Defining the research question
Business context
A mid-sized Estonian trading company began working with a Japanese supplier in late 2019 — coinciding with the EU-Japan EPA coming into force. The company's strategy focuses on distant markets where logistics complexity creates a natural barrier for smaller competitors. The EPA reduced that barrier significantly.
South Korea went through a similar process earlier — EU-Korea EPA came into force in 2011. Does the Korean experience offer a preview of Japan's trajectory?
Research questions
- Did Estonia's trade with Japan increase after the EPA — and how did exports and imports behave separately?
- Which commodity groups grew in the 2019-2022 window, before geopolitical disruption set in?
- How does the Estonia-Korea post-2011 pattern compare to Estonia-Japan post-2019?
- How did COVID-19 and the war in Ukraine affect the data?
Data sources & structure
Sources
All data sourced from Statistics Estonia (Statistikaamet), downloaded manually in Excel from the PxWeb database. Local source chosen over Eurostat — the focus is specifically on Estonia's bilateral trade, and Statistics Estonia provides the most directly accessible data.
| File | Content | Period |
|---|---|---|
| VKK32 BY COUNTRIES | Export/import by partner country, 177 countries | 2004-2025 |
| VKK30 JP | Estonia-Japan trade by CN commodity group | 2004-2025 |
| VKK30 TOTAL | Estonia total world trade by CN group | 2004-2025 |
| VKK30 PARTS OF WORLD | Trade by world region | 2020-2025 |
*all file names are written with an underscore "_" and .xlsx in the end
Known data issues
- Wide format — years stored as columns, not rows
- Missing flow labels — "Exports"/"Imports" only in first row of each block
- ".." placeholder — suppressed values coded as ".." rather than NA
- Regional data gap — VKK30_PARTS_OF_WORLD only available from 2020 in Statistics Estonia's PxWeb database. This is why the regional destination charts (Wood, Fish) cover only 2020–2024 and not the full time series.
Data cleaning, transformation & analytical choices
This section documents every decision made between raw data and final output — what was done, why it was done, and what alternatives were considered and rejected. The goal is full reproducibility: someone else should be able to follow this chain and arrive at the same numbers.
Stage 1 — Structural cleaning
The raw Statistics Estonia export has three structural problems that must be fixed before any analysis is possible.
Wide format
Years are stored as columns (2004 … 2025). Every statistical tool works better with one row per
observation. pivot_longer() converts this to year as a variable. Why not leave it
wide? Filtering, grouping, and computing metrics across years becomes unwieldy — you'd need to
reference 22 column names explicitly in every operation.
Missing flow labels
"Exports" / "Imports" appears only in the first row of each block, not on every row.
fill(.direction = "down") propagates the label downward. Why not manually split the
file? Manual splitting breaks the reproducibility chain — re-downloading the data would require
repeating every manual step.
".." placeholder
Statistics Estonia uses ".." for suppressed or unavailable values — not NA, not 0. These are converted to
NA, not 0. Why NA and not 0? Zero means no trade occurred. ".." means the value
existed but is below the reporting threshold or otherwise unavailable. Replacing with 0 would artificially
inflate baseline averages and distort growth signals.
raw <- read_excel( "VKK30_JP.xlsx", skip = 2, col_names = FALSE )
colnames(raw) <- c( "flow", "commodity", "country", years )
df_clean <- raw %>%
fill(flow, .direction = "down")%>% # fix missing flow labels
pivot_longer(
cols = all_of(years),
names_to = "year",
values_to = "value")%>%
mutate(
value = as.numeric(
ifelse(value == "..", NA, value)))%>%
# ".." → NA, not 0
write_json(
df_clean,
"data/commodities.json",
pretty = TRUE)
Stage 2 — Selecting commodity groups (Pareto methodology)
The raw data contains all 99 two-digit CN codes. Most have zero or suppressed values for Estonia–Japan trade throughout the entire period. Rather than applying arbitrary threshold filters, groups are selected using a composite Pareto score that ranks all growing groups by their combined economic weight and contribution to growth.
