Skip to content
GREEN FUTURETECH.
Explore

Research libraryForecasting From a Cold Start

Chapter 04 of 14 · Part I · Draft web edition

External data can bound a market but cannot nominate a winner

Trade, demographic and digital signals are most credible when each is assigned a limited and testable forecasting role.

Sections
7
Exhibits
6
Minutes
31
Sources
52
EX 16 · Apparel eCommerce baseline 2023 (R26): 20.3 million US$ by reported or illustrative channel

The cloth is woven from this chapter's 6 exhibits: each weft is one exhibit's own data, normalised; each warp group is a section. The probe reads the exhibit's own value where it crosses. Procedural visualisation, not a measurement.

The chapter as a draft: 7 sections, 4 exhibit wefts, one shuttle.

Lead exhibit · 16Ranked barsillustrativeReported apparel inflows need product and provenance detail
HS 6309 worn clothing inflows (illustrative)76Unrecorded informal inflows (illustrative)42Apparel eCommerce baseline 2023 (R26)20.3HS 62 woven apparel imports (illustrative)14.5HS 61 knitted apparel imports (illustrative)9.2

Read-outHS 6309 worn clothing inflows (illustrative): 76 million US$ by reported or illustrative channel

Method and source

Illustrative diagnostic comparison anchored to the sourced 2023 Cameroon apparel eCommerce revenue baseline of US$20.32 million (R26) and structured using UN Comtrade (S41) and ITC Trade Map (S43) tariff boundaries. The slices for HS 6309 worn clothing (76.0), unrecorded informal inflows (42.0), HS 62 woven apparel (14.5) and HS 61 knitted apparel (9.2) are illustrative values constructed to show boundary separation; they are not observed GFT customs extracts or consumer sell-through totals.

Unit: million US$ by reported or illustrative channel. Status: illustrative. Sources: S41, S43, R26. Illustrative figures are teaching examples, not measured factory results.

The chapter's warpBar: share of the reading · square: an exhibit

Section 4.0 · Opening · 1 min · no exhibit of its own · one movement Go to section ↓

The chapter cloth8 warp groups · one pick per paragraph

This paragraph previews 4.1 · Reported trade flows describe supply, not consumer sell-through

External statistics cannot tell a greenfield apparel manufacturer which shirt, dress or uniform variant will clear its first production lot. Customs tables, national accounts, agricultural balances, search indices and weather records measure aggregates at boundaries far removed from a customer choosing a specific garment in a specific channel at a specific price. A disciplined cold-start forecasting system therefore assigns every external series a bounded structural task before any model is fitted. Reported trade flows help bound formal category inflows and unit-value distributions, but they do not measure consumer sell-through.[S41][S43] Demographic and income series scale multi-year market scenarios and screen candidate analogue markets, but their annual cadence cannot drive weekly stock keeping unit (SKU) replenishment.[S44][S45][S46] Agricultural and industrial cotton statistics describe upstream raw-material availability and policy ambition, not downstream garment demand.[R06][S47] Higher-frequency digital and meteorological series enter only as challenger covariates that must beat simple baselines in strictly chronological, ex-ante launch replay.[S09][S34][S35][S0401]

This separation of duties protects a new venture from a frequent cold-start error outside the factory floor: mistaking a large macro or trade number for product-level commercial proof. Fashion demand combines demand uncertainty, short product life cycles and seasonal shifts even when a firm holds years of historical transactions.[S01][S0402] When a firm holds no sales history of its own, judgemental and prior-building methods are required, and every external proxy must carry an explicit boundary definition, an as-of timestamp and a falsifiable test of predictive value.[S08][S37] In Green Future Tech Limited (GFT) programme work for Cameroon's textile sector, prepared on 31 January 2024 with source refreshes flagged in 2026, external market indicators establish why local industrial capacity is strategically relevant while leaving SKU-level demand commitments to calibrated priors, controlled pilots and audited demand capture.[R06][R07]

4.14 min · 2 sub-steps · 1 exhibit

Reported trade flows describe supply, not consumer sell-through

Related · chapter 037 paragraphs step it

Related exhibit 11 · chapter 03Decision mapillustrativeAn analogue scorecard exposes weak comparisons before modelling
Context needs reviewBroadest candidatesReject or repairProduct-only resemblanceProduct and offer match →Channel and market match →Candidate ACandidate BCandidate CCandidate D

Read-outCandidate A: Product and offer match 75, Channel and market match 70. Quadrant: Broadest candidates

Method and source

Illustrative candidates and scores demonstrate screening only. Horizontal positions use the worked weights for category, occasion, fabric and price. Vertical positions combine channel, geography and timing with reviewer-approved illustrative weights. Scores are not probabilities or observed demand.

Unit: illustrative match score from 0 to 100. Status: illustrative. Sources: S02, S03. Illustrative figures are teaching examples, not measured factory results.

Read this exhibit in its chapter ↗
4.1 · Introduction · paragraph 1 of 7

A customs import table records border crossings of classified merchandise, not retail demand. When a planning team extracts apparel import series from the United Nations Comtrade database, it observes data sourced directly from national authorities and submitted to the United Nations Statistics Division under international merchandise trade conventions.[S41] Those records capture formal consignments declared at customs ports. They do not record garments produced and sold by domestic tailors, stock held in wholesale warehouses between seasons, unrecorded cross-border trade, or the date and price at which a final consumer buys an item. Treating an annual or monthly customs inflow as a direct proxy for retail sell-through confuses wholesale supply entering a territory with consumer demand clearing at the point of sale.[S06][S41]

