$). The €225/ton group is heterogeneous: plastic bags (2.696, driven by a very thin sample of four units) and food containers (0.765) show large positive effects, while to-go cups (0.001) and deposit bottles (0.221) are near zero, and plastic wrap ($-0.372$) and wet wipes ($-0.435$) are significantly negative. The tier-pooled test in the lower panel of Table 2 finds an average €450/ton coefficient of 0.403 ($p<0.001$) against an average €225/ton coefficient of $-0.023$ ($p=0.896$), with a tier difference of 0.427 ($p=0.035$). Figure 8 presents this comparison graphically.

Three observations are noteworthy. First, the fee-tier difference is statistically significant and economically large: a 0.43 log-point gap between the two groups is consistent with the view that higher statutory rates generate larger price responses. Second, the negative coefficients for plastic wrap and wet wipes indicate that certain €225/ton categories experienced price declines over the same period, possibly reflecting competitive pressure, product-mix shifts, or the natural-material substitution dynamic described in Section 4.2. Third, the plastic bag estimate of 2.696 warrants caution given a sample of four units; it is excluded from the tier average in the robustness checks below.

Figure 9 plots the five most expensive products per category by mean price, providing a sense of the product landscape underlying each coefficient.

6.2. Balanced-Panel Robustness

The baseline panel is unbalanced: treated observation counts rise sharply from 2023 onward while the control series remains stable (Section 5.5). If lower-priced listings exit disproportionately following the policy, the average treated price rises mechanically without any within-product repricing. To separate these margins, the sample is restricted to units observed in at least 25 of the estimation months—a balanced sub-panel that excludes the most intermittently observed listings, representing a conservative alternative to a strict survival restriction.

Figure 10 compares the three-period TWFE coefficients and the full event-study path between the balanced and unbalanced samples.

The payment-due coefficient rises from 0.394 in the unbalanced panel to 0.541 in the balanced sub-panel; the report-only coefficient increases from 0.158 to 0.219. Point estimates are larger for the longer-lived listings, not smaller, which is inconsistent with the compositional exit hypothesis: if exit of lower-priced listings were driving the baseline result, restricting the sample to survivors would be expected to compress rather than amplify the coefficient. The event-study paths track each other reasonably through the pre-period before diverging in the payment phase, consistent with the balanced sub-panel capturing a set of products whose sellers respond more persistently to the compliance regime. Wider confidence bands in the balanced sample reflect the smaller unit count. Taken together, these results indicate that the baseline estimates are not primarily artefacts of changing sample composition.

6.3. Seller-Type Heterogeneity

The panel combines stand-alone e-tailers and marketplace-based sellers (Amazon, eBay, and similar platforms, identified by retailer slug patterns). The two groups differ in pricing technology, assortment breadth, and compliance cost structure: marketplace sellers operate within platform-level pricing tools and may face different fixed compliance costs per listing compared to stand-alone shops. Figure 11 reports the three-period TWFE coefficients and event-study paths separately for the 203 treated marketplace units and 1850 treated standalone units.

The payment-due coefficient for standalone sellers is 0.494 log points (≈64 percent), roughly six times the marketplace estimate of 0.080 (≈8 percent). The report-only phase shows a similar pattern: standalone sellers register a larger anticipatory response than marketplace sellers in both the point estimate and the event-study path. The marketplace path drifts slightly upward from the reporting onset but remains close to zero through most of the post-payment window, while the standalone path rises more clearly and remains elevated.

Two interpretations are consistent with this pattern. First, standalone sellers may face higher per-unit compliance and repricing costs that are more fully passed on, while marketplace sellers—who route compliance through platform infrastructure—may absorb a greater share of the burden or distribute it across larger volumes. Second, the marketplace sub-sample is approximately ten times smaller (203 versus 1850 units), so the wider confidence bands for that group do not preclude effects of similar magnitude to the standalone estimate. The seller-type split should therefore be interpreted as indicative evidence on the concentration of the pricing response rather than as a definitive decomposition.

