pipeline/src/isora_hf/panel.py

"""Consolidated analysis-ready panel: one row per jurisdiction × fiscal year, headline indicators
as columns with short names, values drawn only from questionnaire generations where the question
is the same. The mapping below is the curated crosswalk; `panel_dictionary` publishes it."""

from __future__ import annotations

from dataclasses import dataclass, field

import pandas as pd

G16, G18, G20 = "ISORA 2016", "ISORA 2018", "ISORA 2020+"
GENERATIONS = (G16, G18, G20)

PCT, RATIO, COUNT, LCU, PERSONS = "percent", "ratio", "count", "local currency units", "persons"


@dataclass(frozen=True)
class PanelColumn:
    name: str
    unit: str
    description: str
    codes: dict[str, str] = field(default_factory=dict)  # generation -> indicator code
    topic: str = ""


def _col(name, unit, description, topic, g16=None, g18=None, g20=None) -> PanelColumn:
    codes = {g: c for g, c in ((G16, g16), (G18, g18), (G20, g20)) if c}
    return PanelColumn(name, unit, description, codes, topic)


PANEL_COLUMNS: tuple[PanelColumn, ...] = (
    # --- Revenue and resources
    _col(
        "net_revenue_lcu",
        LCU,
        "Total net revenue collected by the tax administration (incl. SSC and non-tax revenue where collected)",
        "revenue",
        "80040_3",
        "80040_3",
        "80040_3",
    ),
    _col(
        "gdp_lcu",
        LCU,
        "Gross domestic product (as reported to ISORA)",
        "revenue",
        None,
        None,
        "398_001",
    ),
    _col(
        "government_revenue_lcu",
        LCU,
        "Total government revenue (as reported to ISORA)",
        "revenue",
        None,
        None,
        "398_005",
    ),
    _col(
        "population",
        PERSONS,
        "Total population (as reported to ISORA)",
        "revenue",
        None,
        None,
        "398_003",
    ),
    _col(
        "labor_force",
        PERSONS,
        "Labor force (as reported to ISORA)",
        "revenue",
        None,
        None,
        "398_004",
    ),
    _col(
        "revenue_to_gdp_pct",
        PCT,
        "Net revenue collected by the tax administration as % of GDP",
        "revenue",
        "10790",
        "337_001",
        "337_001",
    ),
    _col(
        "tax_incl_ssc_to_gdp_pct",
        PCT,
        "Tax collected including social security contributions as % of GDP",
        "revenue",
        None,
        "337_002",
        "337_002",
    ),
    _col(
        "tax_excl_ssc_to_gdp_pct",
        PCT,
        "Tax collected excluding social security contributions as % of GDP",
        "revenue",
        None,
        "337_003",
        "337_003",
    ),
    _col(
        "revenue_to_government_revenue_pct",
        PCT,
        "Net revenue collected as % of total government revenue",
        "revenue",
        None,
        None,
        "337_168",
    ),
    _col(
        "pit_share_of_revenue_pct",
        PCT,
        "Personal income tax as % of total revenue collected",
        "revenue",
        "80080_4",
        "337_005",
        "337_005",
    ),
    _col(
        "cit_share_of_revenue_pct",
        PCT,
        "Corporate income tax as % of total revenue collected",
        "revenue",
        "80090_4",
        "337_006",
        "337_006",
    ),
    _col(
        "vat_share_of_revenue_pct",
        PCT,
        "VAT as % of total revenue collected",
        "revenue",
        "80130_4",
        "337_007",
        "337_007",
    ),
    _col(
        "ssc_share_of_revenue_pct",
        PCT,
        "Social security contributions as % of total revenue collected",
        "revenue",
        "80240_4",
        "337_008",
        "337_008",
    ),
    _col(
        "other_taxes_share_of_revenue_pct",
        PCT,
        "Other taxes as % of total revenue collected",
        "revenue",
        None,
        "337_009",
        "337_009",
    ),
    _col(
        "nontax_share_of_revenue_pct",
        PCT,
        "Non-tax revenue as % of total revenue collected",
        "revenue",
        "80250_4",
        "337_004",