The score is the average of two normalised shares:
share_FTA = fta_mean ÷ total_fta_volume— this group's share of total bilateral export volume across all CN groups during the EPA period (2019–2022). Total export volume in this window: 110.6 M€.share_Δ = abs_change ÷ total_growth— this group's share of total absolute growth, measured only across groups that actually grew (k > 1). Total growth across those groups: 44 M€.
Using two shares simultaneously rewards groups that are both large in volume and grew meaningfully. A group that dominates trade but barely moved scores lower than a group that is moderately sized but grew strongly. Both denominators are computed directly from the source data and update automatically if the underlying data changes.
Separate no-trade groups
Groups where fta_mean = 0 throughout the EPA period are excluded — roughly 24 of 98 chapters
in exports. Zero means the relationship doesn't exist. No threshold, just a structural check.
Growing vs declining
Growing groups: k = fta_mean / pre_mean > 1 AND abs_change > 0.
Declining groups are not in the primary analysis but are screened separately for spike signals.
Pareto classification (growing groups)
Group A: cumulative score ≤ 80% — the core EPA growth story, typically 3–5 groups. Group B: score 80–95% OR abs_change ≥ 1 M€ — economically meaningful secondary growth. Tail: score > 95% and abs_change < 1 M€ — not in primary analysis.
Spike detection (declining groups)
Among groups with k ≤ 1, those meeting: YoY 2018→2019 ≥ 2× and
fta_mean ≥ 1 M€ are marked as Spikes. These groups had a strong immediate
reaction to the EPA in 2019 but no structural follow-through — front-loading or announcement effect.
Stage 3 — Choosing the analysis period
Why 2014–2018 as "before", not 2004–2018: Trade structure in 2004–2010 was fundamentally different — Estonia had just joined the EU, volumes were much smaller, and several commodity relationships didn't yet exist. Including that era means comparing incomparable baselines. Five stable years (2014–2018) with the same underlying trade structure is a more honest pre-EPA window.
Why 2019–2022 as the "clean EPA window": Russia's airspace closure in February 2022 introduced a major logistics shock specific to Japan — no other major Estonian partner is as exposed because of the sheer distance involved. The 2019–2022 window captures the EPA effect before that shock. Using 2019–2024 as "post-EPA" would conflate the agreement itself with an unrelated geopolitical event.
Why show both windows: The full post-EPA period (2019–2024) shows the net outcome. The clean window (2019–2022) isolates the trade-policy effect. Showing both is more honest and more useful than picking one.
Stage 4 — Choosing metrics
Three metrics characterise each commodity group. Each answers a different question.
Mean before vs mean after
Average annual exports 2014–2018 vs 2019–2022. Answers: did the level of trade change? Simple to compute and explain. Limitation: doesn't distinguish between "the EPA caused a jump" and "trade was already growing steadily."
CAGR before vs CAGR after
Compound Annual Growth Rate in 2014→2018 and 2019→2022. Answers: did the speed of growth change? A group growing before the EPA isn't surprising — CAGR reveals whether it accelerated. Why CAGR over average YoY? CAGR computes from endpoint to endpoint, ignoring mid-period noise. Average YoY is skewed by single volatile years.
YoY 2018 → 2019
Year-on-year change at the exact moment of the EPA. Answers: was there an immediate jump in the EPA year itself? CAGR smooths over the transition point — YoY shows whether something happened specifically in 2019. A high 2019 YoY combined with low pre-EPA CAGR is the strongest possible event-driven signal.
Stage 5 — Composite score rationale
Why not sort groups by volume alone, or by % growth alone? Both approaches fail in small bilateral corridors:
Volume-only: rewards large but stagnant groups. CN44 Wood would dominate everything — it is already large, but the EPA’s marginal effect is modest relative to its existing scale.
