# Benchmarking golden-source pricing and valuation confidence
A private credit loan sits in your fund at 99.5. Your primary vendor marks it at 98.2. A broker quote says 97. Month-end is in 48 hours, and the NAV (Net Asset Value, the per-share value at which investors subscribe and redeem) has to strike. Which number is right, and how confident are you? That question, not the price itself, is the real job of a valuation team.
This lesson is about building the data machinery that answers it: a defensible pricing hierarchy, a scoring system for vendor prices, and a quantified confidence score you can defend to an auditor.
A golden source is the single, authoritative record for a given data point that all downstream systems inherit. For a liquid US large-cap equity, the golden source is trivial: the closing price on the primary exchange. For illiquid assets, there is no single authority, so you construct one from a hierarchy of inputs.
The assets that cause pain:
The US private credit market alone is commonly estimated at over 1.5 trillion USD in assets under management as of 2024 to 2025 (industry estimate, figures vary by source). These positions rarely trade, so their "price" is an opinion. Your data discipline turns opinions into an auditable number.
Regulators expect a documented waterfall. Under ASC 820 (US GAAP) and IFRS 13 (international), fair value inputs are ranked into three levels:
Your operational hierarchy sits underneath this. A typical waterfall for a bank loan:
1. Composite evaluated price from a primary pricing vendor
2. Secondary vendor evaluated price
3. Average of executable dealer quotes (2 or more)
4. Single dealer indicative quote
5. Model or matrix price from comparable instruments
6. Prior price carried forward (stale, flag it)
Each rung down is a lower confidence tier. The rule of the game: document the rule before you need it, so the price selection is systematic, not a month-end judgment call.
The major evaluated-pricing vendors you will actually name: ICE Data Services, Bloomberg (BVAL), S&P Global Market Intelligence (including the former IHS Markit pricing), and Refinitiv (LSEG). For private assets, valuation specialists like Houlihan Lokey, Kroll, and Lincoln International provide independent marks.
You do not trust a single vendor blindly. You benchmark them against each other and against your independent evaluated mark. Three data-quality metrics do most of the work.
The spread across independent sources for the same instrument.
Worked example, a single high yield bond:
| Source | Price |
|--------|-------|
| Vendor A (BVAL) | 96.80 |
| Vendor B (ICE) | 97.10 |
| Dealer quote 1 | 96.50 |
| Dealer quote 2 | 97.40 |
Mean = (96.80 + 97.10 + 96.50 + 97.40) / 4 = 96.95
Max minus min = 97.40, 96.50 = 0.90 points dispersion.
Dispersion as a percentage of mean = 0.90 / 96.95 = 0.93%.
A rule of thumb some shops use: flag any Level 2 bond where dispersion exceeds a threshold (for example, 1% or 2 points) for analyst review. This is a tolerance break, and it should route automatically to a queue.
How many business days since the price last moved. A private credit mark that has been flat at 99.5 for 40 trading days is not "stable," it is stale. Track days since last price change per position and set a staleness ceiling by asset class (tighter for high yield, looser for a quarterly-marked PE stake).
Over a rolling window, the percentage of instruments where two vendors agree within tolerance.
$$\text{Agreement rate} = \frac{\text{instruments within tolerance}}{\text{total instruments priced by both}}$$
If ICE and BVAL agree within 0.5 points on 940 of 1,000 bonds, agreement rate is 94%. Track this per vendor pair, per sector. A drop from 94% to 85% in leveraged loans over a month is a data signal worth investigating before it hits NAV.
Now combine the signals into a single valuation confidence score per position. This is the deliverable your CFO and auditor want.
A simple, defensible scoring model assigns points across dimensions:
def valuation_confidence(source_count, dispersion_pct,
days_stale, fair_value_level):
score = 100
# Fewer independent sources = lower confidence
if source_count >= 3: score -= 0
elif source_count == 2: score -= 10
else: score -= 25
# Higher dispersion = lower confidence
if dispersion_pct <= 0.5: score -= 0
elif dispersion_pct <= 1.5: score -= 10
else: score -= 25
# Staleness penalty
score -= min(days_stale * 2, 30)
# Fair value level penalty
score -= {1: 0, 2: 10, 3: 25}[fair_value_level]
return max(score, 0)
# Our private credit loan: 1 source, 0% dispersion (no trades),
# stale 40 days, Level 3
print(valuation_confidence(1, 0.0, 40, 3)) # -> 20That loan scores 20 out of 100. Low confidence, correctly flagged, despite a clean-looking flat price. Contrast a Level 2 bond with 3 sources, 0.3% dispersion, priced today: it scores 90. The score exposes the illusion of stability that a static price creates.
Weights are illustrative, not a standard. The point is that the model is transparent, applied consistently, and produces an auditable trail.
🎬 [VIDEO: "Fair Value Measurement (IFRS 13) explained" - youtube.com - a clear walkthrough of the three-level fair value hierarchy and observable versus unobservable inputs]
For the governance backbone behind all of this, the CFA Institute's Global Investment Performance Standards (GIPS) and IOSCO's principles for the valuation of collective investment schemes are the reference frameworks regulators lean on.
Knowledge check
1. The lesson argues that for illiquid assets, the 'real job' of a valuation team is not determining the price itself but rather:
2. Why is establishing a 'golden source' described as a data problem rather than a math problem for illiquid assets?
3. Under ASC 820 / IFRS 13, why would a private credit loan marked by three disagreeing sources (99.5, 98.2, 97) most likely fall into a lower fair-value level than a US large-cap equity?
4. Select ALL correct answers about the categories of assets the lesson identifies as causing valuation pain.
Select all the correct answers.
5. Select ALL correct answers about why a documented pricing hierarchy (waterfall) matters for illiquid valuations.
Select all the correct answers.
Position-level scores are inputs. The board and the fund administrator care about the NAV-level valuation confidence: what share of the fund, by market value, sits in low-confidence buckets?
Worked example, a 500 million USD fund:
| Confidence band | Market value | % of NAV |
|-----------------|-------------|----------|
| High (80 to 100) | 350m | 70% |
| Medium (50 to 79) | 100m | 20% |
| Low (0 to 49) | 50m | 10% |
10% of NAV is low-confidence. If your fund policy caps low-confidence at 15%, you strike. If a market event pushed it to 22%, you escalate to the valuation committee before striking, and you likely widen your price challenge activity: formally querying vendors to justify or revise a mark.
This roll-up becomes the single slide in the monthly valuation committee pack. It converts thousands of price points into a governance decision.
None of this holds without lineage. Every struck price needs a stored record of: which sources were available, which rung of the hierarchy was used, the dispersion, the confidence score, and who approved any override. This is data lineagedata lineageData lineage maps how data moves and transforms across systems, from origin to consumption, showing where it came from, what changed it, and where it goes.View full definition → (the traceable path from source to reported figure).
Regulators expect it. In the US the SEC's Rule 2a-5 under the Investment Company Act (effective since 2022) requires fund boards to oversee a documented fair value process, including testing of pricing vendors. In Europe, AIFMD (Alternative Investment Fund Managers Directive) mandates a valuation function independent of portfolio management. Both are, at their core, requirements for good pricing data governancedata governanceData governance is the set of policies, roles, and processes that ensure data is accurate, secure, well-defined, and used responsibly across an organization.View full definition →.