Home

Working notes: markets, probability, decision systems, and AI applied to finance.

The technical résumé is at pramod-thaz.github.io.

Latest Posts

• The Bayesian Investor: A Plain-Language Primer for the Chief Investment Officer — How your team can learn to think, model, and decide under uncertainty — one analogy at a time.
• The One Idea Behind Everything — Part of the The Bayesian Investor — A Primer for the CIO series.
• Phase 1 — Foundations: Probability, Updating, and the Tools of the Trade — Part of the The Bayesian Investor — A Primer for the CIO series.
• Phase 2 — Core Modeling: From Coin Flips to Real Decisions — Part of the The Bayesian Investor — A Primer for the CIO series.
• Phase 3 — From the Committee to the GPS: Tracking Things That Move — Part of the The Bayesian Investor — A Primer for the CIO series.
• Phase 4 — Putting It All Together: Multi-Sensor Market Prediction — Part of the The Bayesian Investor — A Primer for the CIO series.
• Use Case 1 — “Who Really Moves With Whom”: Portfolio Construction — Part of the The Bayesian Investor — A Primer for the CIO series.
• Use Case 2 — “Which Factors Actually Matter?”: IPO Pricing and Feature Selection — Part of the The Bayesian Investor — A Primer for the CIO series.
• Use Case 3 — “When One Domino Falls”: Cross-Asset Stress Testing With Private-Asset Look-Through — Part of the The Bayesian Investor — A Primer for the CIO series.
• Use Case 4 — Asset Allocation: The Operating System — Part of the The Bayesian Investor — A Primer for the CIO series.
• Use Case 5 — Performance & Risk Measurement and Attribution: The Scorekeeper — Part of the The Bayesian Investor — A Primer for the CIO series.
• Factor Rotation — Regime-Aware Systematic Portfolio — A systematic factor rotation strategy that combines two independent regime signals — one, that is slow-moving & lagged and the other more latent derived micro-structure signal that
• Dashboard Samples & Artifacts — Selected screenshots from work across performance measurement, agent-based simulation, and multi-family operations.

Posts

Working notes, essays, and work samples.

Categories

• Bayesian Workflow — regime-aware systematic investing and probabilistic decision systems
• General — essays, notes, and work samples
• Real Assets — commercial real estate and private markets

Bayesian Workflow

Regime-aware systematic investing and probabilistic decision systems.

The Bayesian Investor — A Primer for the CIO

A plain-language series for CIOs and executives: how a team learns to think, model, and decide under uncertainty — one analogy at a time. Start at the series index.

1. The One Idea Behind Everything
2. Phase 1 — Foundations
3. Phase 2 — Core Modeling
4. Phase 3 — Dynamic Tracking
5. Phase 4 — Applied Integration
6. Use Case 1 — Portfolio Construction
7. Use Case 2 — IPO Pricing
8. Use Case 3 — Cross-Asset Stress Testing
9. Use Case 4 — Asset Allocation
10. Use Case 5 — Attribution

The Bayesian Investor: A Plain-Language Primer for the Chief Investment Officer

How your team can learn to think, model, and decide under uncertainty — one analogy at a time.

This primer explains the four-phase learning plan your team has charted, entirely in plain language. No formulas. No Greek letters. No code. If a technical word appears, it will be explained with an everyday analogy in the same breath. The goal is not to make you a mathematician — it is to give you a mental map of what your team is building, why it matters for public markets and commercial real estate, and how the pieces fit together into one decision-making system.

Start with a belief. Update it as evidence arrives. Always know how sure you are.

Executive Summary: The Roadmap in One Page

The one sentence to take away: Investing under uncertainty is the bookmaker’s craft — start with a belief, update it relentlessly as evidence arrives, and never lose sight of how sure you are — and this roadmap gives your team the tools to practice that craft at scale.