4.2 Customs classifications aggregate heterogeneous garments and channels

International trade platforms organise apparel flows primarily under Harmonized System (HS) Chapter 61 (articles of apparel and clothing accessories, knitted or crocheted) and Chapter 62 (articles of apparel and clothing accessories, not knitted or crocheted), alongside Chapter 63 for other made-up textile articles and worn clothing. The International Trade Centre (ITC) Trade Map system provides both values and quantities for the trade flows of goods across these tariff headings.[S43] Even at the six-digit subheading level, a single customs line groups garments that differ in silhouette, fabric weight, stitch density, branding, sizing, retail channel and target occasion. New-item forecasting research demonstrates that clothing and footwear demand depends on fine-grained merchandising attributes, selling price, product age and store context.[S02][S03] Because a customs declaration strips away those merchandising attributes, a six-digit HS flow can bound the volume of a broad category entering formal distribution, but it cannot identify which design attributes or price points will win inside that category.[S02][S43]

Market context in Cameroon illustrates why channel and product boundaries must remain explicit before any trade series touches a forecast prior. Inherited sector evidence records that Cameroon's domestic textile player CICAM saw its domestic market share fall from about 85 percent three decades ago to about 5 percent in the baseline period, while informal employment accounts for about 90 percent of total employment across the economy.[R11][R16] At the consumer level, inherited market estimates put Cameroon clothing and footwear expenditure at about US$94.70 a person in 2024, corresponding to about 8.0 pieces a person, while domestic apparel eCommerce revenue was estimated at US$20.32 million in 2023 and projected to reach US$33.61 million by 2027 at an annual growth rate of 13.41 percent.[R26] Each figure describes a different boundary: total personal expenditure, physical piece intake, formal online channel revenue, or domestic mill share. None is identical to customs-reported imports under HS Chapters 61 and 62, and none measures the sell-through velocity of a newly launched local garment line.[R11][R26][S41]

Exhibit 16 contrasts these distinct channel and customs buckets on a common monetary scale, joining the sourced 2023 Cameroon apparel eCommerce baseline with illustrative customs and informal inflow slices to show why an analyst must decompose category supply before setting a market bound.

Exhibit 16 establishes a foundational rule for Part I of the workflow: the largest volume entering a clothing market may sit in worn clothing or informal distribution rather than formal new-apparel customs lines, and even formal HS 61 and HS 62 lines represent landed wholesale shipments rather than consumer purchases.[R16][S41][S43] The exhibit cannot establish addressable demand for a specific Cameroonian factory offer; instead, it forces the commercial team to specify which customs headings, price tiers and distribution channels its proposed garment contests.

4.3 Policy and regional trade rules alter supply ceilings without guaranteeing sales

Trade series also reflect tariff regimes and preferential access rules. In the inherited register, Cameroon's eligibility under the United States African Growth and Opportunity Act (AGOA) was terminated effective 1 January 2020 and remained listed as ineligible for calendar year 2025.[R18] Within Africa, textiles and clothing remained among unresolved tariff categories under the African Continental Free Trade Area (AfCFTA), where 92.3 percent of lines had been agreed excluding vehicles, textiles and clothing.[R23] Continental market-research sources in the register estimate the Africa textile market at about US$37.3 billion in 2025, projected to reach US$49.41 billion by 2030 at 4.73 percent a year in one series, and projected to reach US$72.7 billion by 2035 at 6.9 percent a year with fashion and apparel representing 48.1 percent of the 2024 market in a second vendor series.[R24][R25]

A cold-start team treats these vendor projections and trade-policy rules as scenario boundaries rather than model inputs.[S08][S0403] Vendor market-size reports carry grade-C status and use incompatible definitions, so GFT retains both series without averaging them.[R24][R25] Tariff-line resolution or preferential access changes the competitive ceiling of an export channel, belonging in conditional scenarios rather than in the statistical prior for domestic weekly sell-through.[R18][R23][S08]

4.42 min · 2 sub-steps · 1 exhibit

Mirror asymmetries need a reconciliation rule

Step 1 of 6Partner FOB export

Exhibit 17 · illustrativeMirror reconciliation bridges partner exports and reporter imports

05001,0001,5001,200Partner FOBexport+140CIF freightadd-on−210Transitre-exports−85Clearancetiming lag−45HSreclassification1,000Reporter CIFimport05001,0001,500Partner FOB export1,200CIF freight add-on+140Transit re-exports−210Clearance…−85HS reclassification−45Reporter CIF import1,000

Partner FOB export

Partner FOB export: total 1,200. Running total 1,200 illustrative thousand US$ in bilateral HS 62 trade · illustrative

Even when trade data are used strictly to bound formal inflows, a single customs query rarely yields an unambiguous number.

CIF freight add-on

CIF freight add-on: adds 140. Running total 1,340 illustrative thousand US$ in bilateral HS 62 trade · illustrative

Transit re-exports

Transit re-exports: removes 210. Running total 1,130 illustrative thousand US$ in bilateral HS 62 trade · illustrative

Clearance timing lag

Clearance timing lag: removes 85. Running total 1,045 illustrative thousand US$ in bilateral HS 62 trade · illustrative

HS reclassification

HS reclassification: removes 45. Running total 1,000 illustrative thousand US$ in bilateral HS 62 trade · illustrative

Reporter CIF import

Reporter CIF import: total 1,000. Running total 1,000 illustrative thousand US$ in bilateral HS 62 trade · illustrative

Method and source · Exhibit 17

Illustrative reconciliation walk showing how GFT decomposes bilateral trade asymmetries defined by UN Comtrade (S41, S42) and ITC Trade Map (S43). Starting from an illustrative partner FOB export record of 1,200 thousand US$, adding CIF freight and insurance (+140) and deducting third-country transit routing (-210), year-end customs clearance lag (-85) and tariff-line reclassification (-45) reconciles to the reporter CIF import record of 1,000 thousand US$. Values are illustrative accounting steps, not a measured Cameroon customs audit.

Unit: illustrative thousand US$ in bilateral HS 62 trade. Status: illustrative. Sources: S41, S42, S43. Illustrative figures are teaching examples, not measured factory results.

