Lesson 9.3: Quant and discretionary funds, pods and teams
Learn how quant and discretionary funds differ, how multi-manager pods work, why drawdown limits stop skilled managers too, and the roles on a team.
By the end of this lesson you will be able to compare a quantitative fund with a discretionary one on process, scale, risk and failure modes, describe how a multi-manager platform divides capital among independent pods and controls them with risk limits, work out how much diversification across managers raises the Sharpe ratio and how fast that benefit disappears when correlations rise, explain why a drawdown rule removes many skilled managers along with the unskilled, and name the roles on a fund's team. You will show that twenty pods that each have a Sharpe ratio of 1.0 and a correlation of 0.1 combine to a Sharpe ratio of 2.63 but only 1.38 if the correlation rises to 0.5, that a 5 per cent drawdown limit on a pod running at 6 per cent volatility stops 57 per cent of managers with a true Sharpe ratio of 1.0 within three years, and 92 per cent of managers with no skill, and that if one manager in five is skilled, only 13 per cent of those stopped were skilled. You should have read Lessons 3.4, 4.2, 5.3 and 9.1. The mathematics is Sharpe ratios, correlation and simulation. This course is education about how markets and research work, not investment advice.
Lesson 9.1 described how a hedge fund is built and paid. Lesson 9.2 mapped what it trades. This lesson asks how the people inside are organised to find, size and control the bets. Two questions run through it. How do you make decisions: with a model, with human judgement or with both? And how do you combine many decision-makers into one firm that is more reliable than any one of them? The answers explain why the industry has moved towards large multi-manager platforms, and what that does to the people who work in them.
Quantitative and discretionary funds
A discretionary fund's decisions are made by people. A portfolio manager, backed by analysts, studies companies, economies or markets, forms a view and decides what to buy and sell and how much. A quantitative fund's decisions come from models: researchers build and test signals, as in Modules 3 to 5, a system combines them into positions and trades them automatically, and the people supervise the system and keep improving it. In between sits a spectrum: systematic discretionary or "quantamental" funds use models to screen ideas and discretion to decide.
· Discretionary · QuantitativeWhere the edge comes from · Deep knowledge of companies, industries or policy, and judgement about what is priced in · Statistical patterns found across many assets and tested on data, implemented consistently
Decision process · Conviction on a few positions, written theses · Many small positions from signals, sized by rules
Breadth · Low: perhaps 20 to 100 positions · High: hundreds to thousands (Lesson 3.3's fundamental law)
Turnover and horizon · Usually lower, weeks to years · From microseconds to months, depending on the style
Capacity · Limited by the liquidity of concentrated positions and the manager's attention · Limited by costs and impact (Lesson 4.4), often larger for slow strategies
Consistency · Varies with the manager's judgement, mood and incentives · Rules are followed, but the rules can decay
Main risks · Key-person risk, behavioural biases, concentrated losses · Overfitting (Lesson 3.4), crowding, regime change, bugs
Typical failure · A wrong thesis held with conviction · A model stops working, or everyone's models fail together
Scaling the team · Hiring more analysts adds coverage slowly · Hiring more researchers adds signals, with discipline needed to avoid false discoveries
Neither is better. Discretionary investing is stronger where information is qualitative, events are unique, or data is thin: a bankruptcy, a merger, a political shift. Quantitative investing is stronger where there are many repeated situations, a lot of data and a need for consistency: thousands of stocks, daily. Many successful firms use both.
How a quantitative team is organised
A systematic firm is organised like a factory with several stages, matching the pipeline of Lesson 3.2.
Research. Researchers find and test signals, with the care about look-ahead, costs and multiple testing of Modules 3 and 4. A good firm keeps an experiment log of every idea tried, and a second group, independent of the researchers, validates anything that goes live.
Data and infrastructure. Data engineers build and clean the data (Lesson 3.1), and software engineers build the research platform, the backtester, the production system and the monitoring. In a modern firm they outnumber the researchers.
Portfolio construction and risk. Specialists combine signals into a portfolio, with the risk and cost models of Lessons 4.4 and 5.3, and set limits on exposures.
Execution. Traders and algorithms get the orders done with minimal cost (Lesson 5.4). The execution team measures its shortfall and feeds it back to research.
Risk, operations and compliance. Independent teams monitor exposures, reconcile positions and check that rules are followed.
