Analyzing the Airbnb Market: A Multivariate Pricing Strategy for Boston and DC Short-Term Rentals

Executive Summary

Entire homes list for $73+ more per night than shared rooms, and Boston listings run about $18 above Washington, D.C. This Excel regression of 1,382 Airbnb listings shows how room type, city and guest capacity shape price.

Many hosts set prices by gut feel or a flat rule, which can backfire: price a big listing too high and it may sit empty, too low and money is left on the table. Built for an MBA course, this project starts with a one-variable regression and improves it step by step into a model that puts a dollar value on what drives price.

This project investigates three core questions:

1. Room Type & City: How much more do entire homes and private rooms list for than shared rooms, and how much does Boston add over D.C.?
2. Guest Capacity: How does price change as capacity grows, and does it change differently for entire homes?
3. Model Fit: How much of the variation in price can a simple Excel regression explain, and what does it leave out?

---

Skills Demonstrated

• Data cleaning: Handled missing values, treated outliers with the IQR rule and encoded categorical variables.
• Statistical modeling: Built multiple regression models with quadratic and interaction terms and tested significance.
• Model diagnostics: Used residual plots and a correlation matrix to spot problems and improve each version.
• Business communication: Turned coefficients into plain-English pricing advice and a presentation deck.

---

Presentation Deck

A slide summary of the project, from the problem to the recommendations.

---

Step 1: Preparing the Data

1. Cleaned the basics.
   • Dropped ID columns that don’t help predict price (room_id, survey_id, host_id) and set nightly price (USD) as the target.
   • Found 2 missing bedrooms values and filled them with 1. Both were entire homes for 2 guests, so this was a safe call.
2. Dealt with outliers.
   • Prices were heavily skewed by luxury listings (skewness of 3.77, mean of $194.25 vs. median of $129.50).
   • I used the IQR rule (upper cutoff of $372.50) and removed 140 listings, about 10% of the data.
   • The remaining 1,242 listings had a much more balanced distribution (skewness of 0.95), which suits regression.
3. Encoded categories.
   • Turned room type and city into 0/1 variables as is_entireHome, is_privateRoom, is_boston.
   • The baseline for comparison is a shared room in Washington, D.C.
4. Checked for overlapping variables.
   • bedrooms and accommodates were strongly correlated (r = 0.69), so they would blur each other’s effect.
   • I kept accommodates because it relates more closely to price (r = 0.49 vs. 0.40) and dropped bedrooms.

Price Metric	Raw Data (N = 1,382)	Cleaned Data (N = 1,242)	What Changed
Mean price	$194.25	$136.44	Luxury outliers no longer pull the average up
Median price	$129.50	$120.00	The typical price barely moved
Standard deviation	$216.05	$75.79	64.9% less volatility
Skewness	3.770	0.947	Much closer to a normal shape

---

Step 2: Building the Model in Three Rounds

Round 1: Simple Linear Regression

• Predicted price using only guest capacity (accommodates).
• Result: about +$19.96 per extra guest on top of a $73.94 starting point. It was statistically significant but explained only 24.18% of price variation.
• What the residuals showed: the residual plot formed vertical stripes instead of random scatter. That told me important categories, like room type, were missing from the model.

Round 2: Multiple Linear Regression

• Added room type, city, review count (reviews) and overall_satisfaction.
• overall_satisfaction was not statistically significant (p = 0.78), so I removed it.
• Result: Adjusted R² rose to 39.10% and the stripes disappeared. The model still over-predicted prices for larger groups.

Round 3: Non-linear model (final)

• Added two terms to fix the large-group problem:
  • accommodates_squared, so the model can show price gains slowing as groups get bigger.
  • accommodates × is_entireHome, because an extra guest is worth more in an entire home than in a room.
• All seven predictors are significant (p < 0.05). Adjusted R² reached 39.54% and the standard error fell to $58.93.