4.4 · the bodyExhibit 17 above; its figures are illustrative where marked.

Even when trade data are used strictly to bound formal inflows, a single customs query rarely yields an unambiguous number. In UN Comtrade, bilateral trade asymmetries occur when reported exports from country A to country B do not match reported imports to country B from country A.[S42] Where national customs reporting experiences publication lags, incomplete capture or reclassification, partner-reported exports (mirror data) diverge from reporter imports.[S41][S42] Overwriting reporter imports with mirror exports whenever a new extraction runs silently alters historical backtests.[S37][S42]

4.5 Structural causes of bilateral divergence require explicit adjustments

4.5

Before reconciling reporter and partner records, the engineer separates four mechanical causes of bilateral asymmetry:[S42][S43]

¶ 02 · 2 sources

S42S43

  • Valuation basis (CIF versus FOB): Importing countries record cost, insurance and freight (CIF) values, whereas exporting partners record free on board (FOB) values, creating a freight and insurance wedge.[S42]
  • Transit and re-export routing: Consignments shipped through intermediate ports, or landed at Douala for transit to landlocked neighbours, may be attributed differently by origin and consignment rules.[S42]
  • Clearance timing lags: Maritime shipments departing in November or December frequently clear destination customs in January or February, shifting flows across annual boundaries.[S42]
  • Tariff reclassification: An exporter may declare a six-digit HS 61 or HS 62 subheading while the importing authority records mixed lots under another subheading or HS Chapter 63.[S42][S43]
List · 2 sources

S42S43

Exhibit 17 illustrates how these four mechanisms bridge a partner's FOB export total and a reporter's CIF import total.

Exhibit 17 shows why taking a simple average of reporter and partner numbers, or picking whichever number is larger, corrupts the market bound. Because CIF freight works in the opposite direction from transit re-exports and clearance lags, net bilateral discrepancy can be positive or negative depending on the corridor and quarter.[S42] What Exhibit 17 cannot establish on its own is the exact freight ratio or transit share for a corridor without shipping-line and customs-manifest evidence; where those inputs are absent, the discrepancy must remain visible as an uncertainty interval around the inflow bound rather than disappear into a single point estimate.[S08][S42]

¶ 04 · 2 sourcesExhibit 17

S42S08

4.6 Extraction metadata prevents silent backtest rewriting

4.6

National authorities submit late declarations and revisions to UN Comtrade long after the calendar year closes.[S41] If an analyst pulls trade series in 2026 to backtest a 2024 launch origin without recording when each figure was published, the backtest commits temporal data leakage by using revised trade totals unavailable at the decision date.[S37][S41] Every trade extraction in the GFT pipeline therefore locks seven mandatory provenance fields alongside the numeric series:

¶ 05 · 2 sources

S41S37

  1. Reporter jurisdiction and code: The declaring customs territory and its numerical identifier.[S41]
  2. Partner jurisdiction and role: Country of origin, country of consignment or partner export reporter, preserved as separate columns.[S42]
  3. Trade flow and regime: Gross imports, net imports, domestic exports or re-exports.[S41][S42]
  4. HS nomenclature revision: The Harmonized System vintage (such as HS 2017 or HS 2022) used to define the six-digit code.[S41][S43]
  5. Valuation and physical units: Currency basis (CIF or FOB US dollars, or CFA francs at the euro peg of 655.957 XAF per euro) and supplementary quantity units (kilograms, pieces, pairs or dozens).[R32][S43]
  6. Reference period versus publication date: The calendar period of the customs event alongside the release date.[S37][S41]
  7. Snapshot extraction timestamp: The exact as-of date on which GFT froze the table for a forecast version.[S37][S40]
List · 6 sources

S41S42S43R32S37S40

4.73 min · 1 sub-steps · read with a related exhibit

Values and quantities answer different diagnostic questions

Trade databases following the ITC Trade Map convention report both monetary trade values and physical quantities for goods flows.[S43] In apparel diagnostics, value and quantity answer distinct operational questions and fail in different ways. Monetary import value (in US dollars or XAF) reflects total border expenditure at wholesale landed cost, which responds to inflation, exchange-rate movements, valuation rules and shifts between budget and premium garments.[R32][S43] Physical quantity (in net kilograms or supplementary pieces) reflects material volume crossing the border, connecting to spinning, weaving, dyeing and sewing capacity measured in tonnes or pieces a year.[R10][S43]

4.8 Unit-value ratios mix price, weight and product composition

4.8

Dividing reported trade value by reported quantity yields an apparent unit value, such as US dollars per kilogram or per piece.[S43] Commercial teams frequently treat this ratio as a proxy for local selling price or quality tier. That inference is unsafe without independent validation. Customs quantity reporting is often incomplete across supplementary units, grouping multipacks or mixed bale weights under one line.[S43] Even when weight and piece counts are complete, a change in US dollars per kilogram between two periods can reflect a shift in fabric weight (such as lightweight cotton t-shirts versus heavy twill workwear trousers) rather than a change in garment price. Furthermore, a customs CIF unit value measures wholesale landed cost before importer margins, domestic transport, taxes and retail mark-ups.[S03][S43]

¶ 02 · 2 sources

S43S03

Note

This arithmetic demonstrates why GFT rejects single unit-value rankings from trade tables as direct proof of consumer willingness to pay or retail price elasticity.[S02][S03][S43] When an analyst uses Trade Map values and quantities to bound a market, weight in tonnes connects to upstream mill capacity (such as the 12,000 tonnes a year announced for Camtext SA), piece counts connect to per-capita intake benchmarks (such as the 8.0 pieces a person reported for Cameroon in 2024), and monetary values connect to expenditure ceilings (such as the US$94.70 per-person clothing and footwear spend in 2024), each with an explicit conversion range.[R10][R26][S43]

¶ 03 · 5 sources

…mill capacity (such as the 12,000 tonnes a year announced for Camtext SA)R10, R26, S43

S02S03S43R10R26

Key figures from chapter 01 · Exhibit 04Key figuresillustrativeFour questions discipline every cold-start forecast
  • CommitmentGate 1: DecisionWhich lot, purchase or allocation changes if the range shifts?
  • Latent demandGate 2: TargetHow are stockouts, returns and channel timing separated?
  • Prior weightsGate 3: ProvenanceWhich analogues, surveys and judgements form the range?
  • Stop ruleGate 4: ReversalWhat pilot or replay outcome forces a fallback to baseline?