The research process has a hazard of its own. If hundreds of researchers test thousands of ideas, some will pass by luck, so a firm needs governance: a high bar of evidence, an out-of-sample test that is kept clean, a record of what has been tried, and a deliberate "no" for ideas that look too good. The multiple-testing arithmetic of Lesson 3.4 applies to an organisation as much as to one researcher.
How a discretionary team is organised
A discretionary fund is built around one or a few portfolio managers (PMs). Each has a team of analysts, usually organised by sector or region, who research companies and propose ideas, and traders who execute. A chief risk officer sets and monitors limits, and an operations and compliance team handles the rest. The key feature is concentration of judgement: the fund's performance depends heavily on a few people, which is its strength and its weakness. If the star PM leaves, the track record may leave with them.
The multi-manager platform: the pod model
A fast-growing form of hedge fund, often called a multi-manager or pod shop, treats many independent teams as one portfolio. The idea is simple. Instead of betting on the skill of one manager, the firm hires many portfolio managers, each running a pod, a small team with its own strategy, and allocates capital to each. The firm sets risk limits for every pod, and a central risk team watches the whole book and the exposures that the pods share.
The main features:
Allocation. Each PM gets a capital allocation and a volatility target, which can be raised for good performance and cut for poor performance.
Central risk and neutrality. The platform requires pods to be neutral to the market and to factors, such as the beta and sector exposures of Lesson 5.3, so that returns come from stock selection or specific strategies and not from common risks. The central team may hedge the factor exposures the pods leave over.
Strict loss limits. If a pod loses a set share of its capital, its allocation is cut, and if it loses more, it is shut down and the PM is let go. Limits like 5 per cent for a cut and 7 to 8 per cent for termination are used in the industry, though terms vary and are not public.
Pass-through costs and pay. The platform charges investors the management and performance fees of Lesson 9.1, and typically passes the costs of data, technology and staff to the pods. The PM is paid a share of the profits the pod earns, often in the range of 15 to 20 per cent, with much of the risk of a lost job.
Leverage. Because the diversified book has a high Sharpe ratio and a low volatility, the platform can run it at a high leverage to reach a target volatility, which magnifies both the profits and the risk of a shock.
A talent market. Pods compete for capital, and PMs move between platforms, so the firm competes by offering capital, infrastructure and a risk framework.
Why the model works: diversification across managers
The logic is the fundamental law of Lesson 3.3 and the diversification of Lesson 5.1 applied to people. If each pod has a Sharpe ratio S and the correlation of returns between pods is ρ, then the Sharpe ratio of a portfolio of N pods with equal risk is
Sportfolio = S × √(N ÷ (1 + (N − 1) ρ))
Worked example 1. Suppose each pod has a Sharpe ratio of 1.0, which is high but not impossible after costs for a skilled manager, and the average correlation between pods is 0.1, achieved by requiring neutrality and different strategies.
Number of pods · 1 · 5 · 10 · 20 · 50Portfolio Sharpe ratio · 1.00 · 1.89 · 2.29 · 2.63 · 2.91
Twenty pods give a Sharpe ratio of 2.63 from pods that are each only 1.0, and the ceiling as N grows is 1 ÷ √0.1 = 3.16. In rupees, take 20 pods with ₹500 crore each. Each earns on average a Sharpe ratio of 1.0 times its volatility of 6 per cent, 6 per cent a year, or ₹30 crore, so the book earns ₹600 crore. Each pod has a standard deviation of ₹30 crore. With a correlation of 0.1 the variance of the total is 20 × 900 + 20 × 19 × 0.1 × 900 = 52,200, a standard deviation of ₹228 crore, only 2.3 per cent of the ₹10,000 crore of capital. That low volatility is what lets the platform use leverage.
The assumption doing the work is the low correlation. In a stress, correlations rise: pods that were independent hold the same crowded positions, or must sell at the same time (Lesson 9.4). The same twenty pods with a correlation of 0.3 give 20 ÷ (1 + 19 × 0.3) = 3.0 effective bets, so a Sharpe ratio of 1.73, and at a correlation of 0.5 only 1.9 effective bets and a Sharpe ratio of 1.38. A platform sized for 2.6 and running at high leverage faces a very different risk when the real figure is 1.4.
Experiment with the number of pods and the correlation, and see how fast the benefit drains away.
Drawdown limits and the noise in skill
The strict loss limits that platforms use have an effect that is rarely discussed: they cannot separate skill from luck, as Lesson 4.2 would predict. A pod's return over a few months is mostly noise, even for a good manager.
