RQ4 — is energy trade a structurally different market? Same 35 economies, only HS chapter 27 (coal, crude, refined products, gas, electricity). Compare concentration, who leads, and where Austria sits against the all-merchandise network.
Why the node set is held fixed, and the caveats that matter for energy (2022 prices, suppliers outside the frame, pipeline-gas misattribution), are on the Data & method page.
Bridge. HS-2716 electrical energy is the customs view of the same physical flow that austria-energy-analysis measures hourly from ENTSO-E.
The energy subgraph keeps every economy but only the HS-27 flows between them.
all_edges = data_loader.build_edgelist()
energy_edges = data_loader.build_edgelist(products=config.HS_ENERGY)
G = graph.build_graph(all_edges)
E = graph.build_graph(energy_edges)
for label, network in zip(LABELS, (G, E), strict=True):
stats = graph.graph_summary(network)
print(
f"{label:>16}: {int(stats['n_nodes'])} economies, {int(stats['n_edges']):>5} links, "
f"density = {stats['density']:.2f}, "
f"{stats['total_value_kusd'] / 1e6:>6,.0f} bn USD"
)Shares and ranks are computed the same way for both networks, so every row is like-for-like. The Herfindahl index is the sum of squared export shares; its reciprocal is the number of equal-sized exporters that would be as concentrated as what we observe.
Left: how fast cumulative export share piles up as you walk down the exporter ranking — the steeper curve is the more concentrated market. Right: each economy’s share of energy exports against its share of merchandise exports. The dotted diagonal is “energy share = merchandise share”, so everything above it is an economy that punches above its trade weight in energy.
HS-27 is not one commodity. Splitting it by 4-digit heading shows what the network trades and what Austria — with no oil or gas of its own — buys and sells.
flows = energy.load_product_flows()
network_mix = energy.heading_composition(flows)
aut_imports_mix = energy.heading_composition(flows, iso3=FOCUS, direction="imports")
aut_exports_mix = energy.heading_composition(flows, iso3=FOCUS, direction="exports")
mix = (
network_mix[["hs", "heading", "share"]]
.rename(columns={"share": "network"})
.merge(
aut_imports_mix[["hs", "share"]].rename(columns={"share": f"{FOCUS} imports"}),
on="hs",
how="left",
)
.merge(
aut_exports_mix[["hs", "share"]].rename(columns={"share": f"{FOCUS} exports"}),
on="hs",
how="left",
)
.head(6)
.fillna(0.0)
)
mix.style.format(
{"network": "{:.1%}", f"{FOCUS} imports": "{:.1%}", f"{FOCUS} exports": "{:.1%}"}
).hide(axis="index")Who Austria buys HS-27 from and who it sells to, as a share of its own energy imports / exports inside this node set.
def _partner_shares(edges: pd.DataFrame, side: str, partner: str) -> pd.Series:
"""Partner shares of Austria's energy trade on one side of the edge."""
own = edges.loc[edges[side] == FOCUS]
totals = own.groupby(partner)["value_kusd"].sum().sort_values(ascending=False)
return totals / totals.sum()
suppliers = _partner_shares(energy_edges, "importer_iso3", "exporter_iso3")
customers = _partner_shares(energy_edges, "exporter_iso3", "importer_iso3")
top_partners = pd.DataFrame(
{
"rank": range(1, 7),
"supplier": suppliers.index[:6],
"share of AUT energy imports": suppliers.to_numpy()[:6],
"customer": customers.index[:6],
"share of AUT energy exports": customers.to_numpy()[:6],
}
).set_index("rank")
top_partners.style.format(
{"share of AUT energy imports": "{:.1%}", "share of AUT energy exports": "{:.1%}"}
)Intra-set import shares for merchandise vs energy (the blind spot quantified). The interpretation is on Data & method.
sourcing = energy.import_sourcing(focus_iso3=FOCUS, labels=LABELS)
sourcing_view = sourcing.assign(
imports_world_bn=sourcing["imports_world_kusd"] / 1e6,
imports_within_bn=sourcing["imports_within_kusd"] / 1e6,
)[["scope", "importer", "imports_world_bn", "imports_within_bn", "intra_share"]]
sourcing_view.style.format(
{"imports_world_bn": "{:,.0f}", "imports_within_bn": "{:,.0f}", "intra_share": "{:.0%}"}
).hide(axis="index")Generated from the run above, so the numbers always match the tables and the figure.
total_conc = energy.concentration(total_shares, top_k=TOP_K)
energy_conc = energy.concentration(energy_shares, top_k=TOP_K)
total_pos = energy.country_position(total_shares, FOCUS)
energy_pos = energy.country_position(energy_shares, FOCUS)
n_nodes = int(energy_pos["n_economies"])
total_value = float(all_edges["value_kusd"].sum())
energy_value = float(energy_edges["value_kusd"].sum())
top_energy = ", ".join(energy_shares["iso3"].head(TOP_K))
top_total = ", ".join(total_shares["iso3"].head(TOP_K))
def _share(mix_frame: pd.DataFrame, hs: str) -> float:
"""Share of one HS heading in a composition table (0.0 if absent)."""
row = mix_frame.loc[mix_frame["hs"] == hs, "share"]
return float(row.iloc[0]) if len(row) else 0.0
def _value(mix_frame: pd.DataFrame, hs: str) -> float:
"""Value (thousand USD) of one HS heading in a composition table."""