Read-outGate 1: Decision: Commitment. Which lot, purchase or allocation changes if the range shifts?

Method and source

Illustrative governance tiles summarising the four questions required before a cold-start forecast may inform a production commitment. Tiles state governance gates rather than empirical performance metrics.

Unit: mandatory pre-launch contract gates. Status: illustrative. Sources: S08, S14, S33, S36. Illustrative figures are teaching examples, not measured factory results.

Read this exhibit in its chapter ↗

4.93 min · 2 sub-steps · 1 exhibit

Population, income and urbanisation scale priors slowly

Step 1 of 4Year 1

Exhibit 18 · illustrativeSlow market indicators belong above the SKU layer

Population index (S44)100Year 1103Year 2105Year 3108Year 4Real GNI/capita (S45)100Year 1101Year 2103Year 3105Year 4Urban share index (S46)100Year 1101Year 2102Year 3103Year 4eCommerce index (R26)100Year 1113Year 2129Year 3146Year 4
Population index (S44)100Year 1103Year 2105Year 3108Year 4Real GNI/capita (S45)100Year 1101Year 2103Year 3105Year 4Urban share index (S46)100Year 1101Year 2102Year 3103Year 4eCommerce index (R26)100Year 1113Year 2129Year 3146Year 4

Year 1

Year 1: Population index (S44) 100, Real GNI/capita (S45) 100, Urban share index (S46) 100, eCommerce index (R26) 100 (illustrative index (Year 1 = 100)) · illustrative

Macroeconomic and demographic indicators sit at the top of the external evidence hierarchy, above customs categories and far above individual SKUs.

Year 2

Year 2: Population index (S44) 103, Real GNI/capita (S45) 101, Urban share index (S46) 101, eCommerce index (R26) 113 (illustrative index (Year 1 = 100)) · illustrative

Year 3

Year 3: Population index (S44) 105, Real GNI/capita (S45) 103, Urban share index (S46) 102, eCommerce index (R26) 129 (illustrative index (Year 1 = 100)) · illustrative

Year 4

Year 4: Population index (S44) 108, Real GNI/capita (S45) 105, Urban share index (S46) 103, eCommerce index (R26) 146 (illustrative index (Year 1 = 100)) · illustrative

Method and source · Exhibit 18

Illustrative indexed four-year comparison (Year 1 = 100) showing why annual demographic and income series defined by World Bank WDI (SP.POP.TOTL in S44, NY.GNP.PCAP.KD in S45, SP.URB.TOTL.IN.ZS in S46) change too slowly to explain short-cycle apparel variation, whereas channel-specific adoption (calibrated to the sourced 13.41 percent annual growth rate of Cameroon apparel eCommerce from 2023 to 2027 in R26: 100.0, 113.4, 128.6, 145.9) moves far faster. Macro panel trajectories (100.0 to 108.0, 104.5 and 103.3) are illustrative annual paths.

Unit: illustrative index (Year 1 = 100). Status: illustrative. Sources: S44, S45, S46, R26. Illustrative figures are teaching examples, not measured factory results.

4.9 · the bodyExhibit 18 above; its figures are illustrative where marked.

Macroeconomic and demographic indicators sit at the top of the external evidence hierarchy, above customs categories and far above individual SKUs. In Chapter 3, population, income and urbanisation helped screen whether a foreign city or country was an eligible analogue market.[S44][S45][S46] Here, their role as time-series covariates is defined. Because national demographic and income accounts update annually, incorporate census revisions and describe whole-population averages, they can scale multi-year market priors and capacity scenarios, but they cannot explain week-to-week demand variation across styles, colours and sizes.[S01][S44][S45][S46][S0402]

4.10 Indicator definitions restrict what annual macro series can represent

4.10

Each of the three World Development Indicators (WDI) carries a statistical definition that limits its forecasting role:

¶ 02
  • Total population (SP.POP.TOTL): The World Bank defines total population on the de facto definition, counting all residents regardless of legal status or citizenship as midyear estimates.[S44] While SP.POP.TOTL supplies a resident-count denominator for converting per-person intake estimates (such as 8.0 pieces or US$94.70 per person in R26) into a national volume envelope, it treats infants, rural subsistence households and urban salaried buyers as equal units.[R26][S44]
  • GNI per capita in constant 2015 US dollars (NY.GNP.PCAP.KD): Gross national income measures total income earned by residents within an economic territory during an accounting period, divided by midyear population and adjusted to constant 2015 prices.[S45] It does not measure household disposable income within a retail catchment or supply a clothing-specific income elasticity.[S45] In Cameroon, inherited evidence shows wide income dispersion and informality: the statutory minimum wage (SMIG) rose by 5 percent from XAF 41,875 a month in 2023 to XAF 43,969 a month (about US$76) from 23 February 2024, whereas average monthly salary across sectors was reported at about US$359 a month and informal employment stood near 90 percent.[R14][R15][R16] A national mean like NY.GNP.PCAP.KD cannot tell a planner whether a garment priced for formal earners will clear in a specific store.[R14][R15][S45]
  • Urban population share (SP.URB.TOTL.IN.ZS): This series reports the percentage of total population living in urban areas as defined by national statistical offices.[S46] Because administrative criteria differ across countries, cross-country urban comparisons require caution, and a national urban percentage does not measure store footfall in Douala or Yaounde.[S46]

Exhibit 18 illustrates the frequency and magnitude mismatch between these slow-moving annual macro series and faster-moving channel adoption over a four-year window.