Growth-only: rewards small groups with fast growth. CN71 Precious metals grew from near zero — any small absolute increase looks explosive in % terms, but its economic significance is limited until volumes build.
The composite score balances both: a group ranks high only if it is both large in volume and contributed meaningfully to absolute growth. CN84 Machinery illustrates this — modest share_FTA (3.4%) but high share_Δ (7.3%), placing it in Group A above larger but slower-growing groups.
The 80/95% thresholds reflect extreme concentration typical of small-economy bilateral trade: 4 export groups carry ~77% of the composite score; 2 import groups carry ~79%. This is the Pareto principle applied to a niche bilateral corridor. Groups A and B cover the economically significant story; the tail (below 95%, under 1 M€ growth) adds noise without insight.
Bilateral trade overview
The EPA year (2019) brought a visible export uptick — exports averaged 51 M€/year in the post-EPA period vs 36 M€ pre-EPA, a real and measurable shift. The real surprise came in 2022: trade hit an all-time peak at 150 M€, driven by surges in electrical machinery and chemicals. Then it fell sharply and kept falling.
The cause is logistics. Russia's airspace closure in February 2022 eliminated a routing that cut freight transit times by 3–4 hours on a round trip. That sounds minor until you consider that air cargo to Japan is already at the edge of what makes economic sense for many goods. Doubling cargo times made some shipments unviable. Maritime alternatives add 4–6 weeks. No other major Estonian trade partner is as exposed to this as Japan — the distance simply doesn't matter as much for nearby markets.
The import side tells a different story. It's structurally stable, dominated by machinery and electronics, with no dramatic EPA effect visible. Estonia buys from Japan fairly consistently regardless of trade agreements.
Exports & imports — absolute values, M€
Trade balance (exports - imports)
Japan's share in Estonia's total trade
One number worth pausing on: Japan's share in Estonia's total trade has stayed flat at roughly 0.5–0.8% throughout. The EPA created growth in absolute terms — but it didn't shift Japan's strategic weight in Estonia's trade geography. Japan is a real but niche partner, and the agreement didn't change that.
Growth index vs global trade (2019 = 100)
How to read: 2019 = 100 is the baseline for all series. A value of 150 means exports are 50% higher than in 2019; a value of 80 means 20% lower. The dashed Global line is Estonia's total exports to all countries — it shows what "normal" growth looked like without the Japan-specific logistics shock.
Japan vs other trade partners
The most meaningful comparisons are partners where growth can be attributed to trade policy rather than geopolitical accidents. Two partners fit that criterion here.
Singapore — also EPA 2019, also a distant market, also small bilateral volumes. Singapore did not show the same trajectory as Japan. That means the EPA alone doesn't explain Japan's growth. The commodity mix matters: Singapore has almost no wood trade with Estonia. Japan does. The EPA amplified something that was already structurally present.
South Korea — the most relevant benchmark. The EU–Korea EPA came into force in 2011, and its effect took 3–5 years to fully materialise in trade data. Japan is now at year 5–6 post-EPA. The trajectory isn't necessarily finished — but further progress requires the logistics situation (Russia airspace closure) to stabilise first.
Export growth index by partner country (2019 = 100)
How to read: all lines share the same 2019 = 100 baseline. This normalises for country size — a small country and a large one are directly comparable. Japan is the red line. Countries above Japan's line grew faster relative to their own 2019 level; countries below grew slower or declined.
Absolute export values — selected partners, M€
What actually grew — and what didn't
CN44 Wood is roughly half of all Estonian exports to Japan — not because of the EPA, but because of Japan's long-standing demand for Nordic and Baltic timber. The agreement gave an existing relationship an extra push (+50% on the post-EPA average, 36.3 M€ → 54.6 M€), but it didn't create it. This distinction matters: the EPA amplified what was already there. CN44 is a Group A commodity: largest share of both total FTA-period volume and absolute growth.
