How BuyerEdge Calculates Your Bid Strategy: The Full Statistical Methodology

The Information Asymmetry Problem in NSW Property

The NSW residential property market presents a structurally asymmetric information environment. Sellers — and more specifically, their agents — routinely possess materially better market intelligence than buyers: access to offer histories, inspection volumes, vendor timelines, and a continuous stream of comparable transaction outcomes that buyers have no systematic access to. This asymmetry is not incidental; it is the operating model of the traditional property sales process.

Conventional approaches to addressing this asymmetry — agent appraisals, automated valuation models trained on listing data, comparative market analyses derived from current guide prices — share a fundamental data quality deficiency: they conflate supply-side signals (what sellers are asking) with demand-side outcomes (what buyers actually paid). For a buyer attempting to establish a defensible maximum offer, this conflation is not an academic concern. Guide prices in supply-constrained NSW submarkets are routinely set at or below the expected sale price to spark competition — they are strategic anchors, not market prices.

BuyerEdge was built on a single methodological premise: that only registered closed-sale data constitutes reliable ground truth for offer strategy modelling, and that any analysis incorporating guide prices, agent estimates, or listing platform data introduces a systematic bias that is structurally disadvantageous to buyers.

The NSW Valuer General as Data Foundation

The NSW Valuer General, administered by the NSW Government, records residential property transactions across New South Wales. The data records the registered sale price — not the guide price, not the agent's appraised value, but the price at which a binding, settlement-completed transaction occurred.

Our dataset contains approximately 379,000 residential transactions spanning over a decade of NSW market cycles. This longitudinal depth is analytically critical — it provides sufficient historical density to compute robust time-adjustment factors at a local level, rather than relying on national or regional indices that obscure significant intra-market variation.

Every BuyerEdge analysis is derived exclusively from Valuer General data. No guide price data, no listing platform estimates, no agent valuations, and no automated valuation model (AVM) outputs are used at any stage of the calculation pipeline.

Comparable Search: Geographic and Temporal Ladder

For a given subject property, the engine uses a progressive search ladder until at least six comparables are found. The starting configuration is 1.5km radius and 18 trailing months. If that yields fewer than six comparables, the radius is expanded to 3km (still 18 months). If still insufficient, the time window is extended to 24 months at 3km. A final level relaxes property-type grouping (e.g. Semi-D/Terrace/End of Terrace treated as one pool) while keeping 3km and 24 months. There is no 36-month window in the implementation.

Within the selected window, transactions are filtered by property type (with grouping for apartments and for house types), and by bedroom count ±1 where available. The target is 6–20 comparables; the first level that returns at least six is used, and that level's radius and time window are reported in the output (e.g. "1.5km, 18 months" or "3km, 24 months").

Similarity Scoring and Same-Development Boost

Where the subject has floor area, each comparable with floor area receives a similarity score (0–1) from four components: size (40%), distance (30%), recency (20%), and property-type match (10%). Size score is 1 at exact match and 0 at ±100% deviation. Distance score is 1 at 0km and 0 at 5km+. Recency is linear decay: 1.0 at 0 months age, 0 at 36 months — there is no exponential half-life. Type score is 1 for exact match, 0.5 for group match (e.g. Semi-D vs Terrace).

Comparables from the same estate or development receive a 1.5× weight multiplier. Estate tokens are extracted from the address (e.g. "Dun Emer Way" → [dun, emer]); if the subject and a comparable share at least two such tokens, that comparable's weight is multiplied by 1.5 before the weighted median and fair-value calculations. This reflects the higher informational value of sales in the same scheme.

Hard size filters are applied before scoring: comparables outside 0.6× to 2× the subject's floor area are excluded from the valuation comp set. Among the remainder, a high-similarity subset (score ≥ 0.35) is defined; if there are at least four high-similarity comps, that set is used for size-adjusted quantiles and fair value; otherwise the similar-size set is used.

Time-Adjustment and $/m² Trend

Price direction is computed from the local $/m² trend: median $/m² of comparables in the last 12 months vs median $/m² of comparables in the 12–24 month window. If at least four comps exist in each bucket, an annual growth rate is derived and used to time-inflate historical sale prices to a current-date equivalent: price × (1 + annualGrowthRate × (ageMonths / 12)). Where floor area is insufficient for a reliable trend, no time adjustment is applied and the absence is flagged. Trend thresholds: >3% annual → "Rising", <−3% → "Falling", else "Stable".

