Multi-Constraint Bidding (MCB)

• Multi-Constraint Bidding (MCB) is a framework for optimizing bidding strategies under simultaneous constraints such as budgets, ROI, and eligibility, applicable across various auction settings.
• MCB is implemented in diverse environments including digital advertising with RTB systems like HALO, spectrum auctions using SAA methodologies, and procurement auctions with bidimensional mechanisms.
• Recent advances like hindsight-augmented learning and latent graph diffusion enhance MCB by enabling robust online optimization and efficient constraint handling amid dynamic shifts.

Searching arXiv for the cited MCB papers to ground the article and verify metadata. {"query":"(Dong et al., 5 Aug 2025) HALO Hindsight-Augmented Learning for Online Auto-Bidding", "max_results": 5} {"query":"Multi-Constraint Bidding HALO auto-bidding (Dong et al., 5 Aug 2025)", "max_results": 10} Multi-Constraint Bidding (MCB) denotes bidding problems in which an allocation or bidding policy is optimized subject to multiple simultaneously active constraints. In the cited literature, MCB appears in several distinct auction environments: Real-Time Bidding (RTB) for digital advertising, where total budget and ROI targets are central; Simultaneous Ascending Auctions (SAA), where exposure, own price effect, budget constraints, and eligibility management jointly shape feasible strategies; and procurement auctions, where private unit-cost and private capacity define a bidimensional mechanism-design problem. Across these settings, the common structure is not a single auction rule but the coexistence of multiple operational or strategic limits that must be enforced while maximizing value, utility, conversions, or expected reward (Dong et al., 5 Aug 2025, Pacaud et al., 2023, Bhat et al., 2015).

1. Scope and domain-specific meanings

The term MCB is used most explicitly in online advertising. In the RTB formulation of HALO, the advertiser seeks to maximize cumulative value subject to a total budget constraint $B$ and an ROI constraint $r_{\text{target}}$ (Dong et al., 5 Aug 2025). In MCMF, the relevant constraints are framed as multiple Key Performance Indicators (KPIs), including eCPC, eCPM, per-conversion cost (PPC), and daily budget, with a unified optimization-and-budget-management objective rather than a separated pacing layer (Wang et al., 2022). In the latent graph diffusion framework, MCB is formulated as constrained maximization over bidding trajectories under multiple KPI constraints such as budget, CPA, ROI, and CVR (Huh et al., 4 Mar 2025).

The SAA literature uses the same term differently. There, the MCB framework models spectrum auctions as an $n$-player simultaneous-move game with complete information, and the constraints are the exposure problem, the own price effect, budget constraints, and the eligibility management problem (Pacaud et al., 2023). In procurement auctions, the bidimensional mechanism of 2D-OPT and 2D-UCB treats private per-unit cost and private capacity as the two operative dimensions, and the exposition explicitly presents this as a natural extension to a general MCB setting (Bhat et al., 2015).

A common misconception is to treat MCB as a single fixed optimization template. The cited works instead use the term for several related formulations whose shared feature is simultaneous constraint handling, but whose state spaces, action spaces, observability assumptions, and objectives differ materially.

Setting	Constraints or dimensions	Representative formulation
RTB digital advertising	total budget, ROI, PPC, CPA, CVR, daily budget	HALO, MCMF, latent graph diffusion
Spectrum SAA	exposure, own price effect, budget constraints, eligibility management	$SMS^\alpha$
Procurement auction	unit-cost, capacity, unknown quality	2D-OPT, 2D-UCB

2. Canonical RTB formulation

A canonical MCB problem in RTB is defined over an impression sequence $\mathcal I=\{1,\dots,N\}$ within one bidding period. For each impression $i$, the advertiser has a private value $v_i\ge 0$, a market-clearing price $c_i>0$ that is unknown until after bidding, and an action $x_i\in\{0,1\}$ indicating whether the impression is won. Under a second-price auction, the advertiser pays $c_i$ if $r_{\text{target}}$0 (Dong et al., 5 Aug 2025).

The optimization target is

$r_{\text{target}}$1

subject to

$r_{\text{target}}$2

The special case BCB drops the ROI constraint (Dong et al., 5 Aug 2025). Prior closed-form analyses in the static one-shot setting yield

$r_{\text{target}}$3

for BCB, with $r_{\text{target}}$4 chosen so that $r_{\text{target}}$5, and

$r_{\text{target}}$6

for MCB. The stated limitation is that these formulas break down when constraints shift or when one must adapt online over billions of impressions (Dong et al., 5 Aug 2025).