CN85 Electrical machinery is a cleaner EPA signal. Growth of +58% on a real base (9.2 M€ → 14.6 M€) is a genuine post-EPA story — these are products where tariff reduction directly changes the competitive economics. CN71 Precious metals is even more striking: it was near-zero before 2019 (0.04 M€ pre-EPA average) and averaged 7.2 M€/year in the 2019–2022 window, growing to 12.5 M€/year by 2020–2024 — an EPA-created trade segment. Both are Group A.
CN84 Machinery grew over 500% — from a 0.6 M€ pre-EPA base to 3.8 M€ post-EPA. Its composite score earned it Group A placement because of high share_Δ (7.3% of total absolute growth), not volume. A trend worth watching as volumes build.
Group B (CN08 Fruit & nuts, CN90 Optical, CN72 Steel, CN28 Chemicals, CN94 Furniture) shows meaningful but secondary growth — real, but below the 80% Pareto threshold in composite score.
The spikes tell a different story. Fish (CN03) and base metals (CN81) reacted strongly to the EPA in 2019 (YoY ≥2×) but averaged below their pre-EPA baselines across the full EPA period (k < 1). The EPA was noticed in 2019; structural relationships did not follow.
Exports to Japan — key groups, M€
Solid lines = Group A core (CN44, CN85, CN71, CN84) — top 80% of Pareto composite score. Dashed lines = Group B secondary (CN08, CN90, CN72, CN28, CN94) — score 80–95% or abs_change ≥ 1 M€. CN71 Precious metals was near-zero before 2019 and averaged 7.2 M€/year after — an EPA-created segment.
Imports from Japan — key groups, M€
Japan's share in Estonia's total exports by CN group
Pre vs post-EPA
Average annual exports (M€): 2014–2018 vs 2019–2022 (clean EPA window, before airspace closure). World % = same CN group, Estonia total exports to all destinations — to separate Japan-specific growth from global commodity trends.
| Group | Pre-EPA avg 2014–18 |
Post-EPA avg 2019–22 |
Japan % | World % | Signal |
|---|---|---|---|---|---|
| Loading… | |||||
% change post-EPA vs pre-EPA — Japan vs World
If Japan (coloured bar) grew much faster than World (sand bar) for the same CN group, that is a Japan-specific signal — not just a global commodity boom. CN85 Electronics: Japan +58% vs World +4% — Japan-specific. CN44 Wood: Japan +40% vs World +47% — global demand explains it. CN28 Inorg. chemistry: Japan +24% vs World +66% — Japan underperformed world. Spikes (CN03, CN81) are excluded from this chart — their k < 1 FTA average would distort the scale. See the spike chart below.
Trend acceleration: CAGR before vs after EPA
Mean values show that trade grew — but not why. CAGR separates continuation from acceleration:
what matters is whether the rate of growth changed at the EPA breakpoint, not just the level.
The clean EPA window (2019–2022) is used as “after” to isolate the agreement from the 2022 logistics
disruption. CAGR_total spans the full pre+post horizon as a background trend; EPA_accel shows
whether the EPA period outpaced that background.
| Group | YoY 2018→19 | CAGR before 2014→2018 |
CAGR after 2019→2022 |
Signal |
|---|---|---|---|---|
| A CN44 Wood | +1.1% | +7.4% | +22.0% | Accelerated 3× |
| A CN85 Electronics | −38.7% | −8.7% | +18.7% | Reversed — declining to growing |
| A CN71 Precious metals | +734% | n/a (new flow) | +49.9% | EPA-created segment |
| A CN84 Machinery | +292% | +105% | −30.6% | Front-loaded 2019 — structurally growing |
| B CN28 Chemicals | −38.6% | −2.9% | +11.4% | Reversed — declining to growing |
| B CN90 Optical | +52.7% | −8.8% | −4.5% | Jump in 2019, no sustained trend |
| ⚡ Spike CN03 Fish | +130% | −19% | −43% | 2019 spike — no structural change |
CAGR = compound annual growth rate. CN84 Machinery: negative CAGR after 2019 reflects front-loading — a large 2019 shipment (6.6 M€) normalised in subsequent years; the mean-based comparison still shows +521% growth (pre 0.6 M€ → post 3.8 M€). CN71 has no pre-EPA CAGR — the flow did not exist before 2019.
