← Quantery Blog

How to turn a filing into a testable thesis

September 17, 2026 · 9 min readthesis-buildingfilingsresearch

A filing is a better starting point for a thesis than a stock story. It gives you dated facts, a stated problem, and enough detail to ask a precise question. It doesn't give you an edge. The work is to turn one fact in the document into a rule that could have been applied to every similar company before you know what happened next.

Start with the actual filing. State the claim in plain language. Pick an observable proxy from fields that were available at the time. Then compare that rule with a simpler version of the idea in a walk-forward backtest. If it cannot survive that sequence, keep it as a research note. It has not earned a place in a thesis yet.

This is useful after an earnings release, a quarterly report, or an unexpected corporate announcement. It also keeps a timely story from smuggling its outcome into your rules. You can learn from an interesting company without turning its ticker into a recommendation.

Why does a filing make a good research prompt?

A public filing records what management disclosed at a particular point in time. For an operating company, that usually includes the financial statements, management's discussion of the period, and details that turn a broad narrative into things you can inspect. The SEC's Form 10-Q instructions describe the quarterly report used after each of the first three fiscal quarters. You'll find a company through the SEC's EDGAR filing search, which links the reports and exhibits themselves.

The date matters. A useful thesis has to respect when a reader could have known an input. A year-end number that arrived in a February filing cannot justify a simulated decision in January. This is the difference between a tidy retrospective and a rule that could have existed at the time. Trailing versus forward numbers explains why filed, trailing values give a thesis a firmer footing than estimates that later changed.

Use the document to find tension. Revenue can rise while operating cash flow falls. A debt balance can remain high while the business slows. Shares can increase while net income does not. None of those observations settles the investment case. Don't ask them to. Each can become a question about a recurring business condition.

Apple's quarterly report for the quarter ended June 27, 2026 is a compact illustration of the document structure. It presents unaudited condensed consolidated statements of operations, balance sheets, and cash flows. Those three statements give you flows through the business, the financial position at a date, and the bridge between accounting profit and cash. Start with a relationship across those statements. An isolated headline number rarely carries enough meaning.

How do you separate a fact from a claim?

Write two sentences before you open the thesis builder.

The first is the filing fact. It should say only what the document establishes: operating cash flow declined while revenue grew; debt increased after an acquisition; receivables grew faster than sales; share count rose. Preserve the period and source in your notes. Don't add a cause the filing does not establish.

The second is the research claim. It says what you think might recur across companies: businesses with receivables rising faster than revenue may have weaker future cash conversion; businesses funding an acquisition with additional debt may have less balance-sheet room when earnings stall. Keep it in hypothesis territory. The filing supplied the prompt; testing has to do the rest.

The gap between those sentences is where most bad rules enter. "Receivables rose" is observable. "Customers will not pay" is a story about the future. The first belongs in a feature. The second may motivate the research, but it can't be an input unless you can define it from information available on the filing date.

Keep the claim narrow enough to lose. "Companies with weak quality do badly" cannot be tested because weak quality has no settled meaning. "Companies whose trailing operating cash flow is below trailing net income while receivables grow faster than revenue" has a defined condition. It can pass, fail, or prove irrelevant in a test. You aren't trying to make the rule cover every risk in the filing. You're trying to isolate one possible mechanism.

That discipline also protects against hindsight. If the company later missed guidance, got acquired, or recovered, don't tune the condition until it describes that ending. Take the same condition to companies you did not pick first. The original company should become one observation among many.

Which filing line can stand in for the idea?

A proxy is a measurable stand-in for a business idea. It is not the idea itself. You choose it because no filing line arrives with a label saying what it means. That choice is the central judgment in thesis design.

Suppose the filing prompt is concern that reported profit may not be reaching the bank. The claim is not "the company is manipulating earnings." That would be a serious conclusion, and the statement lines alone cannot prove it. A workable claim is smaller: a persistent gap between net income and operating cash flow can identify profits that deserve more inspection.

The proxy could compare trailing operating cash flow with trailing net income. Both inputs come from public financial statements. The cash flow versus net income article explains the accruals gap behind that comparison, while free cash flow adds the capital spending a business needs to fund its assets. The two ratios ask related questions. Don't award both a full score unless you've decided which claim each adds. You don't need every valid ratio in one score.

Other filing prompts have other candidates:

• A concern about refinancing pressure can begin with debt, cash, and earnings before interest, taxes, depreciation, and amortization. Gross debt and net debt covers why cash is not always fully available to offset debt.
• A concern that growth is being bought through equity issuance can use point-in-time share count alongside revenue and cash flow. Share count and dilution explains why per-share results can move when the underlying business does not.
• A concern that a business is funding a large buildout can compare capital expenditures with operating cash flow or depreciation over several filing periods. One period alone may capture timing. The capital cycle is the right context for that comparison.