        "337_004",
    ),
    _col(
        "cost_of_collection_pct",
        PCT,
        "Recurrent (operating) cost of collection: operating expenditure as % of net revenue collected",
        "resources",
        None,
        "337_012",
        "337_012",
    ),
    _col(
        "operating_expenditure_lcu",
        LCU,
        "Operating (recurrent) expenditure of the tax administration",
        "resources",
        "89410",
        "91710_543",
        "337_176",
    ),
    _col(
        "salary_expenditure_lcu",
        LCU,
        "Salary expenditure of the tax administration",
        "resources",
        "89440",
        "91730_543",
        "337_177",
    ),
    _col(
        "ict_expenditure_lcu",
        LCU,
        "ICT operating expenditure of the tax administration",
        "resources",
        "83280_204",
        "91740_543",
        "337_178",
    ),
    _col(
        "capital_expenditure_lcu",
        LCU,
        "Capital expenditure of the tax administration",
        "resources",
        None,
        "91710_544",
        "337_179",
    ),
    _col(
        "salary_share_of_opex_pct",
        PCT,
        "Salary cost as % of operating (recurrent) expenditure",
        "resources",
        "10700",
        "337_013",
        "337_013",
    ),
    _col(
        "ict_share_of_opex_pct",
        PCT,
        "ICT operating cost as % of operating expenditure",
        "resources",
        "10710",
        "337_014",
        "337_014",
    ),
    _col(
        "capex_to_opex_pct",
        PCT,
        "Capital expenditure as % of operating expenditure",
        "resources",
        None,
        None,
        "337_015",
    ),
    _col(
        "total_fte",
        COUNT,
        "Total full-time equivalent staff of the tax administration",
        "staff",
        "83430_206",
        "83430_206",
        "337_180",
    ),
    _col("population_per_fte", RATIO, "Population per FTE", "staff", "10740", "337_010", "337_010"),
    _col(
        "labor_force_per_fte", RATIO, "Labor force per FTE", "staff", "10750", "337_011", "337_011"
    ),
    # --- Staff
    _col(
        "staff_audit_share_pct",
        PCT,
        "% of staff in audit, investigation and other verification",
        "staff",
        "83460_207",
        "94020_207",
        "337_017",
    ),
    _col(
        "staff_debt_collection_share_pct",
        PCT,
        "% of staff in enforced debt collection and related functions",
        "staff",
        "83470_207",
        "94030_207",
        "337_018",
    ),
    _col(
        "staff_hq_share_pct", PCT, "% of staff in headquarters", "staff", None, "337_022", "337_022"
    ),
    _col(
        "hiring_rate_pct",
        PCT,
        "Recruitments in FY as % of staff",
        "staff",
        "10020",
        "337_028",
        "337_028",
    ),
    _col(
        "attrition_rate_pct",
        PCT,
        "Departures in FY as % of staff",
        "staff",
        "10010",
        "337_029",
        "337_029",
    ),
    _col(
        "staff_female_pct", PCT, "% of staff who are female", "staff", "10200", "337_041", "337_041"
    ),
    _col(
        "executives_female_pct",
        PCT,
        "% of executives who are female",
        "staff",
        "10220",
        "337_042",
        "337_042",
    ),
    _col(
        "staff_bachelor_pct",
        PCT,
        "% of staff with a bachelor's degree (or equivalent)",
        "staff",
        "10080",
        "337_043",
        "337_043",
    ),
    _col(
        "staff_master_or_higher_pct",
        PCT,
        "% of staff with a master's degree or higher (or equivalent)",
        "staff",
        "10070",
        "337_044",
        "337_044",
    ),
    _col(
        "staff_under_35_pct",
        PCT,
        "% of staff younger than 35 (sum of <25 and 25-34 bands)",
        "staff",
        None,
        None,
        None,
    ),
    _col(
        "staff_55_or_older_pct",
        PCT,
        "% of staff aged 55 or older (sum of 55-64 and >64 bands)",
        "staff",
        None,
        None,