We simulated it. A pod runs at a volatility of 6 per cent a year, with monthly returns that are normally distributed. The platform fires the manager if the pod's cumulative return falls 5 per cent below its previous peak. We ran 40,000 simulated managers each for three kinds of skill, and counted how many were stopped out within 6, 12, 24 and 36 months.
True Sharpe ratio of the manager · Stopped within 6 months · 12 months · 24 months · 36 months · Still running after 36 months0.0 (no skill) · 22.5% · 50.3% · 79.7% · 91.7% · 8.3%
1.0 (good) · 8.2% · 21.0% · 41.7% · 57.1% · 42.9%
2.0 (exceptional) · 2.4% · 6.2% · 13.6% · 20.2% · 79.8%
The limit does weed out the unskilled: 92 per cent of managers with no skill are gone in three years. But it also removes 57 per cent of managers with a Sharpe ratio of 1.0 and 20 per cent of those with an exceptional 2.0. A drawdown of 5 per cent is less than one year's standard deviation of the pod, so even a good manager runs into it by bad luck, and the stop is tripped by variance more than by incompetence.
What does that mean for the people who survive and for those who are fired? Suppose one manager in five is truly skilled with a Sharpe ratio of 1.0 and the rest have none. Of all those stopped within three years, only 13.5 per cent were skilled, but of those who survived, 56 per cent were skilled. A manager who is fired has most probably been unlucky or ordinary, but 13.5 per cent of the fired are good managers lost to the system, and 44 per cent of the survivors have no skill and will be stopped later. Selection by a loss limit is a noisy filter. If the skilled managers have a Sharpe ratio of 2.0, the filter works better: only 5 per cent of those stopped were skilled, and 71 per cent of survivors are.
None of this makes loss limits wrong. For a platform with many pods, a quick stop limits the damage from a manager who is genuinely losing control, and the cost of firing a good manager by mistake is small when talent is plentiful. But it shapes behaviour: PMs near their limit may cut risk, avoid large positions, or run very low-volatility strategies, which has implications for the capacity and style of what a platform's pods can do. And it shows why a short track record is weak evidence, whichever way you look at it.
The economics for a PM and for the platform
A pod's economics, with illustrative numbers: a PM with ₹500 crore, a volatility of 6 per cent and a true Sharpe ratio of 1.0 expects to earn ₹30 crore a year. At a 15 per cent payout, the expected pay is ₹4.5 crore, with the chance of a stop-out noted above, and no pay in a losing year. The platform keeps the rest, charges investors its own fees and bears the diversified risk. For a PM with no skill, the expected P&L is zero and the expected pay is zero or slightly positive in lucky years. The structure is asymmetric, like the performance-fee option of Lesson 9.1: the PM is paid for gains and fired after losses, which encourages the PM to protect the downside.
Careers and roles
For readers of this course, the useful picture is who does what.
Role · What the person does · SkillsPortfolio manager · Owns the strategy and the P&L, decides risk and sizing · Judgement, market knowledge, risk discipline
Quantitative researcher · Finds and tests signals and models · Statistics, programming, scepticism about backtests (Module 3 and 4)
Quantitative developer or engineer · Builds the research platform, the execution system, the data pipeline · Software engineering, performance, reliability
Data engineer and data scientist · Sources, cleans and structures data, builds features · Data handling, domain knowledge
Trader and execution specialist · Gets orders done at low cost and measures it (Lessons 4.4, 5.4) · Market microstructure, algorithms
Analyst · Researches companies or sectors for a discretionary PM · Accounting, industry knowledge, writing
Risk manager · Measures and limits exposures, runs stress tests (Module 11) · Statistics, independence, communication
Quant strat or structurer (banks) · Builds pricing and risk models for a bank's trading desk (Module 12) · Stochastic calculus, numerical methods (Modules 7 and 8)
Operations, compliance, legal, investor relations · Keep the business running, lawful and funded · Process, regulation, relationships
In India, the employers include asset managers and AIFs, proprietary trading and market-making firms, the trading and risk desks of banks and brokers and the offices of global funds. The mix of roles varies, and so do the terms: non-compete clauses and garden leave are common in the industry, so read a contract with care.
Putting it together: how to read a fund's organisation
Is the edge in people or in a process? A star-manager fund has key-person risk. A research factory has process risk and the need for governance.