6.4. Alternative Outcome Variables and Economic Magnitude

The baseline outcome is ln(average monthly price). Two considerations motivate alternative specifications: log-point effects are harder to assess economically without a reference price level, and duration-weighted averages may assign disproportionate weight to long-lived spells. Figure 12 reports three-period estimates for three outcomes side by side: log price, EUR-level duration-weighted price, and EUR-level unweighted price.

The announcement and report-only coefficients are near zero across all three outcomes, while the payment-due coefficient is positive and statistically significant in log points (0.394, $p<0.05$) and positive but less precisely estimated in EUR levels (€16.5 in both weighted and unweighted specifications). The qualitative timing pattern—payment phase dominant—is robust across all three outcome definitions. The EUR-level estimates are large in absolute terms, reflecting the presence of high-priced products in the sample (to-go cups at €207, tobacco filters at €272) that generate substantial EUR changes even for modest log-point effects.

Figure 13 translates the category-specific log-point coefficients from Table 2 into estimated EUR price changes using pre-treatment category means, $\Delta {\widehat{p}}_c={\overline{p}}_c^{\mathrm{pre}}\times (\exp \left({\widehat{\delta }}_c\right)-1)$.

The EUR magnitudes vary substantially across categories. Food containers exhibit the largest positive effect (€+48.2, $+115$ percent on a pre-mean of €42), followed by balloons (€+18.0, $+56$ percent) and tobacco filters (€+18.1, $+41$ percent). To-go cups are near zero in EUR terms (€+0.0) despite constituting the largest category by unit count. Plastic wrap and wet wipes show negative EUR changes (−€4.3 and −€3.9, respectively), with food containers pulling the pooled estimate upward. This decomposition reveals that the pooled log-point average constitutes a unit-count-weighted mixture of economically diverse outcomes: three categories account for most of the positive signal while two sizeable categories move in the opposite direction.

6.5. Demand-Side Proxy: Listing Frequency and Duration

Price increases that reflect cost pass-through need not coincide with demand contraction in a platform setting where sellers can adjust listing intensity in response to market conditions. The panel contains two demand-adjacent variables available at the unit–month level: the number of distinct price spells in a month ($\ln n_{\mathrm{spells},it}$) and the total active-price days covered ($\ln {\mathrm{days}}_{it}$). Neither constitutes a direct measure of consumer demand—higher spell counts may reflect more frequent repricing rather than greater sales volumes—but both provide an initial indication of whether the policy coincides with retraction or expansion of market presence.

Figure 14 plots event-study coefficients for these two outcomes using the same TWFE specification as the main results with $\tau$ relative to the payment date.

Both $\ln n_{\mathrm{spells}}$ and $\ln \mathrm{days}$ rise after the payment date relative to the control group. Spell counts increase by approximately 0.5–0.9 log points in the post-payment window; days covered rises by a similar magnitude. This pattern is inconsistent with demand contraction as the primary driver of the price increase: if consumers were substituting away from treated products in response to higher prices, market presence would be expected to contract. The evidence is instead consistent with sellers increasing their activity on the platform—raising listing frequency and coverage—around and following the compliance onset, possibly as they update price lists and refresh product pages in response to the administrative burden of the regime.

This finding does not preclude demand contraction at the transaction level, as spell counts and days measure listing activity rather than realized sales. It does indicate, however, that the price increase documented in the main results is not accompanied by visible retraction from the platform on the treated side.

Figure 15 examines whether the payment-due price coefficient is sensitive to controlling for $\ln n_{\mathrm{spells}}$ directly—a test of whether the price effect operates through repricing conditional on listing behavior or partly through the composition of listings.

Adding $\ln n_{\mathrm{spells}}$ as a control leaves the payment-due coefficient essentially unchanged (0.397 versus 0.394, unconditional). The price effect is not driven by shifts in listing frequency within unit–months; it reflects genuine within-unit price adjustment, conditional on continued market presence. The report-only and announcement coefficients are also stable across the two specifications.