As Exhibit 18 makes clear, annual resident population, real GNI per capita and urban share move by only a few index points over four years, while a specific sales channel such as Cameroon apparel eCommerce (R26, growing at 13.41 percent a year from US$20.32 million in 2023 to a projected US$33.61 million in 2027, with user penetration rising from 18.0 percent to 22.7 percent) compounds by nearly 46 index points over the same span.[R26][S44][S45][S46] At the weekly SKU level, variance is larger still and is dominated by product attributes, selling price, item age, store allocation and stock availability.[S02][S03][S06]

¶ 04 · 7 sources

growing at 13.41 percent a year from US$20.32 million in 2023 to a projected…R26, S44, S45, S46

Exhibit 18

R26S44S45S46S02S03S06

4.11 Interpolating annual macro series into weekly models creates spurious fit

4.11

A common machine-learning error is to linearly interpolate annual population or GNI per capita into fifty-two weekly values and feed them into a tree or neural model alongside weekly SKU sales. Because an interpolated annual series is a smooth monotonic trend within each year, a flexible learner uses it as a surrogate time index, memorising in-sample launch ramps and failing when a new style launches under different conditions.[S09][S37] In GFT's architecture, SP.POP.TOTL, NY.GNP.PCAP.KD and SP.URB.TOTL.IN.ZS are restricted to scaling the annual category envelope in multi-year capacity planning and serving as annual market context in hierarchical Bayesian partial pooling.[S28][S44][S45][S46]

¶ 05 · 6 sources

S09S37S28S44S45S46

4.122 min · 1 sub-steps · 1 exhibit

Cotton statistics describe supply conditions rather than garment demand

For an integrated African textile venture, upstream agricultural statistics are prominent in feasibility studies. The Food and Agriculture Organization of the United Nations (FAO) publishes agricultural production statistics through FAOSTAT covering crop commodities from production volumes to harvested areas up to 2024.[S47] In Cameroon, the inherited GFT evidence register records substantial upstream cotton scale: national seed-cotton production reached 330,000 tonnes in 2022, 314,455 tonnes in 2023 and about 340,000 tonnes in 2024, with a 2025 projection above 350,000 tonnes.[R01][R02] Across the CFA-zone cotton belt, output was reported at 2.3 million tonnes in 2024/25, down 11.5 percent.[R05] Within northern Cameroon, SODECOTON supports more than 220,000 cotton producers, operates nine ginning plants and two oil mills, and injects roughly US$100 million a year into rural communities.[R03][R04]

¶ 01 · 6 sources

national seed-cotton production reached 330,000 tonnes in 2022R01, R02

Related · Exhibit 61 · chapter 13

S47R01R02R05R03R04

4.13 Industrial processing targets justify a factory, not a sales forecast

4.13

Upstream abundance coexists with a severe downstream processing gap. About 95 percent of Cameroon's cotton fibre is exported raw against about 5 percent processed locally, while national industrialisation policy under SND30 sets a target of 50 percent local processing by 2030.[R06][R07] Major industrial announcements respond to that gap, including a proposed SODECOTON and CICAM merger targeted by 2030 and the announced Camtext SA integrated project valued at US$300 million (180 billion XAF) on 12 November 2025, designed to process 12,000 tonnes a year, create about 12,000 jobs at Dibamba and 3,000 at Garoua, and reach optimal capacity around 2033 with an estimated return on investment near 11 years.[R08][R09][R10] Meanwhile, historic domestic mill operations faced severe constraints: a July 2023 report put CICAM capacity utilisation at 33.7 percent in Garoua, 48.9 percent at CICAM 1 and 10.83 percent at CICAM 2, with underutilisation costing about 2 billion XAF in 2020 and restructuring estimates ranging from 30.7 billion to 48.2 billion XAF (with one relaunch plan cited at 70.2 billion XAF).[R12][R13]

¶ 02 · 7 sources

About 95 percent of Cameroon’s cotton fibre is exported raw against about 5 percent processed locallyR06, R07

R06R07R08R09R10R12R13

Exhibit 19 compares Cameroon's annual seed-cotton harvests against the planned fibre intake of a major integrated textile project on a common physical scale.

Exhibit 19Bar chartsourcedAggregate cotton output dwarfs individual mill intake
0100200300400330Seed cotton 2022 (R01)314Seed cotton 2023 (R01)340Seed cotton 2024 (R01)350Projected 2025 (R02)12Camtext SA intake (R10)

Read-outSeed cotton 2022 (R01): 330 thousand tonnes a year

Method and source

Sourced comparison from the inherited GFT evidence register: Cameroon seed-cotton production of 330.0 thousand tonnes in 2022, 314.455 thousand tonnes in 2023 and about 340.0 thousand tonnes in 2024 (R01); 2025 seed-cotton projection above 350.0 thousand tonnes (R02); and announced Camtext SA annual fibre processing of 12.0 thousand tonnes a year (R10). FAOSTAT (S47) defines the agricultural production domain boundary. Original publication dates for R01 are unrecorded in the inherited register.

Unit: thousand tonnes a year. Status: sourced. Sources: R01, R02, R10, S47.