Model	R²	Adjusted R²	Std. Error	Key Takeaway
1. Simple linear	24.18%	24.12%	$66.02	Striped residuals: missing variables
2a. Multiple linear (full)	39.35%	39.05%	$59.17	overall_satisfaction not significant
2b. Multiple linear (pruned)	39.35%	39.10%	$59.15	Stripes gone, but large groups over-predicted
4. Non-linear (final)	39.88%	39.54%	$58.93	Captures slowing gains and the entire-home effect

---

The Final Model

Predicted Price = 26.91 + 73.17(EntireHome) + 30.94(PrivateRoom) + 17.82(Boston)
                  - 0.17(Reviews) + 13.56(Accommodates) - 0.59(Accommodates²)
                  + 6.52(Accommodates × EntireHome)

The baseline is a shared room in Washington, D.C.

Variable	Coefficient	Std. Error	t-Stat	P-Value	95% Confidence Interval
Intercept (shared room, D.C.)	26.91	12.61	2.13	0.0331	[$2.16, $51.66]
is_entireHome	+73.17	13.37	5.47	< 0.0001	[$46.94, $99.40]
is_privateRoom	+30.94	11.72	2.64	0.0084	[$7.93, $53.94]
is_boston	+17.82	3.44	5.18	< 0.0001	[$11.08, $24.57]
reviews	-0.17	0.04	-3.85	0.0001	[-$0.26, -$0.08]
accommodates	+13.56	3.36	4.04	< 0.0001	[$6.97, $20.15]
accommodates_squared	-0.59	0.24	-2.49	0.0128	[-$1.05, -$0.13]
interaction_accom_entireHome	+6.52	2.70	2.41	0.0160	[$1.22, $11.82]

What the numbers mean

• Room type matters most. A private room adds $30.94 per night over a shared room. An entire home adds $73.17, which is $42.23 more than a private room, before counting guest capacity.
• Boston costs more. With room type, capacity and reviews held equal, Boston listings average $17.82 more per night than D.C.
• More reviews, slightly lower price. Each extra review goes with a $0.17 lower price, so 100 reviews is about $17 lower. This likely reflects budget listings that book often, not reviews hurting a listing’s value.
• Extra guests add less each time. Guest capacity has a curved effect, not a straight line.
  • For private and shared rooms, the gain per guest shrinks steadily and levels off at around 11 to 12 guests.
  • For entire homes, the interaction term lifts the starting gain to +$20.08 per guest (13.56 + 6.52).
  • Example: a Boston entire home with 24 reviews is predicted at $133.31 for 1 guest, $184.70 for 4 guests and $236.70 for 8 guests.

---

Business Recommendations

1. Look closely at entire homes for 4 to 8 guests. An entire home lists for about $73 more per night than a shared room at baseline, and the gap grows by about $6.50 for each extra guest (roughly $86 at 2 guests).
2. Build the Boston premium into pricing. Anyone comparing the two cities should expect about $17.82 more per night in Boston and factor that in when setting rates or evaluating a property.
3. Use tiered pricing instead of a flat extra-guest fee. The model shows each added guest is worth a bit less than the one before, by roughly $1.18. A flat fee likely overcharges larger groups, so a fee that tapers down as the group grows is a better fit.

---

Bottom Line

Room type and city explain most of what hosts can predict about price. Entire homes carry a premium that widens with group size and Boston runs about $18 above D.C. The model is a useful pricing benchmark, but it can’t show profit or demand, which the data doesn’t include.

---

Limitations

• It predicts price, not profit. The data has no occupancy, booking, or cost information. The model shows where prices run higher, not which listings earn the most.
• Important details are missing. Amenities like parking, renovation quality and distance to transit or landmarks are not in the data. They likely account for much of the 60% of variation the model does not explain.
• City-level view only. The model treats every Boston neighborhood the same and every D.C. neighborhood the same. A next version could add neighborhood or zip-code effects and model ln(Price) to better handle expensive listings.
• One snapshot in time. The data can’t show seasonality, local events, or weekday vs. weekend demand.

---

Data Source & Attribution

Airbnb listing data for Boston and Washington, D.C., based on the open-source Airbnb Data Collection by Tom Slee, which gathers public listings from the Airbnb website. I received the 1,382-listing sample (534 in Boston, 848 in D.C.) through my MBA course, MIS 657: Business Analytics. Prices are nightly rates in US dollars. This project was done for academic purposes only.