Use only a proxy whose inputs the data can support on the historical dates you intend to test. A management-defined organic-sales measure may help you understand a specific release, yet it doesn't live as a consistent, standardized filing field across the market. Keep it as a manual research note. A number beside it doesn't make it a general rule.

Write the smallest version of the rule

Once the claim and proxy are clear, write the smallest thesis that expresses them. The shape below is illustrative. The bundled templates are the reference for exact fields and options, while the Quantery documentation introduces the parts of a thesis.

# Cash backing under pressure: illustrative
params:
  conversion_strong: 1.0
  conversion_ok: 0.70
  receivables_growth_max: 0.10

features:
  net_income_ttm: ttm(net_income)
  operating_cash_flow_ttm: ttm(operating_cash_flow)
  cash_conversion: if(net_income_ttm > 0,
    operating_cash_flow_ttm / net_income_ttm, null)
  revenue_now: newest(revenue)
  revenue_year_ago: lag(revenue, 4)
  receivables_now: newest(receivables)
  receivables_year_ago: lag(receivables, 4)
  revenue_yoy: if(revenue_year_ago > 0,
    revenue_now / revenue_year_ago - 1, null)
  receivables_yoy: if(receivables_year_ago > 0,
    receivables_now / receivables_year_ago - 1, null)
  receivables_outpacing_sales: receivables_yoy - revenue_yoy

criteria:
  cash_backing:
    rules:
      - { when: "is_null(cash_conversion)", score: 0, flag: no_conversion }
      - { when: "cash_conversion >= $conversion_strong", score: 2 }
      - { when: "cash_conversion >= $conversion_ok", score: 1 }
      - { else: 0 }
  collections:
    rules:
      - { when: "is_null(receivables_outpacing_sales)", score: 0, flag: no_comparison }
      - { when: "receivables_outpacing_sales <= $receivables_growth_max", score: 1 }
      - { else: 0 }

gate:
  mode: score
  min_score: 2

The threshold values are proposed assumptions drawn before any result. Keep them round. A precise value found after reviewing the outcome is usually a parameter that memorized one period.

Notice the null guards. A missing line shouldn't become a passing score. It might be absent because of a filing format, an issuer's reporting convention, or incomplete historical coverage. Whatever caused it, the thesis has less evidence in that case. Make it say so.

The code separates cash backing from the receivables comparison. The two can agree or disagree. A company might generate cash despite receivables growth because another working-capital item moved. Another might have weak conversion without an unusual receivables trend because capital needs changed.

What should the backtest compare?

Run the smallest rule against a baseline that answers the claim you made. If the hypothesis is that poor cash backing identifies a condition to avoid, compare the thesis with the same universe before the added condition. Hold the universe, rebalance schedule, and benchmark fixed. That setup gives any difference a chance of describing the extra rule instead of a different market exposure.

Every backtest remains hypothetical and excludes trading costs. It can show whether the condition was associated with different outcomes in the dates and data it saw. It can't establish that the filing line caused those outcomes. Sector concentration, company size, the selected window, and a handful of large moves can still explain the result. What makes a backtest honest covers the data and timing rules that keep this exercise from peeking ahead.

Run a second version with one change. Move the cash-conversion boundary by a modest round amount. Remove the receivables criterion. Shift the start date. This is an ablation test when you remove a component to see whether it was doing the work. If a conclusion disappears when one nonessential choice moves, write down that fragility. How to tell if a backtest result is real lays out the broader perturbation discipline.

Keep the versions that lost too. A filing is interesting because it gives you a real question. Your research becomes useful when its rules can meet that question before the next filing, across companies you haven't already decided to care about.

Keep the filing in the loop after the test

A market-wide rule can surface patterns, but it can't replace reading. When a company qualifies, return to the filing that supplied the inputs. Read the notes, the cash flow reconciliation, and the discussion of the period. Confirm that the ratio reflects the mechanism you thought it represented. A growing receivables balance may be ordinary seasonal timing. A low cash conversion may follow a deliberate inventory build. The screen is a way to prioritize that work.

The same return trip makes the thesis editable. If repeated review shows that the rule is catching companies with a distinct, explainable condition, keep testing it. If it only finds accounting artifacts from a certain industry or falls apart outside one window, narrow the claim or remove it. The rules are yours to inspect, change, and test again. That's a much better use of a filing than treating it as a verdict on a single stock.

Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.

← All articles