A suggested path forward

• Start small: run the coin-flip exercises yourself or with a small team to internalize the three rules of updating. This is the foundation; skip it and the rest is scaffolding.
• Build the alpha/beta discipline first: Phase 2’s asset-pricing model is the highest-value, lowest-risk deliverable — it directly informs how you judge your own performance.
• Prototype one filter: pick a single hidden target (volatility, or a corridor’s true activity level) and track it with a Kalman or particle filter before wiring anything into production.
• Adopt Black-Litterman as the channel for analyst views — it’s the fastest way to make your existing investment process more disciplined without disrupting it.
• Go multi-sensor only after the single-sensor prototype works. Fusion is powerful but demands trust in each sensor first.

The roadmap is long, but it is one idea, repeated with increasing power. Your team already knows the markets. This plan gives them the bookmaker’s discipline — and the machinery to practice it at the scale and speed your firm operates at.

Series contents

1. The One Idea Behind Everything — the bookmaker’s craft, and why every phase is the same move scaled up.
2. Phase 1 — Foundations: Probability, Updating, and the Tools of the Trade — what probability really is, the core move of updating a belief, and the coin-flip mental model.
3. Phase 2 — Core Modeling: From Coin Flips to Real Decisions — models, sampling, MCMC, hierarchical models, regression, and separating skill from luck.
4. Phase 3 — From the Committee to the GPS: Tracking Things That Move — Kalman and particle filters for hidden, moving targets.
5. Phase 4 — Putting It All Together: Multi-Sensor Market Prediction — sensor fusion, Black-Litterman, regime-switching, and the full stack.
6. Use Case 1 — “Who Really Moves With Whom”: Portfolio Construction
7. Use Case 2 — “Which Factors Actually Matter?”: IPO Pricing and Feature Selection
8. Use Case 3 — “When One Domino Falls”: Cross-Asset Stress Testing With Private-Asset Look-Through
9. Use Case 4 — Asset Allocation: The Operating System
10. Use Case 5 — Performance & Risk Measurement and Attribution: The Scorekeeper

Part of the The Bayesian Investor — A Primer for the CIO series.

The One Idea Behind Everything

Before the phases, hold onto a single image.

Imagine a bookmaker setting odds on a match. Before the first whistle, he has a price based on his initial read of the two teams — call it his starting belief. Then the match starts. A striker gets injured. A goal goes in. The crowd changes mood. With every new piece of information, the bookmaker does not throw away his price and start from zero. He adjusts it — smoothly, continuously, weighing each new fact against what he already believed. And crucially, he always knows how sure he is: early in the match his price is nervous and can swing; late in the match, with the score settled, his price is firm.

That is the entire philosophy of this roadmap, in one picture:

Start with a belief. Update it as evidence arrives. Always know how sure you are.

Every phase in this plan is just a more powerful, more automated version of the bookmaker’s craft. Phase 1 teaches the basic move (updating a belief). Phase 2 turns it into a modeling discipline. Phase 3 makes it fast enough to track things that move continuously. Phase 4 wires it into the way your firm actually invests. Let’s walk through each.

Part of the The Bayesian Investor — A Primer for the CIO series.

Phase 1 — Foundations: Probability, Updating, and the Tools of the Trade

What probability really is

We use the word “probability” constantly, but let’s pin it down. A probability is simply a number between zero and one that says how likely something is. Zero means “it won’t happen,” one means “it definitely will,” and everything in between is a shade of maybe. A probability of 0.7 means “about seven times out of ten.” That’s all.

But there is an important fork in the road that determines which kind of probability you’re using, and it matters enormously in investing.

The first kind is frequency probability. It answers questions like “If I flip this coin a million times, how often does it land heads?” It is the probability of repeatable events — the world of coin flips, dice, and long historical records. An actuary counting past insurance claims lives here.

The second kind is belief probability — a measure of confidence about a single, one-off event. “How likely is it that this specific acquisition outperforms its underwriting?” That event happens once. It has no long-run frequency. There is no million-time history of this exact deal to count. And yet you make decisions on it every day. Belief probability is the honest way to put a number on that: it is your degree of confidence, expressed on the zero-to-one scale, informed by whatever you know.

For a CIO, the punchline is immediate: frequency probability is useful, but belief probability is the one your job actually runs on. Every deal, every market call, every regime call is a one-off event. This roadmap is about getting really good at the second kind — while remaining honest that it is a belief, not a fact.

