ParallaxFX · Methodology
Industry Benchmarks
Industry benchmarks let you see how your FX risk position compares against publicly disclosed practice in your industry. This page explains exactly where the data comes from, how it is processed, and — just as importantly — what it does not tell you.
What an industry benchmark is
An industry benchmark is a quartile distribution of a specific FX risk metric — hedge coverage, materiality threshold, currency concentration, and similar measures — for a defined industry, derived from the market-risk disclosures of public companies. When you see a benchmark in ParallaxFX, you are seeing where your value sits relative to that distribution: which quartile you fall in, and the industry median.
We present benchmarks as quartile bands and percentile positions — not as precise industry averages. Given the realities of sample size and disclosure quality, a quartile band is an honest unit of comparison; a single-point average would imply a precision the underlying data does not support.
Why we built them — and why not "peer cohorts"
FP&A leaders ask comparison questions constantly: "Is our hedge coverage typical for our industry?" "Are our materiality thresholds reasonable?" Today the honest answer is usually a guess based on conversations at industry events and advisor opinions. A defensible benchmark — even one with stated limitations — improves the quality of those answers materially.
We deliberately use the term industry benchmarks, not "peer cohorts." A peer cohort would imply a set of comparable private companies like yours. We do not have that, and we will not imply we do. What we have is honest industry context from public disclosure — and that is exactly what we call it.
Data source — SEC EDGAR only
Industry benchmarks are derived from a single source: the "Item 7A — Quantitative and Qualitative Disclosures About Market Risk" sections of Form 10-K annual filings made by US-listed public companies to the U.S. Securities and Exchange Commission, available through SEC EDGAR.
A single data source means a single methodology to defend. We considered combining SEC data with practitioner heuristics from treasury surveys and academic research, and chose not to: those sources carry subscription costs and redistribution restrictions, and the marginal value over what an experienced FP&A leader already knows did not justify building the product on data we could not freely stand behind. SEC filings are public domain, carry no licensing restrictions, and are grounded in primary disclosure.
How the data is processed
Each filing moves through the same pipeline before it contributes to a benchmark:
- Filing discovery. We identify the set of 10-K filings for an industry, filtered by industry classification code and recent fiscal years.
- Item 7A identification. Within each filing, we locate the market-risk disclosure section.
- AI-assisted structured extraction. Item 7A disclosures are unstructured prose. We use a large language model with deterministic settings to extract structured values — hedge coverage percentages, currency lists, hedging stance — each accompanied by the source quote it was based on and a confidence indicator.
- Analyst validation. Every extracted record is reviewed by an analyst against the source filing. Records are marked validated, flagged for correction, or rejected. The model never makes the final call — rejected records are excluded entirely.
- Aggregation. Validated records for an industry are aggregated into quartile distributions (25th / 50th / 75th percentile), with the sample size and aggregation date recorded.
- Confidence assessment. Each distribution is assigned a confidence band based on sample size. Low-confidence distributions are flagged; insufficient ones are not shown.
The language model is used only for extraction — turning prose into structured data. It is never used to generate analysis, decide what a benchmark says, or interpret your results.
Confidence framing and sample sufficiency
Every benchmark is shown with its sample size and a confidence band. We surface a benchmark only when the sample is large enough to stand behind:
| Sample size | Confidence | Behavior |
|---|---|---|
| 30+ filings | High | Shown with full presentation |
| 15–29 filings | Moderate | Shown with a sample-size caveat |
| 10–14 filings | Low | Shown only with an explicit small-sample disclosure |
| Fewer than 10 | Insufficient | No benchmark shown — "industry benchmark unavailable" |
When your industry has too few comparable filings, ParallaxFX tells you so plainly rather than showing a weak number. That is the correct behavior, not a failure — a benchmark you cannot trust is worse than no benchmark.
Industry classification
Public filers are classified by Standard Industrial Classification (SIC) code, cross-referenced against NAICS, and mapped to a small set of plain-language ParallaxFX industry categories — B2B Software, Industrial Manufacturing, Consumer Goods, and so on.
You set your own classification in Company Settings. Because you know your business better than any code does, your selection is authoritative — you can change it at any time, and you can pick the comparison set that genuinely fits how you operate.
Coverage
The initial benchmark corpus is being assembled. As filings are extracted and analyst-validated, this section will list the industries covered, the sample size behind each, and the date of the most recent aggregation. Until an industry reaches the minimum sample threshold, benchmarks for it will show as unavailable.
The initial corpus is point-in-time. Refresh cadence — most likely annual, after each filing season — will be confirmed and stated here once set.
Limitations — read this part
- Public companies only. SEC EDGAR contains public filings. Mid-market private companies — most of our customers — are not in the source data. A benchmark describes what publicly disclosed companies in your industry do, not the full market.
- Sample composition skews larger and more mature. Public filers are larger and more financially sophisticated than the mid-market average. Expect hedge coverage to skew higher and materiality thresholds to skew lower than private-company norms.
- Disclosure quality varies. Item 7A is prose, and companies write it with varying depth and precision. Extraction and validation reduce error but cannot remove it entirely.
- Point-in-time, not continuous. Benchmarks are snapshots from a filing cycle, not live market data.
- Not size- or model-adjusted. Companies within an industry vary in size, profitability, and business model; benchmarks are not adjusted for those differences.
What a benchmark is not
A benchmark is not a statistical sample of every company in your industry, not a prediction of what your company should do, and not a validated best practice. It describes what public companies disclosed — which is useful context for a decision, not a substitute for one. Treat it as one input among several.
Availability and contact
Industry benchmarks are an Intelligence-tier feature. They appear in the product once your company is classified and your industry has a sufficient sample. This methodology page is public — the credibility of a benchmark comes from an open explanation, not from gatekeeping it.
Questions or concerns about a benchmark's accuracy? Email hello@parallaxfx.ai.