The MCMF literature generalizes the RTB control perspective by writing the actual bid in eCPM as

$r_{\text{target}}$7

where $r_{\text{target}}$8 is a bid adjustment factor decided at time $r_{\text{target}}$9 (Wang et al., 2022). The latent graph diffusion formulation uses a trajectory view: for time horizon $n$0, agent $n$1 chooses bid vectors $n$2, and the objective is to maximize expected total reward under example constraints

$n$3

$n$4

This RTB line of work therefore spans both single-period constrained optimization and sequential planning over bidding trajectories (Huh et al., 4 Mar 2025).

3. Online learning and control in advertising MCB

HALO addresses two explicit deficiencies of traditional auto-bidding: severe sample inefficiency and limited generalization under constraint shifts. The paper characterizes the former as “near-zero transfer” across constraint configurations, because most methods perform trial-and-error per $n$5 pair and discard trajectories that overshoot budget or miss ROI. It characterizes the latter as failure to capture the underlying monotonic or derivative relationships in mappings such as $n$6 or $n$7; empirically, about one-third of campaigns adjust constraints mid-flight, causing large KPI volatility in prior methods (Dong et al., 5 Aug 2025).

Its core mechanism is hindsight experience with trajectory reorientation. HALO adopts a Fixed Coefficient Strategy (FCS): at decision step $n$8, it chooses a constant bid multiplier $n$9 for all remaining slots $SMS^\alpha$0 and records realized total cost and value $SMS^\alpha$1 and $SMS^\alpha$2. Each tuple

$SMS^\alpha$3

can be re-labeled as optimal data for budget $SMS^\alpha$4 (Dong et al., 5 Aug 2025). For ROI handling, HALO uses the stated monotonicity of $SMS^\alpha$5 and solves for a shading factor $SMS^\alpha$6 through the integral equation

$SMS^\alpha$7

with $SMS^\alpha$8, $SMS^\alpha$9, and $\mathcal I=\{1,\dots,N\}$0 (Dong et al., 5 Aug 2025).

HALO couples this hindsight reuse with a B-spline functional representation. It learns two continuous functions of remaining budget $\mathcal I=\{1,\dots,N\}$1,

$\mathcal I=\{1,\dots,N\}$2

where each $\mathcal I=\{1,\dots,N\}$3 is represented as a degree-$\mathcal I=\{1,\dots,N\}$4 B-spline with control points:

$\mathcal I=\{1,\dots,N\}$5

The stated advantages are local control, guaranteed smoothness up to order $\mathcal I=\{1,\dots,N\}$6, easy derivative $\mathcal I=\{1,\dots,N\}$7 for computing $\mathcal I=\{1,\dots,N\}$8, and more robust extrapolation than pointwise fits or black-box MLPs (Dong et al., 5 Aug 2025). A small MetaModel outputs the control points $\mathcal I=\{1,\dots,N\}$9 from state features such as time left, pacing, and p-value histograms, and the system is trained end-to-end with multi-anchor supervision from hindsight labels.

MCMF addresses a different triad of difficulties: sparse feedback, budget management separated from the optimization, and absence of bidding environment modeling. Its input is a “merging” vector that concatenates, for each KPI $i$0, the target value $i$1, the accumulated feedback $i$2, and additional accumulated features $i$3, including posterior features such as realized CTR and realized CVR and prior estimates such as average pCTR and average pCVR (Wang et al., 2022). The model is a two-layer MLP with

$i$4

Its cost function integrates KPI tracking error and control effort over a sliding window:

$i$5

The unknown auction-response derivative $i$6 is approximated by a Hebbian-rule sign term,

$i$7

so the updates do not require an explicit auction model (Wang et al., 2022). The paper explicitly states that MCMF never separates utility maximization from pacing, because budget and PPC constraints are folded directly into the same cost.

The latent graph diffusion framework introduces a planning-based latent diffusion model for large-scale auction environments. Each agent constructs a graph whose nodes encode impression opportunities and a special non-exposed node, with edges capturing intra-agent competition and additional cross-agent sharing. An attention-based GNN encoder maps the graph to a latent state, an inverse-dynamics model predicts bid vectors, and self-attention aggregation yields an approximate multi-agent equilibrium representation (Huh et al., 4 Mar 2025). Planning is performed in latent space via a standard Gaussian diffusion process with forward noising

$i$8

and reward alignment fine-tunes the posterior with

$i$9

The explicit aim is to generate auto-bidding trajectories that maximize KPI metrics while satisfying constraint thresholds (Huh et al., 4 Mar 2025).