The core move: updating a belief

Now the bookmaker’s move, formalized gently.

You begin with a prior — your belief before seeing new evidence. Then evidence arrives. You update. What you’re left with is your posterior — your belief after the evidence. And here’s the elegant part: today’s posterior becomes tomorrow’s prior. The loop never ends.

• Prior = your starting belief (e.g., “I think this property’s cap rate compression has a 40% chance of continuing”).
• Evidence = what you observe (e.g., the Fed cuts, transaction volume picks up).
• Posterior = your updated belief (e.g., “now I think it’s a 55% chance”).

Three rules of thumb make the whole machinery intuitive, and they’re worth internalizing because everything later is a fancier version of them:

1. Thin evidence, prior rules. When you have almost no data, your starting belief dominates. This is not a bug — it’s the honest recognition that with little to go on, your informed judgment is the best estimate you have.
2. Strong evidence, data rules. As evidence piles up, the data increasingly dominates and your starting belief fades. The bookmaker’s early nervousness gives way to a settled price.
3. Confidence grows with evidence. A belief updated by a hundred data points is firmer than the same belief updated by ten. The width of your uncertainty narrows — you don’t just get a better guess, you get a guess you trust more.

That last point is the deepest gift of this way of thinking, and it’s the thing most traditional finance tools quietly omit: an honest, ever-present measure of how sure you are. A number without its confidence is a rumor; a number with its confidence is a decision input.

The coin-flip mental model

Your team’s first exercise is trivial on its face: simulate coin flips. But it’s actually the perfect laboratory for the three rules above.

Suppose you believe a coin is fair — a 50/50 starting belief, a prior. You flip it ten times and get seven heads. Should you abandon your prior? No — ten flips is thin evidence, so your belief moves only a little. Now flip it a thousand times and get seven hundred heads. Now the evidence is overwhelming; your belief should move decisively toward “this coin is biased.” And notice the confidence: after ten flips you’d be unsure; after a thousand you’d be certain.

Doing this in simulation — having the computer run tens of thousands of flips at a keystroke — lets your team feel the rules rather than memorize them. It’s the difference between reading about how a river moves and standing in it.

Python, in words

Somewhere in this phase your team will start using Python. Here’s the only thing you need to understand about it as a decision-maker: Python is a general-purpose calculator that can run millions of imaginary what-ifs in seconds. It lets your analysts describe a belief and a piece of evidence in ordinary language-shaped code, and then have the computer do the heavy arithmetic — arithmetic that would take a human a lifetime, by hand, and is prone to human error besides.

Why the Python ecosystem specifically? Because the world’s best Bayesian tooling happens to live in it — free, battle-tested, and shared openly by a huge community of practitioners. Your team isn’t choosing a programming language for the sake of programming; they’re choosing the workshop where the modern tools of this discipline are kept. Phase 1 gets them comfortable walking around that workshop; the later phases put the tools in their hands.

Part of the The Bayesian Investor — A Primer for the CIO series.

Phase 2 — Core Modeling: From Coin Flips to Real Decisions

Coin flips are the sandbox. Now the team climbs out of it and builds models that describe real markets. This is the phase where they work through the Bayesian Analysis with Python material and its companion repository — the practical playbook of this discipline.

What a “model” is, in plain words

A model is a simplified story of how the world produces the numbers you observe. It is not the world. It is a flight simulator — not the actual plane, but a careful imitation that predicts how the plane behaves. Every investment team already runs models in their heads (“if rates rise, cap rates follow, so property values compress”). The difference here is that your team will learn to write those stories down with precision, so the computer can play them forward, test them, and tell you how confident the story makes you.

Every model has parameters — the dials or settings of the story. If the story is “property value depends on interest rates and income,” then the dials are things like “how strongly does each factor move the value?” Estimating means using the evidence to find the best settings for those dials, with an honest statement of how sure you are about each one.

Sampling, and why it’s not cheating

Here is the crucial move that makes modern Bayesian work possible, and it’s worth dwelling on because it is the single most misunderstood idea in the whole roadmap.