Winsorisation and Size-Adjusted Fair Value

Before computing the size-time-adjusted fair value, $/m² values are winsorised: each comp's raw $/m² is capped at the 10th and 90th percentiles of the distribution. This limits the influence of extreme $/m² outliers on the weighted median. A size-elasticity correction is then applied (exponent 0.85: price ∝ m²^0.85) so that larger comps do not linearly inflate $/m² for a smaller subject.

The resulting size-time fair value is a weighted median of time-inflated, winsorised, size-corrected $/m² × subject floor area. This value is then blended with the raw P50 to produce the reported fair value: (1) if size dispersion of the valuation set is high (max/min > 1.8×), blend 75% raw P50 + 25% size-time fair; (2) if the subject's size is far from the median comp size (>35% deviation), blend 40% P50 + 60% size-time fair; (3) otherwise 60% P50 + 40% size-time fair. With only 3–4 size comps, a lighter blend (85% P50 + 15% size-time) is used. This blending avoids over-relying on a single methodology when the comp set is heterogeneous.

Distribution Parameters: P25, P50, P75, IQR

Raw P25, P50, P75 are computed from the full comparable price list (or trimmed-mean approximations when n<8). When at least four size-similar comps exist, size-adjusted quantiles (adjP25, adjP50, adjP75) are computed from time-inflated prices of that subset; the offer strategy anchors to these when available, avoiding distortion from size-dissimilar comps in the raw set. IQR = P75 − P25; it drives both confidence interpretation and the spread of the offer band (entry, sealed, ceiling).

Confidence Scoring (Exact Rules)

Confidence is three-tier and based on count and recency only in the core logic:

• High: ≥10 comparables and weighted median transaction date < 12 months old. Both conditions are required.
• Moderate: ≥6 comparables (with no recency requirement for this tier).
• Low: <6 comparables. The threshold is 6, not 5.

Additional overrides: if the distribution fails sanity checks (P75 < P50 or P50 < P25), confidence is forced to Low. If the price range is very wide (P75/P25 > 1.8) and valuation comp count < 4, confidence is forced to Low. Under Moderate, confidence can be promoted to High when there are ≥8 high-similarity comps and a non–price-only fair value method. Confidence feeds into the Confidence Adjustment Factor (CAF) and into ceiling compression.

Confidence Adjustment Factor (CAF)

The CAF is a multiplier applied to the distribution-derived ceiling and sealed bid to reflect dataset quality. Base values: High = 1.0, Moderate = 0.97, Low = 0.93. If comp count < 8, an additional ×0.98 is applied; if median recency > 12 months, another ×0.98. The final CAF is clamped to [0.90, 1.00]. So under Low confidence with thin or stale data, the ceiling and best-and-final are pulled down by up to 10%.

Active Supply and Months Supply

The system counts active listings in a bounding box around the subject (latitude ±0.027°, longitude ±0.045°, approximating a few km²), filtered by the same property type and bedroom range. This count is divided by the local monthly sales rate (derived from comparables sold in the last 90 days) to obtain months supply, capped at 12. Months supply feeds directly into seller leverage and sealed-bid probability; it is not mentioned in listing copy but is a core input to the strategy. Where 90-day sales are too low to be reliable, months supply is not reported but the leverage formula still uses a capped value.

Seller Leverage (0–10): Actual Inputs

Seller leverage is a weighted sum of three components (total scaled to 0–10):

• Above-asking share (40% weight, max 4 points): Among comparables that have recorded asking prices, the proportion that closed above asking, scaled linearly (e.g. 50% above asking → 2 points). Requires at least five comps with asking data.
• Months supply (40% weight, max 4 points): (1 − monthsSupply/6) clamped to [0, 1], then ×4. Lower supply → higher leverage.
• Price momentum (20% weight, max 2 points): From median sale price in last 90 days vs median in last 180 days; momentum = (median90 − median180)/median180, scaled and clamped. Rising recent prices add up to 2 points.

