3The Multi-Manager Platform

A large multi-manager platform’s regulatory brochure lists, among the expenses its funds bear, the salaries, bonuses and performance-based compensation of the portfolio managers who run its teams, their recruiters’ fees, their computers and their offices. It calls the arrangement an “expense pass-through”. It adds that a portfolio manager is paid a share of the profit of his own team, “without taking into account the performance of other Portfolio Managers or of the Millennium Partners Funds generally”, and that portfolio managers with positive performance “will receive performance-based compensation even if the Millennium Partners Funds’ overall returns are negative”. The investor pays each winning team’s bonus, including in the years the fund loses. This chapter prices that promise.

3.1 Pods, the centre and the platform’s promise

A multi-manager platform (One Quant Book 1, chapter 1) is a hedge fund run as many independent teams, each a pod (One Quant Book 8, chapter 28), under a centre that allocates capital, sets risk limits, nets and hedges the sum, and pays for the infrastructure. Its promise to investors is diversification of talent: fifty teams with modest Sharpe ratios and low correlation add up to a fund with a high one. Its promise to portfolio managers is a formulaic share of what their own team makes, paid whatever the others do.

The table is the platform’s business case and its limit. Correlation, not the number of teams, sets the ceiling: at $\rho=0.1$ it is $1/\sqrt{0.1}=3.16$ times a pod’s Sharpe ratio, and fifty pods already reach 2.91 of it. Past a few dozen teams, adding pods buys little unless they are less correlated with the ones already there; lowering correlation, by hiring different strategies or hedging what they share (Definition 3.8), buys more than hiring.

3.2 Pass-through fees and the investor’s bill

The chapter’s platform (illustrative) runs 50 pods, each with a Sharpe ratio of 1.0 and a volatility of 20% a year on its share of the fund’s capital, correlated at 0.1 through one shared factor. Teams are paid 20% of their own profit, losses carried forward; a team that loses more than 10% of its capital in a year is replaced by a new one. Pods’ costs of 3% a year and the platform’s 1% pass through; the manager takes a 20% performance allocation on what is left. Over twenty simulated years and twenty seeds, a hundred dollars of capital earns this a year (Figure 3.2): $20.09 gross, of which teams receive $4.27, costs take $4.00 and the manager $2.38, leaving the investor $9.44. The fund lost money in 9 of its 400 years.

A classic fund charging 2% of capital and 20% of profit on the same pods would leave its investors $14.47, charging them $5.62. It could not exist: out of $5.62 of fees its manager would have to pay the teams $4.27 and the costs $4.00, losing $2.65 on every hundred dollars. The pass-through is the fee structure a platform needs once talent and infrastructure cost more than a fixed fee can pay; the investor’s question is whether what is left after them, $9.44 here, is worth the fund’s risk.

def payouts(ann: np.ndarray, rate: float, carry: bool = True, fire: float | None = None) -> np.ndarray:
    """Each pod: rate * (profit above the losses it carries forward); carried losses are reduced by profits;
    nothing is clawed back and other pods' results do not count. carry=False pays on each year alone. With
    `fire`, a pod that loses more than `fire` in a year is replaced by a new team, whose carried loss starts at 0."""
    years, n = ann.shape
    c = np.zeros(n)
    out = np.zeros((years, n))
    for t in range(years):
        net = ann[t] - c if carry else ann[t]
        out[t] = rate * np.maximum(net, 0.0)
        c = np.maximum(-net, 0.0) if carry else c
        if fire is not None:
            c[ann[t] < -fire] = 0.0
    return out


def netting_cost(ann: np.ndarray, rate: float, carry: bool = True, fire: float | None = None) -> np.ndarray:
    """What the pods' payouts cost each year beyond rate * max(total profit, 0)."""
    return payouts(ann, rate, carry, fire).sum(1) - rate * np.maximum(ann.sum(1), 0.0)

3.3 Paying winners: the netting problem

At $p=20\%$ the netting cost is 1.7% of gross profit for pods of Sharpe ratio 1, 7.9% at 0.5 and 22.9% at 0.25 (Figure 3.3). Carrying losses forward removes most of it for a team that stays: in the chapter’s platform with no team ever replaced, netting costs 0.1% of gross profit. Replacing losing teams restores it, because a replaced team’s carried loss dies with it and its successor starts at zero: with teams replaced after a 10% loss, the simulated cost is 1.3% of gross profit at a Sharpe ratio of 1 and 6.2% at 0.5, close to the formula’s bound. A platform’s netting risk is therefore set by the quality of its teams and the rate at which it turns them over, not by the payout rate alone.

3.4 Capital allocation and drawdown rules

The centre allocates capital to teams and takes it back. One Quant Book 8, chapter 28 simulated the central trade-off: a drawdown limit (defined there) stops good teams in bad luck and lets dead ones run, and its rate of false stops is set by the limit in units of the team’s own volatility. Platforms describe their rules in general terms; the specific thresholds of named firms are not in their public documents, and this book does not repeat second-hand figures. The model therefore uses an illustrative two-step ladder: a team’s capital is halved after a 10% drawdown within a year and withdrawn after 15%, and restored at the next year’s start.

In the chapter’s platform the ladder cuts 88.4% of team-years and stops 26.6%: with a volatility of 20%, a 10% drawdown within a year is ordinary. The fund’s return falls from 20.1% to 13.4% and its volatility from 6.9% to 4.8%; its Sharpe ratio falls from 2.92 to 2.77. Every pod in this model has a live edge, so the ladder can only cost; chapter 8 measures what it buys when some edges die.

3.5 Centre books and overlays

Fifty pods correlated at 0.1 give a fund whose volatility is dominated by what they share: $\sigma\sqrt{(1+49\rho)/50}$ is 6.9% for pods of 20% volatility, against 2.7% if the shared part is hedged. In the model the overlay leaves the fund’s mean return at 20.1% and lifts its Sharpe ratio from 2.9 to 7.5 (Figure 3.4). That is an upper bound: here all the shared risk is one factor the centre can trade. Real pods share crowded positions and liquidity as well as market and style factors (One Quant Book 7, chapter 28), and only the tradable part can be hedged; the brochure of the hook describes both uses of the centre’s own trades, hedging netted exposures and adding to them.

3.6 Tutorial: the investor’s bill

What to change next. Keep every team forever (exercise 7); correlate the pods at 0.4 and see the overlay’s gain shrink once only part of the shared risk is a tradable factor.

3.7 Build: the platform model

3.8 Exercises

3.9 Problem: Paid for the Winners

3.10 Interview questions