How to pull comps for a house flip

An after-repair value for a flip is a statement about sold houses, not about the house you are standing in and not about the price the seller hopes for. The report builds that value as a weighted median of adjusted sale prices. You can apply the same weights by hand when you are deciding which sales belong on the list and which ones are only noise. This page is the weight table. It does not pull a live feed, and the sample prices below are chosen so the arithmetic is visible.

Start with closed sales

A listing is an ask. A pending sale can still fall apart or change price. The value the formula consumes is a sold price, then an adjusted price if bedroom or bathroom counts differ from the subject. Active and pending listings can tell you what else is for sale. They do not get a weight in this median. If your list is mostly listings, you do not have an ARV yet. You have a set of asks, and the flip profit worksheet will treat whatever number you type as if buyers already paid it.

Price per square foot is stored on each comp and shown for context. The median itself is taken on the adjusted sale price, not on the price-per-square-foot rank. Ordering by price per foot would pick a different house whenever the comps differ in size. A small house with a high price per foot can outrank a closer match if you sort the wrong column.

Recency, distance, and similarity

Every comp starts at a weight of 1. Three multipliers then stack. Recency uses the sold date. A sale within 3 months is multiplied by 3. A sale within 6 months, but older than 3, is multiplied by 2. A sale within 12 months is multiplied by 1.5. Anything older than 12 months is multiplied by 0.8. Older sales are not deleted. They are simply lighter, so a thin market does not collapse to “no value,” and a stale sale cannot outvote a new one unless it is the only evidence you have.

Distance is in miles. A sale at or inside 0.25 miles is multiplied by 3. Inside 0.5 miles the multiplier is 2.5. Inside 1 mile it is 2. Inside 2 miles it is 1.5. Farther than 2 miles it stays at 1. A missing distance, a distance that is not a finite number, or a distance of zero or below gets a neutral 1x as well. Zero is treated as unknown because that is how a provider stores a gap, and a missing distance must not win the quarter-mile bonus. The check for “unknown” has to happen before the quarter-mile test, because a missing value would otherwise look closer than 0.25.

Similarity is a score from 0 to 1 that the comp selection already attached. If that score is above zero, the weight is multiplied by 1 plus the score. A similarity of 0.8 becomes a 1.8 multiplier. A similarity of 0 adds nothing beyond the recency and distance you already applied. The three multipliers compound. They are not averaged, and they are not a 1-to-10 star rating you can swap in from a listing site.

A two-comp weighted median

Take two adjusted prices, invented so you can see the median move. Comp A sold for an adjusted $210,000, two months ago, 0.2 miles away, with similarity 0.8. Comp B sold for an adjusted $180,000, ten months ago, 1.5 miles away, with similarity 0.4.

• Comp A weight is 3 (recency) × 3 (distance) × 1.8 (similarity) = 16.2.
• Comp B weight is 1.5 × 1.5 × 1.4 = 3.15.
• Total weight is 19.35. Half of that is 9.675.

Sort by price, low to high, and walk until the cumulative weight reaches half the total. Comp B is first. Its 3.15 is still under 9.675, so the median is not $180,000. Comp A’s weight pushes the cumulative total past halfway, so the weighted median is $210,000. A plain average of the two prices is $195,000. The average let the older, farther sale pull the value down by $15,000. The weighted median stayed with the closer, newer sale because that sale holds most of the weight. Add a third comp and the same walk applies: sort by adjusted price, accumulate weight, stop at half.

The reported range is not a second opinion. After the median is chosen, the report looks at the standard deviation of the adjusted prices and sets a low and a high around the median, clamped so the low is not below the cheapest comp and the high is not above the dearest. A wide range means the sold prices disagree. It does not mean you should pick the high end for a flip because the profit looks better there. Profit calculated at the high end of a wide range is the optimistic case. The MAO calculator will obediently raise the offer if you feed it that high end.

Bedroom and bathroom adjustments

Before the median, a comp can be adjusted so a 2-bedroom sale is comparable to a 3-bedroom subject. The adjustment is minus (comp beds minus subject beds) times a per-bed dollar rate, and the same shape for baths. A comp with fewer beds than the subject gets a positive adjustment: it would have sold for more if it had matched. A comp with more beds gets a negative adjustment.

The dollar rate prefers evidence from the comp set itself. The report looks at pairs of comps that differ in bedrooms but match on bathrooms, are within 20 percent of each other on square footage, and were built within 10 years. It needs at least two such bedroom pairs and two bathroom pairs. The rate is then the median of (price difference / feature difference), clamped between $5,000 and $50,000 per bedroom and between $2,500 and $30,000 per bathroom. If there are not enough pairs, it falls back to 3 percent of the median comp sold price per bedroom and 1.5 percent per bathroom, with the same clamps. The fallback percent is of the median sold price of the comps, not of your hoped-for ARV. The pair arithmetic, a set the gates drop, and the sign of the adjustment are worked out on bedroom and bathroom comp adjustments.

Do not also “feel adjust” for beds after this step. A second haircut for the same bedroom gap double-counts it. Condition, a busy road, and a finished basement are not in this bed-and-bath adjustment. If you need to drop a comp for those reasons, drop it or lower its influence by leaving it out. Inventing an extra dollar tweak and still calling the result the report’s median is a different number.

What the confidence score is made of

Confidence is a 0 to 100 score on the comp set, not a promise about the next sale. Up to 30 points come from how many comps you have: the count times 6, capped at 30, so five comps already fill that bucket and a sixth does not add count points. Up to 25 points come from the average similarity, times 25. A comp with no similarity stored is treated as 0.5 for this average only. Up to 25 points come from the share of comps that sold within six months. Up to 20 points come from how tightly the prices cluster. The report uses the coefficient of variation, standard deviation divided by the mean, and scores max(0, 20 − that ratio × 100). A wild spread can zero out those 20 points even when you have a lot of recent sales.

A high confidence score means the sold prices agree and the sales are recent and numerous. It does not mean the rehab will cost what you hoped, and it does not set the wholesale fee. Once you trust the ARV enough to type it into an offer, the fee conversation is about the buyer’s ceiling, which is the subject of how to price a wholesale assignment fee.

Questions