6.6. Summary of Additional Analyses

Four findings emerge from this section. First, category-level pass-through is tied to the fee schedule: the €450/ton tier (balloons, tobacco filters) shows a tier-average effect of 0.403 log points against $-0.023$ for the €225/ton group, a statistically significant difference of 0.427 ($p=0.035$). The monotone fee-tier pattern supports a causal interpretation of the pooled estimate. Second, the baseline estimates do not primarily reflect a composition artifact: the balanced-panel sub-sample of long-lived listings produces larger coefficients, not smaller, suggesting that the pooled effect is, if anything, attenuated in the full unbalanced panel by the presence of short-lived, intermittently observed listings that may not have adjusted prices. Third, the pricing response is concentrated among standalone e-tailers (0.494) rather than marketplace sellers (0.080), pointing to differences in compliance cost structure or pricing technology between the two segments. Fourth, the price increase coincides with rising rather than falling platform presence among treated products and is robust to controlling for listing frequency, which together rule out the most direct form of the demand-contraction hypothesis and confirm that the estimated effect reflects within-unit repricing rather than compositional shifts.

7. Conclusions

This paper asked whether the Austrian SUP compliance regime—reporting obligations, system-participation requirements, and per-quantity fee settlement introduced under Directive (EU) 2019/904—passes through to posted online retail prices. The pooled answer is yes: treated products are on average about $4.1\%$ more expensive after the first payment date relative to the control group. The more revealing finding is in the timing. The sequential TWFE model puts the larger price adjustment in the reporting phase—around $8.1\%$— with an incremental payment-phase effect of $5.6\%$. Sellers began repricing before any fee was due, consistent with anticipatory pass-through of expected compliance costs rather than a contemporaneous reaction to realized payments. For balloons—a category with direct and transparent SUP exposure—event-study coefficients exceed $50\%$ in the months immediately following the first payment date and remain elevated through most of the post-treatment window before fading around December 2024.

The additional analyses in Section 6 sharpen this picture along four dimensions. Category-level pass-through is tied to the statutory fee schedule: the €450/ton tier (balloons and tobacco filters) averages 0.403 log points against $-0.023$ for the €225/ton group, a difference of 0.427 ($p=0.035$) that supports a causal reading of the pooled estimate. Restricting to a balanced sub-panel of long-lived listings raises the payment-due coefficient from 0.394 to 0.541, arguing against compositional exit as the primary driver of the baseline result. The pricing response is concentrated among standalone e-tailers (payment-due coefficient 0.494) rather than marketplace sellers (0.080), pointing to differences in compliance cost structure or pricing technology between the two segments. And listing frequency and days covered both rise after the payment date rather than fall, which is inconsistent with demand contraction and consistent with sellers actively updating price lists and refreshing listings in response to the compliance burden.

Four conclusions follow. First, SUP compliance costs in Austrian online retail are not fully absorbed by producers: a meaningful share reaches posted prices, and the pass-through is larger for the more heavily taxed categories. Second, the reporting obligation alone—before any fee is due—is sufficient to trigger price adjustment, consistent with administrative salience rather than only the monetary charge driving firm behavior [7]. Third, the pooled estimate is a lower bound for the most exposed categories: the €450/ton tier, balloons in particular, shows effects many times larger, and the parallel analysis in Reichel [10] finds full-period contrasts of around 20 index points for Austrian balloons in a price-index design. Fourth, the demand proxies available in the platform data—spell counts and days covered—point toward repricing and listing expansion rather than market retraction on the treated side, suggesting that the price increase is not primarily a composition artifact.

7.1. Limitations

The main limitations are discussed where they arise—the control-group distance and baseline-survivor restriction in Section 4.2, the identification threats including lagged energy confounds, composition effects, and demand-side shifts in Section 5.10. Two points are worth restating. The estimand is internal to the treated-versus-non-SUP comparison and should not be read as representative of the full Austrian online retail market. The data contain prices but not quantities, so pass-through can be measured but consumption responses, substitution toward reusable alternatives, and downstream environmental effects cannot.

7.2. Directions for Future Research

Several of the extensions flagged in the body of the paper remain open. On the demand side, transaction-level click data, which would allow a cleaner test of whether the price increases documented here translate into substitution toward non-SUP alternatives—the mechanism the regulation is designed to activate. The listing-frequency results in Section 6.5 are suggestive but cannot settle this, since spell counts measure market presence rather than sales. Scanner data, platform transaction records, or linked administrative data on reported quantities would convert the price-incidence evidence into a welfare and environmental assessment.