To know what your model believes, you could try to compute the answer with pure mathematics — but for real-world models that math is usually impossibly hard. So instead, your team does something almost cheeky: they sample — that is, they draw possible outcomes at random from the model, the way rolling a die over and over lets you feel its odds without calculating them. Draw enough samples and the picture emerges: the distribution — the full chart of possible outcomes and how likely each is — reveals itself from the frequencies of the draws.

Think of it as estimating the shape of a mountain range by taking thousands of random hikes and recording where you end up, rather than trying to derive the mountain’s shape from a surveyor’s formula. It sounds crude. It is not crude; it is a recognized, rigorous, and powerful technique — and it works precisely because randomness, repeated enough times, stops being random and becomes a reliable portrait of the underlying pattern.

The hiker in the fog: MCMC

The specific sampling technique your team will master is called MCMC (Markov chain Monte Carlo). Let’s unpack that awful acronym gently, because the idea behind it is actually lovely.

Imagine a hiker lost in a foggy landscape. She cannot see the whole terrain — the mountain peaks, the valleys, the shape of everything. She can only feel the ground directly under her feet. Her goal: figure out where the high ground is (where the likely answers live) and how much of the terrain is high vs. low. Her method: take a step in a random direction. If the ground rises, take the step — always. If the ground falls, take the step sometimes — because a dip might be a valley between two peaks, and crossing it could lead to even higher ground. Over thousands of steps, she spends time on each part of the landscape in proportion to how high it is. That visiting pattern is the answer.

MCMC is exactly this hiker. It explores the landscape of possible dial settings, spending time where the answer is most likely, and in doing so it produces a complete map of what the model believes — including all the uncertainty. The three letters just describe the hiking rules: it walks a chain of steps where each step depends only on where it currently stands (Markov), and it uses random draws (Monte Carlo, named after the casino — because it gambles its way to the answer). For an investor, the practical meaning is: your team can get a full, honest picture of what the data and the model together imply — without needing a single closed-form formula.

The committee vs. the organization: hierarchical models

The next big concept is the hierarchical model, and the analogy is organizational.

Imagine a firm with many small regional offices, each managing a handful of properties or a handful of positions. Each office has very little data on its own. A naive analyst in one office might conclude “with only five data points, I can’t estimate our vacancy risk at all.” But here’s the insight: these offices are not isolated islands. They share the same economy, the same market forces, the same broad dynamics.

A hierarchical model says: let every office learn from the whole firm, and let the whole firm learn from every office. Each office still gets its own tailored estimate — Boston isn’t forced to be identical to Houston — but it is gently pulled toward the firm-wide pattern, with the strength of that pull proportional to how little local data it has. When an office has lots of data, it stands on its own. When it has almost none, it leans on the firm.

Why this matters for a CIO: it’s the principled version of “borrowing strength.” Small samples no longer mean guesswork. A portfolio of many small exposures becomes a source of shared wisdom, not a collection of lonely estimates. This is how you get defensible numbers out of thin slices of data — and in commercial real estate, where individual deals are few and idiosyncratic, that is most of the job.

Drawing the line: regression

The team will also master regression — a word that sounds technical and means something simple. Regression is drawing the best-fitting line through history. If you plot historical data points — say, interest rates against cap rates — regression finds the straight line that comes as close as possible to all the points at once, and it draws a shaded band around that line showing how confident it is.

The line is the trend; the band is the honesty. A tight band means the relationship is clear and the forecast is trustworthy. A wide band means the data is scattered and the forecast should be treated as a guess. For a decision-maker, never look at the line without looking at the band.

The payoff: separating skill from luck

The finance application your team will build in this phase is the alpha/beta model, and it deserves a plain explanation because it’s genuinely important.

Think of any portfolio’s return as a coin with two sides. One side is beta — the part of your return that simply comes from the market moving. If the market rises and you’re invested, you rise with it; that’s not skill, that’s exposure. The other side is alpha — the part of your return that comes from your decisions, over and above what the market gave you for free.

