FIN-EQR-001 · Securities-firm earnings forecast · A-share listed broker · 2026E–2028E
On a real brokerage earnings-forecast template (23 worksheets, about 18,000 formulas), build a securities-firm revenue split, a 2026E–2028E earnings forecast, balanced three statements and a peer-average PE valuation, delivered as an Excel model that updates when assumptions change.
The model works on a real brokerage earnings-forecast template: 23 worksheets and about 18,000 formulas that originally relied on a third-party valuation plug-in. The plug-in is removed and replaced by an input workbook and ordinary Excel formulas.
Revenue must be split by securities business — brokerage, investment banking, asset management, credit, investment income — and operating expenses and credit impairment forecast separately. Falling back on “volume × price” or “cost + expense” counts as a severe error.
Target PE must be the peer average and EPS strictly attributable net profit ÷ total shares. Substituting DCF, FCFF or WACC for the PE conclusion triggers the severe-error cap.
Claude Opus 4.8 finished in 28 minutes and scored 0.772; GPT-5.6-sol used the full 30-minute limit and scored 0.583 on its saved workpaper. The reference workpaper scores 1.000, leaving clear headroom.
An excerpt of the instruction the agent actually receives, showing the task setup and constraints.
You are a financial modelling analyst covering the securities industry. Using the company materials, the model input workbook and the earnings-forecast template in the package, build a 2026E–2028E earnings forecast and PE valuation model for XX Securities. The model must run without installing or calling any valuation plug-in.
Complete the front-end revenue split first, then roll it up into the income statement. Split at least: brokerage, investment banking, asset management, credit business, investment income and fair-value gains, and other business or other operating income.
For each business, state the historical source, the forecast drivers and the formula. Business drivers such as commission rates, client assets, margin-financing balances, stock-pledge volume, underwriting volume, AUM and investment yield may be used. The sum of the segments must equal operating revenue in the income statement. Do not rely only on a manufacturer's revenue growth rate or “volume × price” logic without explaining its meaning for a securities business.
The profit forecast must reflect a securities firm's operating structure, treating operating and administrative expenses, credit impairment losses and other securities-specific costs or gains separately. Operating and administrative expenses and credit impairment losses need their own line items and reviewable forecast logic, and may not be merged into a general company's “cost + expense”.
Map general-company line items approximately to securities-firm items and keep the mapping notes; balance the balance sheet; check that the cash-flow statement reconciles to the change in cash and cash equivalents; avoid circular references.
NOTEEnglish translation of the original Chinese instruction; the deliverable list and path requirements are omitted.
Everything visible inside the agent container, with statistics generated from the task package. Original files ship with the full sample package.
| File | Type | Size | Bytes |
|---|---|---|---|
| 002945_earnings_forecast_model.xlsxEarnings-forecast template: revenue split, sales and asset forecasts, three statements, valuation, peer analysis, sensitivity and more — 23 worksheets | xlsx | 23 sheets | 436 KB |
| 002945_model_inputs.xlsxModel inputs: historical data, business mapping, model definitions, peer PE, sensitivity ranges | xlsx | 6 sheets | 15 KB |
Full criteria text, weights and reference answers ship with the complete data package.
Trajectory statistics from frontier models in real agent harnesses. One turn is one model call; the context curve shows how state accumulates over a long-horizon task.
| Model | Agent turns | Tool calls | Cumulative prompt tokens | Peak context | Time | Reward |
|---|---|---|---|---|---|---|
| Claude Opus 4.8Claude Code · Harbor | 36 | 2.86M | 123K | 27m 54s | 0.772 | |
| GPT-5.6-solCodex · Harbor | 25 | 1.93M | 102K | 30m 04s | 0.583 · Hit time limit |
| Task ID | FIN-EQR-001 |
|---|---|
| Task format | Harbor task.toml · schema 1.4 |
| Package version | 1.0.0 |
| Agent time limit | 30 min |
| Verifier time limit | 5 h |
| Agent container | 2 vCPU · 8 GB memory · 30 GB storage |
| Scoring isolation | Verifier runs as a separate stageJudge credentials are injected only into the verifier and are invisible to the agent |
| Instruction length | 1,774 characters |
| Difficulty tier | A3 |
| Workpaper size | Template of 23 worksheets · about 18,000 formulas |
| Reference solution | oracle reward 1.000 |
Leave your details and we will send the full sample package for this task (instruction, attachments, verifier and reference trajectories) and walk you through volume and customisation for similar tasks.