On the supply side, the seller-type split in Section 6.3 shows that standalone e-tailers drive the result while marketplace sellers show much smaller effects. Understanding whether this reflects differences in compliance cost structure, pricing technology, or the platform’s own role in absorbing or transmitting regulatory costs is a natural next step. A structural pass-through model that maps the Austrian fee schedule onto observed price changes by seller type and category would also allow a sharper test of whether the magnitude of the response is consistent with full, partial, or more-than-full pass-through given the statutory rate.

Finally, adding 2025 data would allow cleaner separation of permanent price-level changes from temporary spikes, and reveal whether competitive pressure eventually erodes the pass-through visible in the post-payment window.

7.3. An Ex Post Identification Challenge: The 2026 Petrochemical Shock

Any extension of this analysis into 2026 data will face a severe new identification problem. The geopolitical disruption of early 2026, and the associated near-closure of a key transit route for seaborne oil and liquefied natural gas, generated a sharp simultaneous increase in petrochemical feedstock costs—naphtha, polyethylene, polypropylene, PET—of a magnitude with no close historical precedent [25,26]. Because SUP products are manufactured primarily from these feedstocks, any study extending an Austrian panel into 2026 will observe a cost shock entirely unrelated to SUP compliance obligations that will affect treated categories—food containers, beverage cups, bottles, wraps—far more than a non-SUP control group. This is structurally identical to the 2022–2023 energy confound discussed in Section 5.10, but larger in magnitude and more concentrated in the specific inputs that go into SUP products. The category-level heterogeneity results in Section 6.1 make this especially relevant: the same categories that show the largest pass-through—food containers, balloons, tobacco filters—are also among the most petrochemical-intensive, so a feedstock shock would generate exactly the treated–control divergence a DiD design would otherwise attribute to the regulation.

Four design adaptations would be needed for any evaluation covering 2026 data. First, the control group should be selected with explicit attention to petrochemical input intensity, not only platform-level comparability, so the feedstock shock affects treated and control categories as symmetrically as possible. Second, input-price controls should include product-specific resin series—polypropylene, PET, polyethylene—rather than only broad energy indices. Third, any event window covering 2026 should treat the shock onset as a named identification threat rather than absorbing it silently into time fixed effects. Fourth, a triple-difference design exploiting cross-category variation in resin intensity alongside the SUP treatment variable could partially separate the regulatory effect from the feedstock shock, given a credible measure of category-level input intensity.

The 2022–2023 Ukrainian energy shock and the 2026 petrochemical disruption are two episodes from the same underlying source of identification risk. Research designs in petrochemical-intensive sectors should treat input-price heterogeneity as a first-order threat rather than a robustness footnote.

7.4. Broader Context

The finding that Austrian online retailers pass SUP compliance costs through to posted prices extends a body of evidence showing that producer-facing environmental charges are not silently absorbed in the supply chain [2,3,4,5,6,7]. The fee-tier monotonicity result—larger pass-through for €450/ton than €225/ton products—provides within-sample evidence that the magnitude of the regulatory burden matters, complementing the cross-study comparisons in the broader literature. The paper also complements Reichel [10], who reaches similar qualitative conclusions using a price-index event-study design and finds pooled DiD contrasts of around 13 index points over twelve months and 19 index points over the full period. Two studies using different outcome metrics, aggregation approaches, and control group definitions consistently find positive price incidence for Austrian SUP products in online retail, and the category and seller heterogeneity documented here suggests that the aggregate figure masks substantial variation in who bears the cost and how quickly it is passed on.

If the goal of SUP charges is to shift relative prices toward non-regulated alternatives, the evidence suggests the mechanism is at least partially operative in Austrian online retail. Whether the response is large enough to materially reduce SUP consumption or generate environmentally meaningful reductions in litter cannot be assessed without a greater quantity of data—that remains the binding constraint for any full welfare evaluation of the regime.