Pareto classification summary
Loading…
Note: Group A average % change is dominated by CN71 (from near-zero base, +19,000%) and CN08 (+2,474%). Excluding those two: CN44 +50%, CN85 +58%, CN84 +521%.
EPA spikes: front-loading without structural growth
CN03 Fish and CN81 Base metals both reacted to the EPA announcement in 2019 with a sharp one-year jump (YoY ≥2×, fta_mean ≥1 M€ threshold). But their average across the full EPA period remained below pre-EPA levels (k < 1). This is a “front-loading” or “announcement effect”: buyers exploited the new tariff in year one, but no durable supply chain adjustment followed.
Export spikes — CN03 Fish & CN81 Base metals
CN03 Fish: 2019 surge reflects front-loading ahead of tariff cuts taking effect; competition from other EU exporters prevented structural market share gains. CN81 Base metals: single-year opportunistic shipment pattern; no consistent buyer relationship established.
Was it the EPA or just good times? — Japan vs World comparison
The core analytical risk is confounding: trade might have grown for reasons unrelated to the EPA — general economic expansion, currency effects, or sector-wide demand cycles could produce similar charts. The Japan vs World bar chart tests this: if Japan-channel growth in a CN group substantially exceeds that group's global export trend in the same period, the excess is likely Japan-specific (agreement-driven).
Export structure — 2023 snapshot (Treemap)
Regional context
CN44 Wood — export destinations 2020-2024
CN03 Fish — export destinations 2020-2024
Regional breakdown data is only available from Statistics Estonia from 2020 onward — hence the shorter time range compared to other charts.
Estonia → Japan exports vs. same CN group to partner countries — index 2018 = 100
All four charts show Estonian exports — not imports. The question is: did the Japan export channel for each commodity group accelerate after 2019 in a way that other markets did not? Japan’s red line is the EPA channel; grey dashed lines are major non-EPA or pre-existing markets (Sweden, Finland, Germany, South Korea, USA). When Japan diverges upward after 2019 while other countries stay near 100, that acceleration is Japan-specific — not a global sector boom, and consistent with an EPA tariff effect.
CN44 Wood — Japan vs. partner countries
Japan peaked in 2022 at index 184 (75 M€ vs 41 M€ in 2018) then fell sharply — driven by the Russian airspace closure, not EPA reversal. Sweden and Germany also rose in 2022 (global timber demand), confirming the 2022 peak is partly structural. The post-2019 Japan ramp is steeper than peers.
CN85 Electronics — Japan vs. partner countries
Japan’s 2022 peak (index 339) is dramatically above all peers. Sweden and Finland stayed near 100–140; Germany declined then recovered. Japan-specific acceleration post-2019 is the clearest EPA signal in this dataset.
CN84 Machinery — Japan vs. partner countries
Japan spiked to index 392 in 2019 (large one-year shipment) then normalised. The mean-based comparison still shows +520% growth because the pre-EPA base was tiny (0.6 M€). Other markets show no similar spike — this was Japan-specific.
CN71 Precious metals — Japan vs. partner countries
CN71 is the most dramatic EPA-created flow: Japan index reaches ~9,700 by 2022 (12.5 M€ vs 2018 base of 0.13 M€) because the 2018 base was near-zero (37k EUR). Switzerland and Hong Kong were already active precious metals buyers before 2019 — Japan was not. The EPA opened that channel.