4. Auction-theoretic and mechanism-design variants

In spectrum auctions, the MCB framework of Pacaud, Bechler, and Coupé’choux models SAA as a finite extensive-form game in which all bidders simultaneously choose subsets of remaining items, subject to hard budget and eligibility constraints (Pacaud et al., 2023). If bidder $v_i\ge 0$0 temporarily holds items $v_i\ge 0$1 and has eligibility $v_i\ge 0$2, its action set must satisfy

$v_i\ge 0$3

and

$v_i\ge 0$4

Payoff at auction close is

$v_i\ge 0$5

The algorithm $v_i\ge 0$6 is a Simultaneous-Move Monte Carlo Tree Search (SM-MCTS) procedure with a risk-averse reward

$v_i\ge 0$7

a learned closing-price forecast, UCB-style selection, legal-action expansion, point-price-prediction rollouts, and back-propagation of risk-averse returns (Pacaud et al., 2023). The stated purpose is simultaneous mitigation of the exposure problem, own price effect, budget exhaustion, and eligibility management.

The procurement literature frames a different MCB problem: a buyer procures $v_i\ge 0$8 identical units from heterogeneous strategic agents with private cost and capacity and unknown quality (Bhat et al., 2015). In the offline version, with known quality, the mechanism 2D-OPT uses an allocation rule and payments that maximize the auctioneer’s expected utility under Bayesian Incentive Compatibility (BIC) and Individual Rationality (IR). The key quantity is the virtual cost

$v_i\ge 0$9

and the allocation rule computes

$c_i>0$0

sorts agents in decreasing $c_i>0$1, and greedily takes as many units as allowed until $c_i>0$2 units are chosen or $c_i>0$3 becomes nonpositive (Bhat et al., 2015). The payment rule is

$c_i>0$4

In the online version, 2D-UCB learns unknown qualities through UCB estimates

$c_i>0$5

combined with self-resampling so that the induced mechanism remains Stochastic-BIC and IR (Bhat et al., 2015).

These auction-theoretic variants show that MCB is not restricted to ad-delivery platforms. The same label is used for simultaneous-move strategic games, constrained search over legal actions, and truthful multidimensional mechanisms with online learning.

5. Structural guarantees and optimality claims

HALO’s analysis is built around the fixed-coefficient paradigm. Lemma 1 gives an FCS error bound: any fixed-coefficient strategy that exactly exhausts budget is within $c_i>0$6 of the omniscient optimum $c_i>0$7. Lemma 2 states the BCB optimality condition: if $c_i>0$8, then $c_i>0$9 is optimal on $x_i\in\{0,1\}$0. Lemma 3 states ROI monotonicity: $x_i\in\{0,1\}$1 is non-increasing in $x_i\in\{0,1\}$2, which guarantees a unique shading factor $x_i\in\{0,1\}$3 for ROI compliance. Lemma 4 and Theorem 1 provide the closed-form integral equation for $x_i\in\{0,1\}$4 and prove the global optimality of the two-stage rule that deploys $x_i\in\{0,1\}$5 if ROI is nonnegative and otherwise shades by $x_i\in\{0,1\}$6 (Dong et al., 5 Aug 2025). The paper’s interpretation is that every rollout can be turned into correct training data for any budget/ROI pair.

The procurement mechanism relies on envelope conditions. A mechanism is BIC and IR if, for each agent, interim allocation is non-increasing in reported cost, interim offered utility is non-decreasing in reported capacity and nonnegative, and the offered utility satisfies the integral identity

$x_i\in\{0,1\}$7

Under a regularity assumption that $x_i\in\{0,1\}$8 is non-decreasing in cost and non-increasing in capacity, the resulting mechanism is DSIC, IR, and utility-maximizing for the auctioneer (Bhat et al., 2015).

The SAA framework does not present an analogous closed-form optimality theorem in the provided summary, but it does encode feasibility directly in the action sets and embeds risk-aversion into the reward function. Its empirical-game analysis states that every unilateral deviation to $x_i\in\{0,1\}$9 is strictly profitable against any profile of rivals using the listed baselines, making $c_i$0 a clear Nash equilibrium (Pacaud et al., 2023).

A plausible implication is that MCB research splits into two broad proof styles. One style proves monotonicity, envelope identities, or near-optimality under simplified policy classes. The other enforces constraints algorithmically through action-space restriction, trajectory re-labeling, or reward alignment and evaluates performance empirically.