Time-on-market for the subject property is not used. Asking price relative to median is not a direct input; it affects the report narrative and the "asking position" (above/within/below range) but not the leverage score itself.

Sealed Bid Probability (0–75%)

A separate metric, sealed bid probability, estimates the likelihood of a sealed/best-and-final process. It is a linear combination (clamped 0–0.75): 15% weight from above-asking share, 35% from (1 − monthsSupply/6), 25% from momentum, 25% from leverage/10. So tight supply, high above-asking rates, positive momentum, and high leverage all increase the score. This value is reported as a percentage and drives sealed-bid guidance in the narrative (e.g. when ≥20%, the report notes that if a sealed bid is called, the modelled best-and-final figure is the data-derived position).

Deriving the Offer Band: Entry, Sealed, Ceiling

The three outputs — Opening Position (entry), Best-and-Final (sealed), Statistical Ceiling (hardCeiling) — are derived in a fixed order so that entry ≤ sealed ≤ hardCeiling is always satisfied. All values are rounded to the nearest $5,000 (or $2,500 when asking < $350k).

Opening Position (Entry)

Entry is not directly anchored to P25. It is computed as sealed − (iqrFactor × IQR), where iqrFactor depends on asking position (below/within/above). A floor of 95% of the anchor quartile (sP25 when using size-adjusted quantiles) is applied, so P25 acts as a lower bound, not the anchor. Entry is then capped by asking price when asking ≥ sP25 (asking-price discipline): the opening offer must not exceed the list price except in "bait" pricing (asking < sP25). After all adjustments, entry is again constrained to be ≤ sealed and ≥ floor.

Statistical Ceiling and Best-and-Final

The initial ceiling is sP75 for within/below asking position, or sP50 + 0.25×IQR for above-priced listings. The sealed bid is placed between sP50 and sP75 depending on position (below: near median; above: sP50 + 0.2×IQR; within: sP50 + 0.5×IQR). Both are then multiplied by the CAF. For listings at or below median (and not genuinely underpriced), additional caps apply: sealed is capped as a percentage above asking (e.g. 8% High, 6% Moderate, 4% Low), and ceiling is capped similarly (15%/12%/8%). Final ceiling and sealed are rounded to $5k; entry is forced ≤ sealed and ≥ 95% of sP25.

Post-Processing: applyFinalConstraints

After the raw offer strategy is computed, a six-constraint layer is applied to catch upstream noise (e.g. no floor area → raw quantiles from size-dissimilar comps → inflated ceiling):

• C0 (no-adj backstop): When size-adjusted quantiles are missing, ceiling is capped at asking × 1.15 (Low), 1.20 (Moderate), or 1.25 (High).
• C1 (distribution clamp): Ceiling ≤ adjP75 × 1.01 (Low), 1.02 (Moderate), or 1.03 (High).
• B1 (opening ≤ asking): Opening is capped at asking unless seller leverage ≥ 8, High confidence, and asking < adjP25 (bait exception).
• B2 (best-and-final premium cap): Best-and-final cannot exceed asking by more than a confidence- and property-type-dependent percentage (e.g. houses: 12%/8%/5% for High/Moderate/Low; apartments slightly lower); a small density bonus applies in high-density markets.
• A (monotonic ordering): entry ≤ bestFinal ≤ ceiling enforced after other clamps.
• E (minimum separation): bestFinal ≥ entry + increment (one step).

Any applied clamp is logged for debugging; the user-facing output is the final, constraint-satisfying triple (entry, sealed, ceiling).

What This Methodology Does Not Do

BuyerEdge does not predict the price at which a specific property will sell. It does not incorporate subjective quality factors — renovation standards, aspect, garden condition, internal layout — that are not captured in the Valuer General data. It does not constitute financial, legal, mortgage, or professional property advice. And it does not account for the buyer's personal financial circumstances.

What it does provide is an analytically defensible, data-grounded framework for entering a negotiation with a defined position, a quantified ceiling, and a structured escalation plan — replacing a reactive, emotionally driven bidding process with one anchored to what buyers have actually paid for comparable properties in the recent past. Try our free tools: over-asking probability, overpay risk, and negotiation leverage; or get the full framework in a single property report.