COVID, war, and logistics
COVID-19 caused a brief dip in 2020 — smaller than expected. Wood exports proved resilient, which makes sense: construction timber demand didn't collapse the way consumer discretionary spending did. The pandemic is visible in the data but not dramatic.
The war is a different story. The closure of Russian airspace in February 2022 hit Japan harder than almost any other trade partner, for one simple reason: distance. Air cargo to Japan was already operating near the edge of economic viability for many product categories. The Russian airspace route shaved 3–4 hours off a round trip — enough to make the difference between viable and marginal for some shipments. After closure, round-trip cargo times roughly doubled; maritime alternatives add 4–6 weeks and change the economics entirely.
This is the key confounding factor in reading the post-2022 data. The EPA is still in force. Tariffs are still lower. But the logistics infrastructure that would allow exporters to take advantage of those lower tariffs is under structural strain that has nothing to do with the trade agreement.
There's an uncomfortable inversion here: the same logistics barriers that are suppressing overall trade volumes are simultaneously raising the competitive moat for those who have already built the capability to navigate them. Declining volumes mean fewer active competitors, not a worse market — for the right type of goods.
Conclusions
EPA worked — moderately and selectively
Post-EPA average exports +40% vs pre-EPA (36 M€ → 51 M€/year in the clean 2019–2022 window). Real growth in wood (+50% mean, +22% CAGR), electrical machinery (+58% mean, reversed from declining), chemicals (+24% mean, accelerating to 6.6 M€ by 2020–2024). Consistent with policy intent — but concentrated in groups that already had trade relationships before 2019.
Pareto concentration confirms EPA signal
4 Group A exports (CN44, CN85, CN71, CN84) carry ~77% of the composite score — and all 4 grew substantially post-EPA. Group B (5 further groups) adds secondary growth. The 2 spike groups (CN03, CN81) reacted in 2019 but showed no structural follow-through (k < 1). This pattern — concentrated in tariff-covered groups, absent in structurally untouched ones — is consistent with an agreement-driven effect.
Logistics beats tariffs post-2022
Russia's airspace closure hit Japan harder than any other major partner. Trade peaked at 150 M€ in 2022, then fell even as the EPA remained fully in force. The agreement works — but only when logistics allow exporters to act on it.
CAGR tells a different story than mean values
CN85 Electronics and CN28 Chemicals show negative YoY in 2019 — but their CAGR flipped from negative (pre-EPA: −8.7%, −2.9%) to strongly positive (post-EPA: +18.7%, +11.4%). The EPA effect for these groups is not a one-year jump but a multi-year reversal of direction.
Japan ≠ Asia
South Korea grew while Japan declined post-2022. Distance, routing, and freight infrastructure are country-specific. Treating Asian markets as a single block leads to wrong conclusions about where trade agreements are working.
The Korea precedent
EU–Korea EPA (2011) took 3–5 years to show full effect. Japan is at year 5–6. The trajectory isn't necessarily finished — but further progress requires the logistics situation to stabilise first.
Counter-cases matter
Fish exports fell 48% post-EPA. Dairy barely moved. The EPA removes tariffs — it does not remove competition from other EU exporters who now have the same access. Where Estonia isn't competitive on quality or price, lower tariffs change nothing.
Current conditions = selective opportunity
Declining volumes mean fewer active competitors for those who can navigate the logistics barrier. Niche, high-value, non-time-sensitive goods are best positioned to benefit from lower tariffs without being penalised by higher freight costs.
The EPA worked — trade grew in the right sectors and in the right direction. The Pareto composite score makes that hard to attribute to coincidence: growth is concentrated in exactly the groups where tariff cuts were largest, and absent in groups with minimal EPA exposure. But "working" and "transformative" are different things. Japan remains a small but strategically interesting partner, and the current combination of lower tariffs and higher logistics barriers may create better entry opportunities than the headline numbers suggest — for exporters who can operate at that distance.