        None,
    ),
    # --- Segmentation, registration, filing
    _col(
        "lto_fte_share_pct",
        PCT,
        "FTEs in the large taxpayer office/program as % of total FTEs",
        "segmentation",
        "10760",
        "337_045",
        "337_045",
    ),
    _col(
        "lto_revenue_share_pct",
        PCT,
        "Net revenue administered by the large taxpayer office/program as % of total net revenue",
        "segmentation",
        "10550",
        None,
        "92280_28_1",
    ),
    _col(
        "lto_corporate_taxpayers_share_pct",
        PCT,
        "Corporate taxpayers managed by the LTO/program as % of active corporate taxpayers",
        "segmentation",
        "10240",
        "337_046",
        "337_046",
    ),
    _col(
        "active_pit_taxpayers_pct_labor_force",
        PCT,
        "Active PIT taxpayers as % of labor force",
        "registration",
        "10770",
        "337_059",
        "337_059",
    ),
    _col(
        "active_pit_taxpayers_pct_population",
        PCT,
        "Active PIT taxpayers as % of population",
        "registration",
        "10780",
        "337_058",
        "337_058",
    ),
    _col(
        "inactive_pit_register_pct",
        PCT,
        "Inactive taxpayers as % of PIT register",
        "registration",
        None,
        "337_053",
        "337_053",
    ),
    _col(
        "inactive_cit_register_pct",
        PCT,
        "Inactive taxpayers as % of CIT register",
        "registration",
        None,
        "337_054",
        "337_054",
    ),
    _col(
        "inactive_vat_register_pct",
        PCT,
        "Inactive taxpayers as % of VAT register",
        "registration",
        None,
        "337_055",
        "337_055",
    ),
    _col(
        "active_taxpayers_cit",
        COUNT,
        "Number of active CIT taxpayers",
        "registration",
        None,
        None,
        "95860_37",
    ),
    _col(
        "active_taxpayers_pit",
        COUNT,
        "Number of active PIT taxpayers",
        "registration",
        None,
        None,
        "95860_38",
    ),
    _col(
        "active_taxpayers_vat",
        COUNT,
        "Number of active VAT taxpayers",
        "registration",
        None,
        None,
        "95860_39",
    ),
    _col(
        "active_taxpayers_paye",
        COUNT,
        "Number of active PAYE (employer withholding) taxpayers",
        "registration",
        None,
        None,
        "95860_40",
    ),
    _col(
        "on_time_filing_cit_pct",
        PCT,
        "CIT returns filed on time as % of returns expected",
        "filing",
        "88140_37",
        "88140_37",
        "88140_37",
    ),
    _col(
        "on_time_filing_pit_pct",
        PCT,
        "PIT returns filed on time as % of returns expected",
        "filing",
        "88140_38",
        "88140_38",
        "88140_38",
    ),
    _col(
        "on_time_filing_vat_pct",
        PCT,
        "VAT returns filed on time as % of returns expected",
        "filing",
        None,
        "88140_39",
        "88140_39",
    ),
    _col(
        "on_time_filing_paye_pct",
        PCT,
        "PAYE returns filed on time as % of returns expected",
        "filing",
        "88140_40",
        "88140_40",
        "88140_40",
    ),
    _col(
        "efiling_cit_pct",
        PCT,
        "CIT returns filed electronically as % of returns received",
        "filing",
        None,
        None,
        "111_200",
    ),
    _col(
        "efiling_pit_pct",
        PCT,
        "PIT returns filed electronically as % of returns received",
        "filing",
        None,
        None,
        "111_201",
    ),
    _col(
        "efiling_vat_pct",
        PCT,
        "VAT returns filed electronically as % of returns received",
        "filing",
        None,
        None,
        "111_202",
    ),
    # --- Payment and arrears
    _col(
        "on_time_payment_cit_pct",
        PCT,
        "CIT payments received on time as % of payments due",
        "payment",
        "10430",
        "337_085",