The question every investor must answer honestly: did we add value, or did the market just go up? A naive analysis can fool itself, especially in a bull market. The Bayesian version separates the two cleanly, and — this is the key — it puts an honest uncertainty band around the answer. “Your alpha is positive” is one thing. “Your alpha is positive, but the band is wide enough that we can’t actually be sure” is a completely different, and far more valuable, sentence. This phase gives your team the ability to deliver the second kind of sentence.

Part of the The Bayesian Investor — A Primer for the CIO series.

Phase 3 — From the Committee to the GPS: Tracking Things That Move

There is one problem with the hiker-in-the-fog approach: it is thorough, but it is slow. The hiker needs tens of thousands of steps to map the landscape. For a one-time analysis, that’s fine. But what if the landscape itself is moving? What if you need to track an unobserved target in real time — the market’s hidden volatility, the “true” inflation rate, the fair value of an asset that is only indirectly observable?

This phase is about the pivot: from a slow, thorough committee review to a fast, streaming instrument.

Why the pivot is necessary

Think of the difference between a formal committee review and a GPS navigation system.

A committee gathers, considers all the evidence, deliberates, and produces a considered judgment. It is excellent — but it is slow, and by the time it finishes, the situation may have changed. A GPS does something different: it never stops updating. Every second it receives new measurements, fuses them with its current estimate, and produces a current position with a circle of uncertainty around it — and it does this continuously, forever, as the target moves.

Some economic targets are committee-shaped (you analyze them occasionally, in depth). But many — especially volatility, the market’s hidden fear gauge — are GPS-shaped. They change continuously, they’re never directly observable, and decisions need a now answer, not a last-quarter answer. This phase builds the instruments for GPS-shaped problems.

The Kalman filter: the GPS itself

The centerpiece is the Kalman filter, and the analogy is almost too good: it is a GPS, mathematically.

Here’s the problem it solves. You have many noisy sensors — each one individually unreliable. A GPS has satellite signals that bounce and drift; your sensors might be price data, option-implied signals, transaction prints, all of them noisy. You also have a belief about where the target is and how it’s moving. How do you combine all of it into one good estimate?

The Kalman filter does two things in a loop, forever:

1. Predict: “Given where I think the target is and how it’s moving, where should it be now?” (The bookmaker’s prior, updated for time passing.)
2. Correct: “Now the sensors have spoken. How much should I trust them versus my prediction?”

And here’s the elegance: the filter weighs each source by how trustworthy it is. A very noisy sensor gets little weight; a very precise one gets a lot. If the sensors suddenly disagree wildly with the prediction, the filter notices and starts trusting them less. It is self-correcting — a smooth, continuously updated estimate that never panics and never ignores evidence.

For your firm: this is how you track an unobservable target like market volatility — the hidden state of fear and calm that drives option prices and risk budgets — in near real time, from noisy observable signals, with an honest uncertainty circle that shrinks as the sensors get better.

The particle filter: thousands of tiny scouts

The Kalman filter is brilliant but has limits: it assumes the target moves in fairly smooth, well-behaved ways. What if the truth is wilder — jumping regimes, sudden shifts, non-linear dynamics? Enter the particle filter, which is the scout version.

Instead of one estimate with one uncertainty circle, imagine dispatching thousands of tiny scouts, each carrying a guess about where the target is. The scouts wander according to your model of how the target moves. Then the evidence arrives — and each scout is graded on how well its guess matches what was actually observed. The scouts with good guesses survive and multiply; the ones with bad guesses die off. Repeat, and the surviving scouts crowd around the truth, wherever it is — even if the truth jumps around in wild ways.

It’s evolution, applied to estimation: survival of the fittest guess. Where the Kalman filter is the elegant GPS for smooth motion, the particle filter is the rugged tracker for unpredictable motion. Your team will learn both, because real markets do both: sometimes smooth, sometimes jumpy.

The payoff: volatility, and beyond

The concrete finance payoff in this phase is dynamic tracking of hidden, time-varying market volatility — the market’s shifting mood of fear and greed that no one can observe directly but everyone’s decisions depend on. With Kalman and particle filters, your team can estimate it continuously, in real time, from the noisy signals the market leaves behind, with an honest statement of how sure the estimate is at every moment.

