Compute GSP vs VCG payments with reserves

Quick Overview

This question evaluates competency in auction theory and mechanism design, focusing on revenue and welfare analysis with quality-adjusted bids, reserve pricing, and strategic behavior such as bid shading and budget pacing.

Compute GSP vs VCG payments with reserves

Company: Roblox

Role: Data Scientist

Category: Statistics & Math

Difficulty: medium

Interview Round: HR Screen

Consider a single query with 2 ad slots. Position CTRs are [0.10, 0.06]. Three advertisers have bids b and quality scores q as: A: b=3, q=2.0; B: b=4, q=1.5; C: b=5, q=1.0. Reserve price r=1.0. Ranking is by b·q. Answer: 1) Under GSP with quality scores, determine the allocation and per-click prices. Use the standard payment formula p_i = (next advertiser’s b·q)/q_i (respecting reserve r). Compute expected revenue and welfare (sum over slots of CTR_slot × value per click where value≈true bid; assume bids reflect values for this part). 2) Under VCG with quality scores and the same reserve, compute each winner’s per-click price and the platform’s expected revenue; compare to GSP. 3) Show a profitable bid shading deviation for B (if one exists) under GSP given A and C bid truthfully. Identify a shading range that preserves B’s slot while reducing price. 4) If each advertiser has a daily budget cap that induces pacing (effective bid multiplier α∈(0,1]), explain how pacing interacts with the quality-adjusted ranking and which mechanism (GSP vs VCG) better preserves truthful reporting under budget constraints.

Quick Answer: This question evaluates competency in auction theory and mechanism design, focusing on revenue and welfare analysis with quality-adjusted bids, reserve pricing, and strategic behavior such as bid shading and budget pacing.

|Home/Statistics & Math/Roblox
Roblox logo
Roblox
Oct 13, 2025
mediumData ScientistHR ScreenStatistics & Math
6
0

You are analyzing a single-query auction with 2 ad slots. Position CTR multipliers are [0.10, 0.06]. Three advertisers have bids b and quality scores q:

  • A: b=3, q=2.0
  • B: b=4, q=1.5
  • C: b=5, q=1.0

Reserve price r = 1.0. Ads are ranked by the quality-adjusted score s_i = b_i · q_i. Assume ties are broken in favor of higher quality q (A over B). Assume the standard separable-click model: the expected click probability for advertiser i in slot s is CTR_s × q_i.

Answer the following:

  1. GSP with quality scores
    • Determine the allocation and per-click prices using the standard payment formula p_i = max(r, (next advertiser’s b·q)/q_i).
    • Compute the platform’s expected revenue per impression and total welfare per impression, where welfare = sum over filled slots of (CTR_slot × q_i × value per click) and value per click ≈ bid (assume bids reflect values for this part).
  2. VCG with quality scores (same reserve)
    • Compute each winner’s per-click price and the platform’s expected revenue per impression. Compare to GSP.
  3. Bid shading for B under GSP
    • Given A and C bid truthfully, identify whether a profitable bid shading deviation exists for B under GSP. Specifically, is there a shading range that preserves B’s slot while reducing B’s per-click price? If not, explain why; if yes, give the range and the impact on B’s utility.
  4. Budgets and pacing
    • Suppose each advertiser has a daily budget cap that induces pacing via an effective bid multiplier α ∈ (0, 1], so the effective rank score is α·b·q. Explain how pacing interacts with quality-adjusted ranking and which mechanism (GSP vs. VCG) better preserves truthful reporting under budget constraints.
Loading comments...