In a forecasting model, FAOSTAT crop series and national cotton balances inform fibre procurement and rural income calendars in northern cotton provinces (where SODECOTON injects roughly US$100 million a year), never SKU demand.[R01][R03][R04][S47] Treating the 50 percent SND30 processing target or a factory's nameplate tonnage as a demand forecast violates the demand contract established in Chapter 2.[R06][R07][R10]

¶ 01 · 7 sources

…provinces (where SODECOTON injects roughly US$100 million a year)R01, R03, R04, S47

R01R03R04S47R06R07R10

4.142 min · 2 sub-steps · 1 exhibit

Search and weather must earn incremental value

Digital search queries and meteorological records arrive at weekly or daily frequency. Dynamic regression and multichannel fashion studies note that external predictor variables such as school and business holidays, commercial events (Easter, Christmas), economic conditions, search behaviour and weather may explain part of historical variation.[S04][S34][S35][S0403] Yet high frequency does not guarantee predictive skill for a cold-start maker. Every high-frequency external signal must be formulated as an ex-ante feature and tested against naive, seasonal-naive and attribute baselines before influencing a production or replenishment commitment.[S09][S10][S33][S0401]

4.15 Web search volume is a near-term attention proxy, not a cold-start guarantee

4.15

Research on predicting consumer behaviour with web search shows that in the absence of other data sources, or where small improvements in predictive performance are material, search queries provide a useful guide to the near future.[S34] That finding carries four strict boundary conditions for a new apparel maker:

¶ 02 · 1 source

S34

  • Horizon decay: Search queries measure active user intent days or a few weeks before purchase; they do not forecast seasonal assortment demand six months ahead when fabrics are ordered.[S34][S0402]
  • Brand versus category terms: A greenfield brand generates negligible brand-specific search before launch, and generic category queries reflect all competitors and informal sellers rather than intent to buy the new maker's offer.[S34]
  • Penetration bias: In Cameroon, where Statista estimates apparel eCommerce user penetration at 18.0 percent in 2023 rising toward 22.7 percent in 2027, search queries over-represent urban digital buyers and under-represent offline channels.[R26][S04]
  • Index normalisation leakage: Public search tools return indices scaled from 0 to 100 over the requested window. Pulling a multi-year index in one request scales early weeks by future peak volume after the forecast origin, introducing preprocessing leakage.[S37]
List · 5 sources

S34S0402R26S04S37

4.16 Weather anomalies explain modest timing shifts rather than structural demand

4.16

Using monthly retail sales and weather data from the National Weather Service, Federal Reserve research finds that unusual weather has a modest but significant role in explaining monthly sales fluctuations, while lagged effects frequently offset the initial impact across a quarter.[S35] Weather variables therefore obey four rules:

¶ 03 · 1 source

S35

  1. Use anomalies against seasonal norms: Raw temperature or rainfall mainly repeats the annual calendar, which a seasonal-naive benchmark captures at zero external-data cost.[S09][S35]
  2. Match stations to selling cities: National averages merge Garoua, Douala and Yaounde; weather series must match the city where the garment is sold.[S04][S35]
  3. Separate ex-ante weather forecasts from ex-post station actuals: Genuine ex-ante forecasts use only information available in advance and require forecasts of the predictors, whereas ex-post forecasts plug in realised predictors after the fact.[S0401] Models must be trained and tested on as-of weather forecasts.[S37][S0401]
  4. Enforce complexity limits: Weather features that add pipeline fragility without improving decisions over a simple baseline are rejected.[S33][S35][S39]

Exhibit 20 synthesises the external data stack into four distinct evidence layers, defining where each external family stops.

Exhibit 20LayersderivedEach external signal stops at a defined evidence layer
Layer 1: Supply context (excludedfrom demand)FAOSTAT crop and cotton outpu…Raw export vs local processin…Announced mill capacity and u…Layer 2: Annual market ceiling andscreenDe facto population SP.POP.TO…Real GNI per capita NY.GNP.PC…National urban share SP.URB.T…Layer 3: Category inflow andunit-value boundsUN Comtrade reporter and mirr…ITC Trade Map HS 61 and 62 fl…Per-person apparel spend base…Layer 4: Near-term ex-ante challengerinputsWindow-frozen web search indi…City weather anomaly forecast…Channel holiday and event cal…

Read-outLayer 1: Supply context (excluded from demand): FAOSTAT crop and cotton output (S47, R01), Raw export vs local processing (R06, R07), Announced mill capacity and use (R10, R12)

Method and source

Derived GFT architecture mapping external data families to their maximum permitted forecasting role. Upstream agricultural data (S47) remain in supply context; World Bank macro indicators (S44) stop at annual market scaling and analogue screening; UN Comtrade trade series (S41) stop at category inflow bounds; and search, weather and calendar series (S04, S34, S35) enter near-term SKU models only as as-of audited challengers (S37).

Unit: permitted forecasting scope. Status: derived. Sources: S04, S34, S35, S37, S41, S44, S47.

Exhibit 20 enforces a strict upward firewall: Layer 1 never crosses into demand estimation, Layers 2 and 3 shape only annual market bounds and hierarchical priors, and only Layer 4 signals may compete as short-horizon temporal features alongside product attributes and launch history.[S02][S24][S37] Even inside Layer 4, architectures with feature-selection gates such as Temporal Fusion Transformers (TFT) must prove in held-out launch replay that external inputs improve accuracy and calibration over models that suppress them.[S12][S24][S0403]

¶ 01 · 5 sources

Layer 1 never crosses into demand estimationS02, S24, S37

Exhibit 20

S02S24S37S12S0403

4.173 min · 1 sub-steps · 1 exhibit

A covariate registry prevents a proxy from becoming a story

In a small industrial analytics team, external covariates create two compounding hazards: hidden technical debt and temporal data leakage. Machine-learning systems research warns that solutions providing a tiny accuracy benefit at the cost of massive increases in system complexity are rarely wise practice.[S39] Methodological audits show that data leakage is a spurious relationship between independent variables and the target variable arising as an artefact of data collection, sampling or preprocessing strategy.[S37] External covariates are prime carriers of leakage because macroeconomic tables are revised retroactively, trade statistics arrive months after the customs month closes, search indices rescale across query windows, and weather archives record realised actuals rather than forecasts available at the decision origin.[S37][S41][S0401]