        "337_085",
    ),
    _col(
        "on_time_payment_pit_pct",
        PCT,
        "PIT payments received on time as % of payments due",
        "payment",
        "10440",
        "337_084",
        "337_084",
    ),
    _col(
        "on_time_payment_vat_pct",
        PCT,
        "VAT payments received on time as % of payments due",
        "payment",
        "10460",
        "337_087",
        "337_087",
    ),
    _col(
        "on_time_payment_paye_pct",
        PCT,
        "PAYE payments received on time as % of payments due",
        "payment",
        "10450",
        "337_086",
        "337_086",
    ),
    _col(
        "epayment_by_number_pct",
        PCT,
        "Electronic payments as % of payments (by number)",
        "payment",
        None,
        "337_090",
        "337_090",
    ),
    _col(
        "epayment_by_value_pct",
        PCT,
        "Electronic payments as % of payments (by value)",
        "payment",
        None,
        "337_091",
        "337_091",
    ),
    _col(
        "arrears_to_revenue_pct",
        PCT,
        "Closing stock of arrears at year end as % of total revenue collected",
        "arrears",
        "10560",
        "337_092",
        "337_092",
    ),
    _col(
        "collectable_arrears_share_pct",
        PCT,
        "Collectable arrears as % of closing stock of arrears",
        "arrears",
        None,
        "337_093",
        "337_093",
    ),
    _col(
        "cit_arrears_pct_of_cit_collected",
        PCT,
        "CIT arrears as % of CIT collected",
        "arrears",
        None,
        "337_094",
        "337_094",
    ),
    _col(
        "pit_arrears_pct_of_pit_collected",
        PCT,
        "PIT arrears as % of PIT collected",
        "arrears",
        None,
        "337_095",
        "337_095",
    ),
    _col(
        "vat_arrears_pct_of_vat_collected",
        PCT,
        "VAT arrears as % of VAT collected",
        "arrears",
        None,
        "337_097",
        "337_097",
    ),
    _col(
        "arrears_growth_excl_noncollectable_pct",
        PCT,
        "Year-end arrears as % of previous year-end arrears (excluding non-collectable)",
        "arrears",
        None,
        None,
        "337_102",
    ),
    # --- Audit and disputes
    _col(
        "audit_assessments_to_collections_pct",
        PCT,
        "Additional assessments from all audits and verification actions as % of tax collections",
        "audit",
        None,
        None,
        "337_158",
    ),
    _col(
        "audit_hit_rate_pct",
        PCT,
        "Audits resulting in an adjustment as % of audits completed",
        "audit",
        None,
        None,
        "337_173",
    ),
    _col(
        "cit_assessments_pct_of_cit_collected",
        PCT,
        "CIT additional assessments as % of CIT collected",
        "audit",
        None,
        "337_121",
        "337_121",
    ),
    _col(
        "pit_assessments_pct_of_pit_collected",
        PCT,
        "PIT additional assessments as % of PIT collected",
        "audit",
        None,
        "337_122",
        "337_122",
    ),
    _col(
        "vat_assessments_pct_of_vat_collected",
        PCT,
        "VAT additional assessments as % of VAT collected",
        "audit",
        None,
        "337_124",
        "337_124",
    ),
    _col(
        "internal_review_cases_per_1000_taxpayers",
        RATIO,
        "Internal review (administrative review) cases initiated per 1 000 active PIT and CIT taxpayers",
        "disputes",
        "10650",
        "337_125",
        "337_125",
    ),
    _col(
        "independent_review_to_internal_review_ratio",
        RATIO,
        "Cases under independent review relative to internal review cases",
        "disputes",
        None,
        None,
        "337_126",
    ),
    _col(
        "appeals_won_by_administration_pct",
        PCT,
        "Cases resolved by higher appellate court in favour of the administration as % of cases resolved",
        "disputes",
        None,
        None,
        "337_127",
    ),
)

