Research
Mandate without Managers: Automated Market Makers as Verifiable Portfolio Products
Target-weight mandates, such as a 60% equity / 40% fixed income allocation, are usually delivered through mutual funds and ETFs that must be actively and costfully rebalanced.

- arXiv
- 2608.02917
- Published
- 2026-08-03
- Authors
- Zachary Feinstein, Ionut Florescu, Sean O'Leary
AI summary
Overview
Research area: Trading and market microstructure, with a portfolio-management interpretation of automated market makers (AMMs); the paper sits at the intersection of DeFi mechanism design, asset management, and empirical fund performance evaluation.
Technical level: Advanced. Section 2 is a formal mathematical treatment (invariants, convex optimization, KKT conditions, three appendix proofs), while Section 3 is an empirical backtest that a non-specialist can follow.
Scope (one sentence): The paper recasts the geometric mean market maker (G3M) as a programmatic, verifiable target-weight portfolio product, proves that a new multi-asset fee structure bounds portfolio mis-weighting, and backtests simulated G3M portfolios against three real funds (VBIAX, EQL, EDOW) on returns and tracking error.
What This Paper Is About
Target-weight mandates, such as a 60% equity / 40% fixed income allocation, are usually delivered through mutual funds and ETFs that must be actively and costfully rebalanced. The authors argue that an AMM's mathematical invariant does two jobs at once: it sets swap terms (the usual "decentralized exchange" reading) and it determines the reserves held on behalf of liquidity providers (a portfolio rule). The goal is to show, theoretically and empirically, that a fee-bearing G3M can hold a stated allocation with compliance verifiable directly from the pool's observable holdings, and that it can compete with incumbent funds on both tracking error and compound annual growth rate.
Key Contributions
- A multi-asset, path-independent fee structure for the G3M. The authors define a fee parameter γ ∈ [0,1) and split every market operation into a fee-free proportional mint/burn component and a residual, fee-bearing non-proportional swap. Proportional liquidity provision is shown to be fee-free for every γ, the γ = 0 case recovers the standard fee-free G3M, and the fee generates a strictly positive bid-ask spread that cannot be reduced by routing through an intermediate asset.
- A closed-form no-arbitrage region in weight space for the multi-asset G3M. Theorem 5 proves that no profitable arbitrage exists if and only if every realized weight satisfies ŵᵢ ≥ (1 − γ)wᵢ, and Corollary 6 converts this into an explicit band, ŵᵢ ∈ [(1 − γ)wᵢ, γ + (1 − γ)wᵢ]. This extends the two-asset result of prior work to N assets under a different fee construction.
- The first direct comparison of an AMM-held portfolio against traditional mutual funds and ETFs. Three case studies (VBIAX, EQL, EDOW) are evaluated on annualized tracking error against two mandate definitions and on CAGR, with the G3M subjected only to arbitrage order flow.
- An reinterpretation of loss-versus-rebalancing (LVR). The transfer from liquidity providers to arbitrageurs is framed not purely as adverse selection but as the execution cost of maintaining the portfolio mandate, with the appropriate benchmark being alternative mechanisms for holding the same exposure.
Main Findings
- Bounded mis-weighting is guaranteed ex ante. Under competitive arbitrage, realized weights lie in the simplex (1 − γ)w + γ cl 𝒲, and total allocation drift is bounded by 2γ(1 − minᵢ wᵢ).
- The fee parameter behaves as a rebalancing band, not an exchange spread. The authors state that the relatively high dominating fee ranges in Table 1 are better interpreted as parameters of the portfolio rebalancing band than as competitive exchange spreads.
- Dominance depends on the benchmark definition and, for one fund, on TE frequency. Against VBIAX, the G3M never matches tracking error using daily TE (dominance region ∅), but Pareto-dominates for fees in [2.73%, 3.90%] when TE is computed monthly. The authors attribute the difference to negative autocorrelations in daily active returns, which overestimate TE for all funds.
- Against EQL, dominance holds against the economic mandate but not the legal index NAV. With TE measured against the underlying index (legal mandate), the G3M cannot match EQL's NAV tracking error (region ∅). With TE measured against daily rebalancing of the underlying sector indices (economic mandate), the G3M dominates even the NAV position for fees in [3.22%, 7.09%]. Against EQL's market-value series on the legal mandate, the region is [3.56%, 10%], where 10% is the upper end of the sampled fee range.
- Against EDOW, results follow the same pattern. The G3M cannot match EDOW's NAV tracking error on the legal index benchmark (region ∅), but dominates the NAV position on the economic mandate for fees in [3.32%, 9.90%]. Against EDOW's market value on the legal mandate, the region is [3.44%, 10%].
- Tracking error does not increase monotonically with the fee. The γ = 0 G3M holds target weights perfectly but transfers the full value of each rebalancing opportunity to arbitrageurs; positive fees allow temporary mis-weighting while letting LPs retain part of the rebalancing proceeds, producing non-monotonic frontiners.
- Mutual-fund-based G3M positions show lower TE than ETF-based ones (VBIAX case). The authors conjecture this reflects market noise in ETF prices that is not visible in a mutual fund's daily NAV.
- Market-making profit is negative by construction in these simulations. Because only arbitrage flow is used, fees minus LVR are non-positive, yet the total G3M position can still outperform incumbent funds. The authors note, subject to future study, that realized LVR net of fee revenue annualizes to a cost of the same order of magnitude as the incumbent fund's expense ratio; the actual expense ratios and LVR magnitudes are not reported.