4.18 Every external feature must pass a bitemporal registration gate

4.18

To prevent an external proxy from becoming a post-hoc story, GFT requires every external series to be logged in a governed Covariate Registry before feature engineering or training begins.[S36][S37][S40] The registry treats time as bitemporal: every record stores both the valid time (the calendar period the event describes) and the as-of availability time (the timestamp at which that value became observable to the pipeline).[S37] In expanding-window chronological cross-validation, where successive training sets are supersets of those that come before them, a feature value is admissible at the forecast origin only if its as-of availability timestamp is on or before that origin, and for future horizons only if it is a known calendar structure or an ex-ante predictor forecast issued by that origin.[S13][S37][S0401]

¶ 02 · 5 sources

S36S37S40S13S0401

The GFT five-step admission test for external covariates

  1. Boundary and layer assignment: Map the candidate series to Layer 1, 2, 3 or 4 in Exhibit 20 and reject any supply-side statistic proposed as a direct SKU demand predictor.[S47]
  2. Bitemporal as-of audit: Verify that historical snapshots preserve publication lags, customs revisions and ex-ante predictor forecasts as they stood at each historical decision origin, with scaling parameters fitted strictly inside each expanding training window.[S13][S37][S0401]
  3. Baseline ablation in launch replay: Run the cold-start backtest with and without the candidate covariate against naive, seasonal-naive and attribute-only baselines, scoring mean absolute scaled error (MASE, where defined), bias and quantile calibration.[S09][S10][S12]
  4. Ex-ante versus ex-post degradation check: Compare backtest error using realised ex-post covariate values against backtest error using realistic ex-ante covariate forecasts to measure how much predictor noise erodes skill at the required horizon.[S0401]
  5. Complexity and drift sign-off: Confirm with named commercial and technical owners that the decision improvement exceeds the maintenance and concept-drift burden; attach an automatic fallback to the covariate-free baseline if the external feed stalls.[S33][S36][S38][S39][S40]
  6. GFT note

    In GFT's Cameroon textile programme, external statistics are separated strictly by decision horizon. Feasibility and capacity-sizing models use UN Comtrade and ITC Trade Map series, World Bank demographic and income indicators, Statista expenditure baselines and national cotton-processing benchmarks to bound the macro opportunity and stress-test multi-year scenarios.[R06][R26][S41][S43][S44][S45][S46] None of those series is permitted to nominate a winning garment style or set a first-lot SKU production commitment. Before launch, SKU priors come from audited analogue distributions (Chapter 3) and calibrated primary choice research (Chapter 5); after launch, short-term covariates such as search or weather enter the forecasting service only if they pass the bitemporal registry gate and beat simple baselines in held-out launch replay.[S09][S34][S35][S37] This paper provides an engineering and forecasting methodology, not investment, legal or tax advice.

Exhibit 21 specifies the mandatory schema of the GFT Covariate Registry across the external data families examined in this chapter.

Exhibit 21RegisterderivedA covariate registry binds every external proxy to an as-of rule
Covariate family
Merchandise trade (HS 61 to 63)
Primary source
UN Comtrade and ITC Trade Map (S41, S43)
Cadence and lag
Annual or monthly; multi-month customs lag
Leakage control (S37, S0401)
Freeze snapshot date; retain reporter and mirror columns (S42)
Permitted role
Category inflow and unit-value bounds

Read-outCovariate family: Merchandise trade (HS 61 to 63). Primary source: UN Comtrade and ITC Trade Map (S41, S43). Cadence and lag: Annual or monthly; multi-month customs lag. Leakage control (S37, S0401): Freeze snapshot date; retain reporter and mirror columns (S42). Permitted role: Category inflow and unit-value bounds

Open the full register (6 rows)
Merchandise trade (HS 61 to 63)UN Comtrade and ITC Trade Map (S41, S43)Annual or monthly; multi-month customs lagFreeze snapshot date; retain reporter and mirror columns (S42)Category inflow and unit-value bounds
Population and urbanisationWorld Bank WDI SP.POP and SP.URB (S44, S46)Annual midyear estimate; census revisionsUse vintage release date; forbid weekly linear interpolationMarket ceiling and analogue market screen
Real income per capitaWorld Bank WDI NY.GNP.PCAP.KD (S45)Annual constant-price national accountsLock national accounts release date; never use ex-post GDPMulti-year scenario scaling only
Cotton and crop outputFAOSTAT and national ginning reports (S47, R01)Annual crop campaignFirewall from demand target; log campaign publication dateFibre supply context and rural calendar
Web search query volumeSearch query index APIs (S34)Weekly; near-real-time index rescalingScale strictly inside each training fold up to originShort-horizon challenger for active offers
City weather anomaliesMeteorological station feeds (S35)Daily or weekly; forecast error grows with horizonTrain and test on as-of weather forecasts, not ex-post actualsShort-horizon replenishment modifier
Method and source

Derived GFT governance specification combining external source definitions (S34, S35, S41, S44, S47) with leakage controls (S37) and ex-ante predictor rules (S0401). Every external feature must populate these columns before entering an evaluation fold.

Unit: mandatory governance fields by external source family. Status: derived. Sources: S34, S35, S37, S41, S44, S47, S0401.

Exhibit 21 turns methodological caution into an auditable engineering gate. When an external feed fails its bitemporal check, or when removing the feature leaves launch-replay error and planning decisions unchanged, the feature is retired from the model pipeline rather than retained for narrative appeal.[S33][S37][S39] Continuous monitoring tracks summary statistics of every retained covariate and triggers a notification or automatic rollback to the covariate-free baseline when an upstream agency alters a definition, delays a release or introduces a structural break.[S38][S40]

By closing Part I with a governed covariate registry, the cold-start maker establishes a complete outside view: a demand contract that distinguishes latent demand from sales (Chapter 2), an analogue prior that preserves launch dispersion (Chapter 3), and external market bounds that respect provenance and chronology (Chapter 4). Part II turns from borrowed context to owned pre-launch and pilot evidence, showing how structured primary research and controlled release experiments convert market bounds into style-level purchase signals.