row = mix_frame.loc[mix_frame["hs"] == hs, "value_kusd"]
return float(row.iloc[0]) if len(row) else 0.0
elec_network = _share(network_mix, "2716")
elec_in, elec_out = _share(aut_imports_mix, "2716"), _share(aut_exports_mix, "2716")
gas_out, refined_out = _share(aut_exports_mix, "2711"), _share(aut_exports_mix, "2710")
electricity = flows.loc[flows["hs"] == "2716"]
elec_customers = ", ".join(
electricity.loc[electricity["exporter_iso3"] == FOCUS]
.groupby("importer_iso3")["value_kusd"]
.sum()
.sort_values(ascending=False)
.index[:4]
)
gas_in_bn = _value(aut_imports_mix, "2711") / 1e6
gas_out_bn = _value(aut_exports_mix, "2711") / 1e6
elec_in_bn = _value(aut_imports_mix, "2716") / 1e6
elec_out_bn = _value(aut_exports_mix, "2716") / 1e6
intra = sourcing.set_index(["scope", "importer"])["intra_share"]
node_set = f"{n_nodes}-economy set"
paragraphs = [
f"**A different market.** Energy is {energy_value / total_value:.0%} of the value in this "
f"network but a structurally different market. The top {TOP_K} exporters carry "
f"**{energy_conc['top_k_share']:.0%} of energy exports** ({top_energy}) against "
f"{total_conc['top_k_share']:.0%} for merchandise as a whole ({top_total}); the Herfindahl "
f"index is {energy_conc['hhi']:.3f} vs {total_conc['hhi']:.3f}, i.e. "
f"{energy_conc['effective_exporters']:.0f} effective suppliers instead of "
f"{total_conc['effective_exporters']:.0f}. The leaders change too: the manufacturing hubs "
f"that dominate merchandise trade cannot sell resources they do not have, so resource "
f"exporters take their place — Russia and Norway sit far above the diagonal in the "
f"right-hand panel, Germany and China well below it. Concentration, not connectivity, is "
f"again the risk — the energy graph is nearly complete (density "
f"{float(len(energy_edges)) / (n_nodes * (n_nodes - 1)):.2f}), so everyone trades with "
f"everyone; the volume just sits on very few sources.",
f"**Caveat on the year.** {config.YEAR} is the energy-crisis year: BACI values are USD, so "
f"the energy subnetwork's {energy_value / total_value:.0%} value share reflects prices as "
f"much as volumes, and Russia still leads the {config.YEAR} table despite sanctions that "
f"were only partly in force. Read the concentration ranking as a snapshot of a market under "
f"stress, not a structural constant.",
f"**Austria.** Austria is {int(energy_pos['export_rank'])}th of {n_nodes} by energy exports "
f"({energy_pos['export_share']:.1%} of the network's energy) and "
f"{int(energy_pos['import_rank'])}th by energy imports — versus "
f"{int(total_pos['export_rank'])}th and {int(total_pos['import_rank'])}th for merchandise. "
f"It is a net energy importer ({energy_pos['net_export_kusd'] / 1e6:+,.1f} bn USD). The "
f"ranking flatters it: Austria produces almost no fossil fuel, so its energy exports are "
f"re-exports and transit — {elec_out:.0%} electricity, {gas_out:.0%} gas and "
f"{refined_out:.0%} refined products passed on to neighbours. Its imports are equally "
f"intermediated: {suppliers.iloc[0]:.0%} arrive from {suppliers.index[0]}, which is a "
f"routing fact, not a production one.",
f"**The frame misses the real suppliers.** Only {intra[(LABELS[1], node_set)]:.0%} of this "
f"node set's energy imports come from inside it, against "
f"{intra[(LABELS[0], node_set)]:.0%} for merchandise; for Austria it is "
f"{intra[(LABELS[1], FOCUS)]:.0%} vs {intra[(LABELS[0], FOCUS)]:.0%}. Europe's crude and LNG "
f"largely originate outside the {n_nodes} economies modelled here (the Gulf, Kazakhstan, "
f"North and West Africa). The energy subgraph therefore describes how energy is "
f"*redistributed* within Europe, not where it ultimately comes from — a full dependence "
f"analysis needs the world node set.",
f"**Pipeline gas is badly captured — do not read dependence off it.** Austria's recorded "
f"HS-2711 (gas) imports inside this node set are only {gas_in_bn:,.1f} bn USD while its gas "
f"exports are {gas_out_bn:,.1f} bn, which would make Austria a net gas exporter in "
f"{config.YEAR}. It was not: it burned mostly Russian pipeline gas. Customs statistics "
f"attribute a flow to the declared consignment partner, so gas crossing hubs and traded "
f"contracts lands on the wrong edge. Treat merchandise data as evidence about *value "
f"routed*, and use physical flow data for dependence.",
f"**Bridge to [austria-energy-analysis](https://github.com/fatemeh-studio/"
f"austria-energy-analysis).** Electricity (HS-2716) is {elec_network:.0%} of the energy "
f"network's value but {elec_out:.0%} of Austria's energy exports and {elec_in:.0%} of its "
f"energy imports — {elec_out_bn:,.1f} bn USD out against {elec_in_bn:,.1f} bn in, an almost "
f"perfectly balanced position with its neighbours ({elec_customers}). That is the same "
f"physical flow the sibling project measures hourly from ENTSO-E: there, Austria's renewable "
f"share of electricity reaches 86% (2024), gas generation falls by a third, and each GW of "
f"wind+solar cuts the day-ahead price by 14.3 EUR/MWh. Read together: this project locates "
f"Austria in Europe's energy exchange (mid-sized, intermediating, import-dependent), the "
f"other one explains what the exchange is for.",
]
display(Markdown("\n\n".join(paragraphs)))