And volatility is just the first target. The same machinery tracks any hidden, moving economic quantity: the “true” rate of inflation, the fair value of an asset whose worth is only glimpsed through noisy transactions, the health of a property market read through noisy data feeds. Any time the thing you care about is invisible but moving, this is the toolkit.

Part of the The Bayesian Investor — A Primer for the CIO series.

Phase 4 — Putting It All Together: Multi-Sensor Market Prediction

Now the roadmap pays off. Phases 1 and 2 gave you beliefs and models; Phase 3 gave you speed. Phase 4 fuses them into a working investment system, applied to the real problems of public markets and commercial real estate.

Sensor fusion: the air-traffic controller

The master image for this phase is the air-traffic controller in a busy control tower.

The controller has many sources of information: radar blips, radio calls from pilots, satellite weather feeds, flight plans filed hours ago. No single source is complete. Radar can be fuzzy; a pilot’s radio call is instantaneous but partial; weather feeds are global but indirect. The controller’s job is to take all of these noisy, partial streams and build one consistent picture of what is in the air right now — and then act on it.

This is exactly what your firm does with markets. Consider a commercial real estate thesis: should you invest in a certain corridor? The evidence streams in from many directions — satellite imagery of parking lots (how full are the lots?), foot traffic data (how many people walk the streets?), credit-card and leasing data (are tenants actually paying?), macro signals (is the economy expanding?). Individually, each feed is noisy and incomplete. Fused together, weighted by reliability, they become a picture: is this corridor genuinely getting busier, or is the satellite just catching a holiday weekend?

Sensor fusion is the discipline of combining many noisy, partial views into one weighted, confident view. It is the same move the bookmaker makes — many small pieces of evidence, each weighed by trust, merged into a continuously updated belief. In Phase 4, your team wires the Phase 1 updating loop to the Phase 3 streaming instruments, and points them at the Phase 2 models. The satellite feed doesn’t replace the foot-traffic feed; it corrects and is corrected by it. That’s fusion.

Black-Litterman: the portfolio doctor

One of the most famous tools in this phase is the Black-Litterman model, and it deserves a story because it’s genuinely intuitive.

Imagine a patient — your portfolio — whose treatment plan is currently the market’s equilibrium. An equilibrium, in this context, is simply the price the market as a whole has settled on for every asset — the balance point where the crowd’s buying and selling have canceled each other out. There’s a standard prescription: the market portfolio, priced at that settled state. It’s a reasonable starting point, but it reflects the average opinion of everyone, not yours. Now suppose your analysts walk in with a strongly-held, well-researched second opinion: “We believe energy will outperform,” “We think European real estate is mispriced.”

The naive responses are both wrong. Ignore the analysts, and you’re just a passive index fund with a fixed view. Follow them blindly, and you’re betting the firm on possibly overconfident opinions. Black-Litterman is the portfolio doctor who reconciles the two:

• Start with the market’s equilibrium — the baseline prescription.
• Let the analysts’ views adjust that baseline — but only by an amount proportional to how confident each view is.
• A confident, well-supported view moves the portfolio meaningfully. A shaky, speculative view barely nudges it.
• The result: a portfolio that respects both the market’s wisdom and your team’s edge, never letting either run the show alone.

For a CIO, this is the difference between opinion and process. Your analysts will always have views; the question is whether those views enter the portfolio as disciplined, confidence-weighted adjustments or as unexamined convictions. Black-Litterman is the disciplined channel.

Regime-switching: the model that notices the weather changed

Finally, the roadmap’s most advanced application: regime-switching — and the analogy is a well-tuned thermostat that knows more than temperature.

Markets don’t behave the same way forever. There are regimes — periods where the market is fundamentally in a different mood: risk-on rallies, risk-off selloffs, quiet sideways drift, crisis conditions. The rules that govern returns in one regime are different from another. A model that assumes “the market always behaves like it did last year” is like planning a year of outdoor events using last year’s weather.

A dynamic Bayesian network — and that phrase is simpler than it sounds — is a model that tracks these hidden states. “Bayesian” means it updates with evidence, as we’ve established. “Network” means it tracks several related things at once — volatility, the correlations between assets (how closely their prices move together), momentum — as one connected system. And “dynamic” means it does this over time. Together: a system that watches many market signals at once and quietly maintains a belief about which regime we’re in right now.