How many independent bets does it make? More independent bets mean a higher Sharpe ratio for the same skill (Lesson 3.3), and the number of true independent bets, not positions, is what counts.
How does it control loss and leverage? Understand limits, and who has the authority to enforce them.
How does it measure and reward skill? A system that rewards short-term P&L without a long record will reward luck.
What happens to correlation in stress? The diversification that justifies leverage may vanish when it is needed.
The code below computes the diversification from many pods, simulates the drawdown rule and the Bayesian arithmetic, and works out the pod economics.
python
import numpy as np
from math import sqrt
1. diversification across pods
for n in (1, 5, 10, 20, 50):
print(f"{n:>2} pods, pod Sharpe 1.0, correlation 0.1: portfolio Sharpe {1.0 sqrt(n / (1 + (n - 1) 0.1)):.2f}")
for rho in (0.1, 0.3, 0.5):
neff = 20 / (1 + 19 rho)
print(f"20 pods, correlation {rho}: effective bets {neff:.2f} | portfolio Sharpe {sqrt(neff):.2f}")
the book in rupees: 20 pods of Rs 500 crore, each 6% volatility and a Sharpe ratio of 1.0
cap, vol, n, rho = 500.0, 0.06, 20, 0.1
podsd = vol cap
totalvar = n podsd 2 + n (n - 1) rho podsd 2
print("expected profit", n vol cap, "crore | standard deviation", round(sqrt(totalvar), 1), "crore =", round(sqrt(totalvar) / (n cap) 100, 2), "% of capital")
2. a 5% drawdown limit on a pod with 6% volatility
rng = np.random.defaultrng(12)
M, months, vol, stop = 40000, 36, 0.06, -0.05
monthlysd = vol / sqrt(12)
firststop = {}
for sharpe in (0.0, 1.0, 2.0):
r = rng.normal(sharpe vol / 12, monthlysd, (M, months))
cum = np.cumsum(r, axis=1)
peak = np.maximum.accumulate(np.maximum(cum, 0), axis=1)
hit = (cum - peak) <= stop
first = np.where(hit.any(axis=1), hit.argmax(axis=1) + 1, 999)
firststop[sharpe] = first
print(f"true Sharpe {sharpe}: stopped within 6, 12, 24, 36 months:", [round(float((first <= m).mean()) 100, 1) for m in (6, 12, 24, 36)],
"| still running after 36 months", round(float((first > 36).mean()) 100, 1), "%")
3. who is stopped, if one manager in five is skilled?
t0 = (firststop[0.0] <= 36).mean()
for skill in (1.0, 2.0):
tk = (firststop[skill] <= 36).mean()
prior = 0.2
print(f"skilled managers have Sharpe {skill}: share of the stopped who were skilled {prior tk / (prior tk + (1 - prior) t0):.3f}"
f" | share of the survivors who are skilled {prior (1 - tk) / (prior (1 - tk) + (1 - prior) (1 - t0)):.3f}")
4. pod economics
sharpe, pay = 1.0, 0.15
profit = sharpe 0.06 500
print("expected pod profit", profit, "crore | expected manager payout", round(pay profit, 2), "crore | years to see the skill at t = 2:", 4 / sharpe 2)
Common mistakes
Treating quantitative and discretionary funds as rivals. They suit different problems, and good firms use both.
Believing a model removes the need for judgement. Choosing what to model, when it has stopped working and how much to risk are all judgements.
Assuming independent pods stay independent. Correlations rise in stress, and the diversification is lowest when it is needed.
Sizing leverage on the calm-period Sharpe ratio. The platform's volatility is low only while the correlation is.
Reading a stop-out as proof of no skill. A 5 per cent limit stopped 57 per cent of managers with a Sharpe ratio of 1.0.
Reading survival as proof of skill. Of the survivors, 44 per cent in our example had no skill.
Ignoring key-person risk. The strategy may walk out of the door.
Running a research factory without governance. The number of ideas tested multiplies the false discoveries.
Paying for P&L without a record. It rewards luck and encourages hidden tail risk.
Key takeaways
Discretionary funds decide by judgement and conviction on fewer positions. Quantitative funds decide by models across many positions, with a research, data, portfolio, execution and risk organisation. Each has its own failure mode, and key-person risk, overfitting and crowding are the main ones.
A multi-manager platform allocates capital to pods, each with its own strategy, neutral to common risks, under strict loss limits and central risk control, with managers paid a share of profits.