6. Empirical evidence, recurring difficulties, and research directions

The HALO evaluation uses AuctionNet with 21 days, approximately 500K impression events per day, and 48 advertisers per day, split into 14 days train and 7 days test. In MCB mode, the reported metrics are average conversions when the ROI constraint is satisfied (“Conv”) and compliance rate (C.R.); in BCB mode, they are conversions and cost-to-budget ratio (C/B). Across budgets from 50% to 150% of a normalized baseline, HALO attains the highest Conv in MCB while keeping $c_i$1 in all budgets, and in BCB it reaches $c_i$2 with the highest Conv in every scenario. The reported final realized ROI/target is $c_i$3 for HALO versus $c_i$4 for BC, $c_i$5 for IQL, and $c_i$6 for LP (Dong et al., 5 Aug 2025).

MCMF is evaluated on the iPinYou open dataset and in Alibaba RTB production. On iPinYou, under Single/Adequate conditions, MCMF achieves 33 conv at PPC approximately 786 FEN versus RL’s 33 conv at PPC approximately 1519 FEN; under Multi/Tight, it achieves 26 conversions with PPC approximately 707 FEN versus RL’s 3 conversions at PPC approximately 1724 FEN. Across all four setups, the paper reports that MCMF maximizes conversions, never exceeds the PPC target, and spends exactly the allotted budget. In the reported online A/B test with 7 K SKUs, 22 K advertisers, 4 M bids per day, and $c_i$7 requests per day, over 15 days MCMF produced $c_i$8 conversions, $c_i$9 ROI, and $r_{\text{target}}$00 cost (Wang et al., 2022).

In SAA experiments on realistic instances with $r_{\text{target}}$01 and $r_{\text{target}}$02, when all four bidders play $r_{\text{target}}$03, the average per-bidder utility is roughly $r_{\text{target}}$04, versus $r_{\text{target}}$05 under EPE and negative under SB. Exposure frequency falls below $r_{\text{target}}$06 for $r_{\text{target}}$07, compared with $r_{\text{target}}$08–$r_{\text{target}}$09 for the baselines, and expected losses are $r_{\text{target}}$10–$r_{\text{target}}$11 lower. Average price paid per licence is $r_{\text{target}}$12–$r_{\text{target}}$13 lower than under SB or SCPD and $r_{\text{target}}$14–$r_{\text{target}}$15 lower than under EPE when facing the same opponents (Pacaud et al., 2023).

The latent graph diffusion framework reports results on AuctionNet and a synthetic auction dataset. On AuctionNet, against DiffBid, the reported metrics are Return $r_{\text{target}}$16 versus $r_{\text{target}}$17, CPA $r_{\text{target}}$18 versus $r_{\text{target}}$19, ROI $r_{\text{target}}$20 versus $r_{\text{target}}$21, CVR $r_{\text{target}}$22 versus $r_{\text{target}}$23, budget adherence approximately $r_{\text{target}}$24 for both, and social welfare $r_{\text{target}}$25 versus $r_{\text{target}}$26. On the synthetic dataset, the corresponding results are Return $r_{\text{target}}$27 versus $r_{\text{target}}$28, CPA $r_{\text{target}}$29 versus $r_{\text{target}}$30, ROI $r_{\text{target}}$31 versus $r_{\text{target}}$32, CVR $r_{\text{target}}$33 versus $r_{\text{target}}$34, budget adherence approximately $r_{\text{target}}$35, and social welfare $r_{\text{target}}$36 versus $r_{\text{target}}$37 (Huh et al., 4 Mar 2025).

Several recurring technical difficulties emerge directly from these studies. The RTB literature emphasizes sparse and delayed feedback, severe sample inefficiency, mid-flight constraint changes, and catastrophic extrapolation outside the training range (Dong et al., 5 Aug 2025, Wang et al., 2022). MCMF explicitly notes that the choice of $r_{\text{target}}$38, $r_{\text{target}}$39, $r_{\text{target}}$40, and $r_{\text{target}}$41 requires careful tuning, and that the sign-based gradient approximation can be noisy under extremely erratic auction environments (Wang et al., 2022). The procurement formulation identifies multidimensional ironing, multi-dimensional critical-value integrals, and regret control when multiple dimensions affect ranking gaps as key technical challenges for general MCB extensions (Bhat et al., 2015). Taken together, these results indicate that the central research problem in MCB is not merely constrained optimization, but robust adaptation under shifting constraints, partial feedback, strategic interaction, and domain-specific feasibility rules.