¶ 02

Check the record

The chapter keeps its sources within reach.

Every knot is a source this chapter cites; select one to read its record, or filter the register by grade. Publication dates and inherited-source limits stay explicit. Read the paper's full evidence summary.

52 sources · 269 citation marks on this pageMost cited: S37, 18 times

Grade AGrade BGrade CKnot size: citations on this page

All 52 records shown

R01 · Grade C · inherited record

Business in Cameroon

Inherited evidence R01

Cameroon seed-cotton production 2022-2024: 330,000 t (2022), 314,455 t (2023), about 340,000 t (2024).

Publication date unrecorded. Inherited source-text record. The original publication has not been independently reopened for this web edition.

Chapter source notes

S41 United Nations Statistics Division, UN Comtrade Help Center. 2026-10-30.

S43 International Trade Centre. 2014.

S44 World Bank, World Development Indicators. 2026-10-30.

S45 World Bank, World Development Indicators. 2026-10-30.

S46 World Bank, World Development Indicators. 2026-10-30.

R06 Cameroon Trade Hub. Publication date unrecorded; inherited source-text check.

S47 Food and Agriculture Organization of the United Nations. 2025.

S09 PLOS ONE, Public Library of Science. 2018-03-27.

S34 Proceedings of the National Academy of Sciences. 2010.

S35 Board of Governors of the Federal Reserve System, Finance and Economics Discussion Series. 2000.

S0401 OTexts. 2021-05-31.

S01 InTech, International Journal of Engineering Business Management. 2013.

S0402 arXiv. 2022-11-09.

S08 International Journal of Forecasting, Elsevier; full text through arXiv. 2022.

S37 Cell Press, Patterns; open full text from PubMed Central. 2023-08-04.

R07 Ecofin Agency. Publication date unrecorded; inherited source-text check.

S06 Lancaster University Management School, Munich Personal RePEc Archive. 2018-11-19.

S02 arXiv. 2019-06-27.

S03 arXiv. 2020-07-10.

R11 Business in Cameroon; MMI News. Publication date unrecorded; inherited source-text check.

R16 WageIndicator. Publication date unrecorded; inherited source-text check.

R26 Statista. Publication date unrecorded; inherited source-text check.

R18 Congressional Research Service IF10149; AGOA (Wikipedia). Publication date unrecorded; inherited source-text check.

R23 MacMap; AfCFTA. Publication date unrecorded; inherited source-text check.

R24 Mordor Intelligence. Publication date unrecorded; inherited source-text check.

R25 Market.us. Publication date unrecorded; inherited source-text check.

S0403 OTexts. 2021-05-31.

S42 United Nations Statistics Division, UN Comtrade Help Center. 2026-10-30.

R32 Wise; XE. Publication date unrecorded; inherited source-text check.

S40 Google Cloud Documentation. 2024-08-28.

R10 Ecofin Agency. Publication date unrecorded; inherited source-text check.

R14 WageIndicator. Publication date unrecorded; inherited source-text check.

R15 wage.is. Publication date unrecorded; inherited source-text check.

S28 Journal of Industrial Ecology, Wiley. 2022-10-17.

R01 Business in Cameroon. Publication date unrecorded; inherited source-text check.

R02 Fibre2Fashion; Business in Cameroon. Publication date unrecorded; inherited source-text check.

R05 Ecofin Agency. Publication date unrecorded; inherited source-text check.

R03 SODECOTON. Publication date unrecorded; inherited source-text check.

R04 Financial Fortune Media. Publication date unrecorded; inherited source-text check.

R08 Business in Cameroon. Publication date unrecorded; inherited source-text check.

R09 Ecofin Agency. Publication date unrecorded; inherited source-text check.

R12 Business in Cameroon. Publication date unrecorded; inherited source-text check.

R13 Business in Cameroon; MMI News. Publication date unrecorded; inherited source-text check.

S04 MDPI, Processes. 2021-09-03.

S10 International Journal of Forecasting, Elsevier; abstract through RePEc IDEAS. 2018.

S33 Google for Developers. 2025-08-25.

S39 Neural Information Processing Systems Foundation, NIPS 2015. 2015.

S24 arXiv. 2020-09-27.

S12 International Journal of Forecasting, Elsevier; abstract through RePEc IDEAS. 2006.

S36 National Institute of Standards and Technology, U.S. Department of Commerce. 2023.

S13 scikit-learn. 2026-10-30.

S38 ACM Computing Surveys; open record from Bournemouth University. 2014-04-30.

Keep the thread · chapter 04 of 14

Chapter 05 picks up the thread.

Move the shuttle along the rack to preview any chapter: its sections, exhibits and reading time. Draft web edition. This research is not investment, legal or tax advice; illustrative models are not measured factory results.

The paper's rack · 14 boltsBolt length: reading time

05Part II: Earn evidence before scale

Primary research turns intent into an explicit prior

Stated interest becomes useful only when sampling, incentives and calibration connect it to a future purchase decision.

  • 5.1Respondent segments follow use occasions and buying roles
  • 5.4Conjoint separates attribute preferences before launch
  • 5.7Choice responses need calibration against purchase settings
  • 5.9Price research produces scenarios, not a universal elasticity
Reading time
31 min
Exhibits
4
Sections
7
Continue to chapter 05
← Previous: chapter 03, Analogue products provide the first outside viewContinue the threadChapter 05 · Primary research turns intent into an explicit prior

Find a thread.

Type to search.