Diversification across pods gives S × √(N ÷ (1 + (N − 1)ρ)). Twenty pods with a Sharpe ratio of 1.0 and a correlation of 0.1 give 2.63, but 1.73 at a correlation of 0.3 and 1.38 at 0.5. The high leverage a platform runs depends on the correlation staying low.
A 5 per cent drawdown limit on a pod at 6 per cent volatility stopped 92 per cent of managers with no skill, 57 per cent of those with a Sharpe ratio of 1.0 and 20 per cent of those with 2.0 within three years. It cannot separate skill from luck.
If one manager in five is skilled (Sharpe 1.0), only 13.5 per cent of the stopped were skilled, and 44 per cent of the survivors were not.
A fund's team spans portfolio managers, researchers, developers, data engineers, traders, risk, operations and compliance. In India, employers include asset managers and AIFs, proprietary and market-making firms, bank desks and global funds' offices.
Check yourself
1. Sixteen pods each have a Sharpe ratio of 1.2 and an average correlation of 0.05. What is the portfolio Sharpe ratio?
Neff = 16 ÷ (1 + 15 × 0.05) = 16 ÷ 1.75 = 9.14. Sharpe ratio = 1.2 × √9.14 = 3.63.
2. In stress the correlation rises to 0.4. What is the new Sharpe ratio?
Neff = 16 ÷ (1 + 15 × 0.4) = 16 ÷ 7 = 2.29. Sharpe ratio = 1.2 × √2.29 = 1.81.
3. Why does a drawdown limit that is smaller than the pod's annual volatility stop many skilled managers?
Over a few months a pod's return is mostly noise, and a fall of a size that is a fraction of a year's standard deviation is common even for a manager with a positive true Sharpe ratio.
4. A researcher at a quant fund tests 500 variants of an idea and brings the best, which has a Sharpe ratio of 1.5 over four years, to the portfolio manager. What should the firm's governance ask?
How many variants were tried (500), what the Sharpe ratio would be from luck alone (about 3.0 × 0.5 = 1.5 for the best of 500, from Lesson 3.4), and whether it survives an untouched out-of-sample test. On these facts it is not evidence of skill.
5. What is the key-person risk in a discretionary fund, and how can a firm reduce it?
The returns depend on one or a few people, so their departure or error can end the strategy. The firm can document the process, build a team with several decision-makers, diversify across managers and align incentives with long vesting and co-investment.
Try this
Set aside forty-five minutes. You need Python and the code in this lesson.
1. Run the code. Check the diversification table, the stop-out table and the Bayesian shares.
2. Change the stop from −5 per cent to −8 per cent. How does the share of skilled managers stopped change, and the share of survivors who are skilled?
3. Increase the pod volatility to 10 per cent keeping the 5 per cent stop. What happens, and why does that change the behaviour of a manager near the limit?
4. Use the diversification simulator to find the number of pods and the correlation at which the portfolio Sharpe ratio is 2.0.
5. Compute the effect of a correlation shock: take 30 pods with a correlation of 0.1 and raise it to 0.35. By what factor does the book's volatility rise?
6. Read the careers pages of a few quant and multi-manager firms and list the roles they hire for, with the skills asked.
7. Write a short note: whether you would prefer to work in a discretionary, quantitative or multi-manager setting, and what evidence you would ask for about the firm's process.
A good result is a diversification calculation you can do by hand, a stop-out table you can explain and an informed view of how a team is organised. The last lesson of this module studies what happens when these organisations fail: the leverage, crowding and operational lessons of LTCM, the quant quake of 2007 and others.
Read more
Andrew Lo's Adaptive Markets and his writings on hedge funds discuss performance evaluation and the role of luck. Cliff Asness's essays on quantitative investing and Ernest Chan's Quantitative Trading describe how quant firms are organised. Scott Patterson's The Quants tells the story of the rise of quantitative funds, and Zuckerman's The Man Who Solved the Market describes one systematic firm's research culture. Marcos López de Prado's Advances in Financial Machine Learning has chapters on the organisation of research and the problem of false discoveries, and Campbell Harvey and Yan Liu's papers on multiple testing in finance explain why governance matters. For the pod model, industry coverage in the financial press is the best public source, and the terms of any one platform are private.
Tags: Quant Trader Multi-manager funds Pod model Quant funds Discretionary funds