- Path independence holds only under a specific condition. Lemma 4 shows the fee-bearing update is superadditive across a split trade, with equality (for γ in (0,1)) if and only if all sub-operations share a common minimizing asset.
Methodology in Plain English
The paper has two halves.
The theory half starts from the standard G3M invariant, the weighted geometric mean of reserves, which in a fee-free pool makes the pool's realized portfolio weights exactly equal the chosen weights w regardless of prices. The authors then add fees in a way that does not penalize ordinary deposits and withdrawals: any trade is decomposed into the largest proportional scaling common to all assets (treated as a free mint or burn of LP shares) plus a residual that changes the portfolio mix (treated as a swap, and charged). They then solve the arbitrageur's problem of extracting maximum value from the pool subject to that fee rule, using a convex program and its KKT conditions, to derive exactly when arbitrage stops being profitable. That stopping condition is the "no-arbitrage region" that lets an outside observer check, from the pool's reserves and market prices, whether the mandate is being respected.
The empirical half simulates G3M pools that hold the same assets as three real funds and compares them over fee values from 0% to 10%. Three design choices matter: the pools receive only arbitrage order flow (no uninformed or noise trading), which produces a conservative bound on fee revenue; arbitrageurs trade at most once per day at the market close price; and dividends are reinvested into the asset that paid them so total-return series can be used. Each simulated portfolio is scored on compound annual growth rate and annualized tracking error against two targets: the fund's legal index, and a daily-rebalanced version of the fund's stated economic mandate. A portfolio only "dominates" the fund if it wins on both metrics at once. Data come from Bloomberg.
Case specifics: VBIAX (over $50 billion in assets under management) tracks 60% equity (Morningstar US Total Market Index, CRSPTMT) and 40% fixed income (Bloomberg US Aggregate Float Adjusted Index, LBUFTRUU), tested from January 2, 2014 through June 30, 2026, with G3M positions built from either VTSAX/VBTLX mutual funds or VTI/BND ETFs. EQL targets equal weights in all 11 GICS sectors (total return index NYXLEWTR), tested June 19, 2018 through May 29, 2026, with a G3M holding all 11 Sector SPDR ETFs. EDOW targets equal weights in the 30 DJIA components (DJIEWTR), tested November 11, 2024 through June 29, 2026, a window chosen so the index components were held steady, with a G3M holding all 30 components.
Why This Matters
Impact on research. The paper reframes the AMM not as an exchange with inventory but as a portfolio technology, which changes what "success" means for a liquidity pool: the relevant benchmark becomes an alternative vehicle for the same exposure rather than a hypothetical dealer with no inventory objective. It also supplies the first quantified tracking error for an AMM-generated position and gives a closed-form, multi-asset characterization of the fee-induced no-arbitrage region, which earlier work had only in a two-asset setting.
Real-world applications:
- Tokenized target-weight funds. A mandate and its permissible rebalancing band can be encoded directly in the pool invariant, with compliance checkable from observable reserves and contemporaneous prices rather than from manager attestations.
- Cheaper rebalancing for retail allocation products. The fee parameter doubles as a rebalancing-band control, so a product designer can tune how tightly weights are held versus how much rebalancing value is retained by holders.
- Regulatory and investor monitoring. Corollary 6 gives an auditable band; an observer can flag a pool as out of compliance without access to internal records.
- Benchmark design for AMM evaluation. Comparing against funds with real assets under management (VBIAX, EQL, EDOW) rather than stylized alternatives sets a higher and more economically meaningful bar for AMM performance claims.
Industry relevance. The authors note that incumbent funds hold real assets under management, revealing genuine economic demand for these mandates, so an AMM that delivers the same mandate is competing in an existing market rather than creating a new one. The observation that net LVR annualizes to a cost of similar order to incumbent expense ratios suggests the comparison is economically live, though the paper does not report the expense ratios or the LVR magnitudes themselves.
Future Directions
- Decompose the cost comparison explicitly. The claim that net LVR is of the same order of magnitude as incumbent expense ratios is flagged as subject to future study; quantifying both sides directly is the obvious next step.
- Add realistic order flow beyond arbitrage. All simulations use arbitrage-only flow, which makes market-making profit non-positive by construction; measuring the contribution of uninformed or noise flow is needed to assess LP economics.
- Model the legal-versus-economic mandate gap. The G3M could not match the NAV tracking error of the legal index benchmarks for EQL and EDOW, and the paper treats a continuously maintained target-weight portfolio and a legally specified index as distinct interpretations of one mandate; reconciling them is an open design question.
- Test the mechanism against other mandate types and other fee/band calibrations. The empirical range stops at the 10% upper end of sampled fees, and several dominance regions are open-ended at that boundary, so the behavior of the frontier above 10% is not reported.
Target Audience
Researchers in market microstructure and DeFi mechanism design; quantitative portfolio managers and product designers interested in on-chain or tokenized target-weight vehicles; regulators and compliance professionals who need auditable mandate-adherence rules; and technically trained finance readers who want a rigorous but empirically grounded account of whether AMMs can function as portfolio infrastructure rather than only as exchanges. The theory sections require comfort with constrained optimization and invariance arguments, but the empirical sections and the dominance table are accessible to practitioners.
Authors’ abstract
Automated market makers (AMMs) are typically interpreted and evaluated as decentralized exchanges. Herein, we take the perspective envisioned by Balancer that an AMM can also be viewed as a portfolio technology that programmatically enforces an economic mandate. In particular, we follow the geometric mean market maker (G3M) invariant employed by that protocol in order to enforce a target-weighted portfolio. We introduce a multi-asset fee structure to the G3M under which…
The full abstract on arXiv