# Columns computed as sums of published components (generation -> list of codes).
COMPOSITE_COLUMNS: dict[str, dict[str, list[str]]] = {
    "staff_under_35_pct": {
        G16: ["10090", "10100"],
        G18: ["337_031", "337_032"],
        G20: ["337_031", "337_032"],
    },
    "staff_55_or_older_pct": {
        G16: ["10130", "10140"],
        G18: ["337_035", "337_036"],
        G20: ["337_035", "337_036"],
    },
}

# Ratios that cannot be zero for an operating tax administration. The ISORA 2016 derived
# indicators in particular publish an exact 0 when one of the inputs was not reported.
IMPLAUSIBLE_ZERO_COLUMNS: frozenset[str] = frozenset(
    {"revenue_to_gdp_pct", "population_per_fte", "labor_force_per_fte", "salary_share_of_opex_pct"}
)

ATTRIBUTE_COLUMNS = [
    "jurisdiction_code",
    "jurisdiction_name",
    "fiscal_year",
    "survey_round",
    "questionnaire_generation",
    "imf_region",
    "world_bank_region",
    "income_group_wb",
    "income_group_wb_classification_fy",
    "income_group_wb_current",
    "member_oecd",
    "member_eu",
    "member_iota",
    "member_ciat",
    "member_adb",
]


def _values(obs: pd.DataFrame, generation: str, code: str, unit: str) -> pd.Series:
    sub = obs[
        (obs["questionnaire_generation"] == generation)
        & (obs["indicator_code"] == code)
        & (obs["value_status"] == "value")
    ]
    col = "value_local_currency_units" if unit == LCU else "value_numeric"
    return sub.set_index(["jurisdiction_code", "fiscal_year"])[col].dropna()


def build_panel(
    obs: pd.DataFrame, jurisdictions: pd.DataFrame, income_by_year: pd.Series
) -> tuple[pd.DataFrame, pd.DataFrame]:
    base = (
        obs[
            [
                "jurisdiction_code",
                "jurisdiction_name",
                "fiscal_year",
                "survey_round",
                "questionnaire_generation",
            ]
        ]
        .drop_duplicates(["jurisdiction_code", "fiscal_year"])
        .sort_values(["jurisdiction_code", "fiscal_year"])
        .reset_index(drop=True)
    )
    attrs = jurisdictions.set_index("jurisdiction_code")
    base["imf_region"] = base["jurisdiction_code"].map(attrs["imf_region"])
    base["world_bank_region"] = base["jurisdiction_code"].map(attrs["world_bank_region_current"])
    base["income_group_wb"] = income_by_year.reindex(
        pd.MultiIndex.from_frame(base[["jurisdiction_code", "fiscal_year"]])
    ).to_numpy()
    base["income_group_wb_classification_fy"] = ["FY" + str(y + 2)[2:] for y in base["fiscal_year"]]
    base["income_group_wb_current"] = base["jurisdiction_code"].map(
        attrs["world_bank_income_group_current"]
    )
    for m in ("member_oecd", "member_eu", "member_iota", "member_ciat", "member_adb"):
        base[m] = base["jurisdiction_code"].map(attrs[m])
    base = base.set_index(["jurisdiction_code", "fiscal_year"])

    dictionary_rows = []
    labels = obs.drop_duplicates(["questionnaire_generation", "indicator_code"]).set_index(
        ["questionnaire_generation", "indicator_code"]
    )["indicator_label"]
    for column in PANEL_COLUMNS:
        series = []
        sources = COMPOSITE_COLUMNS.get(column.name)
        for gen in GENERATIONS:
            if sources:
                codes = sources.get(gen, [])
                parts = [_values(obs, gen, c, column.unit) for c in codes]
                if not parts:
                    continue
                combined = pd.concat(parts, axis=1).dropna().sum(axis=1)
                series.append(combined)
                code_txt = " + ".join(codes)
                label_txt = " + ".join(labels.get((gen, c), "") for c in codes)
            else:
                code = column.codes.get(gen)
                if not code:
                    continue
                series.append(_values(obs, gen, code, column.unit))
                code_txt, label_txt = code, labels.get((gen, code), "")
            dictionary_rows.append(
                {
                    "column": column.name,
                    "topic": column.topic,
                    "unit": column.unit,
                    "description": column.description,
                    "questionnaire_generation": gen,
                    "indicator_code": code_txt,
                    "source_label": label_txt,
                    "value_source": "value_local_currency_units"
                    if column.unit == LCU
                    else "value_numeric",
                    "zeros_treated_as_missing": column.name in IMPLAUSIBLE_ZERO_COLUMNS,
                }
            )
        if series:
            stacked = pd.concat(series)
            stacked = stacked[~stacked.index.duplicated(keep="last")]
            if column.name in IMPLAUSIBLE_ZERO_COLUMNS:
                stacked = stacked.where(stacked != 0)
            base[column.name] = stacked.reindex(base.index).to_numpy()
        else:
            base[column.name] = pd.Series(dtype="float64")
    panel = base.reset_index()
    counts = {c.name: int(panel[c.name].notna().sum()) for c in PANEL_COLUMNS}
    dictionary = pd.DataFrame(dictionary_rows)
    dictionary["n_non_null_in_panel"] = dictionary["column"].map(counts)
    return panel[ATTRIBUTE_COLUMNS + [c.name for c in PANEL_COLUMNS]], dictionary