The payoff is the quiet announcement: “Based on the evidence, we now believe we’re in a risk-on regime.” And because it’s Bayesian, that announcement always comes with its confidence: “we’re fairly sure” versus “we’re not sure yet — too early to say.” Your risk budgets, your allocations, your hedging decisions can then respond to the regime — not reactively after a crisis, but as the evidence shifts, the way the bookmaker’s price moves through the match.

The full stack: how the four phases compose

Let’s show how it all connects, because a CIO needs to see the machine, not just the parts.

• Phase 1 taught the atom: start with a belief, update with evidence, carry your confidence.
• Phase 2 built the molecules: models of markets and portfolios, estimated by sampling, with hierarchical structure that borrows strength and uncertainty bands that tell the truth.
• Phase 3 added the engine: filters that keep those models current as the world moves, tracking the invisible (volatility, fair value) in real time.
• Phase 4 assembled the system: sensor fusion feeding the models, Black-Litterman turning views into disciplined portfolio adjustments, regime-switching re-weighting the firm as the market’s mood turns.

The bookmaker, in other words, has been given an entire trading desk. Every piece is the same move — belief, evidence, update, confidence — scaled up from a coin flip to a global investment platform.

Part of the The Bayesian Investor — A Primer for the CIO series.

Use Case 1 — “Who Really Moves With Whom”: Portfolio Construction

What it does. This builds a map of direct relationships between asset classes — how tightly each one is tied to the others once you’ve accounted for everything else — and then rebuilds that map separately for each market regime (bull, bear, neutral). The map then drives a defensive portfolio: the set of holdings that keeps the overall book as calm as possible.

The method. (Bayesian sparse precision estimation with a graphical horseshoe prior for the map, then regime-conditional minimum-variance portfolio weights from it.)

The analogy — the party map. Imagine a party where everyone is in the same room. Two people chat, but only because they both happen to know the host — not because they’re actually friends. A naive observer watching the room might conclude “these two are close.” A sharper observer notices: remove the host from the equation, and they have nothing to say to each other. That’s the difference this tool makes. Most traditional tools look at the whole room and call it a relationship; this tool asks which bonds are real and which are just both people standing next to the same host.

Why it matters for investing. Correlation — how closely two things have moved together in the past — is a noisy guide. It can be fooled by a shared driver (both assets responding to “risk-on” or “risk-off” sentiment) and by plain randomness. The Bayesian version asks a sharper question: if I hold everything else fixed, is this pair genuinely tied? And it answers with a confidence level — “we’re 95% sure this link is real” — not a yes/no guess. For a CIO, that’s the difference between building a book on rumors about who’s friends and building it on verified relationships.

Why it beats the traditional approach. Traditional analysis that simply measures how closely things have moved together in the past treats every link in the room as equally real, so noise gets mistaken for connection and the map is cluttered with fake friendships. This tool cleans the map — it lets the data decide which links survive, with honest uncertainty on every one. Just as importantly, it recognizes that the map changes with the weather: in calm markets, assets drift apart (diversification works); in stress, they suddenly lock together (diversification fails — exactly when you need it most). A traditional tool assumes one fixed map for all seasons and gets blindsided. This one redraws the map as the season turns.

Part of the The Bayesian Investor — A Primer for the CIO series.

Use Case 2 — “Which Factors Actually Matter?”: IPO Pricing and Feature Selection

What it does. When a company goes public, there are dozens of candidate explanations for how much its shares will pop on day one. This tool takes all the candidates, lets the model itself discover which few genuinely drive the outcome, and then groups the deals into peer clusters discovered in the data rather than taken from an industry label.

The method. (Bayesian horseshoe regression for feature selection, then a stochastic block model for data-driven peer clustering.)

The analogy — the needle-finder. Imagine a workshop table covered in hundreds of dials, and you’re told only a handful actually do anything; the rest are decorative. The Bayesian tool is a mechanic who starts with the dials all turned to “off,” then lets the evidence earn each dial being turned on. A dial only lights up if the data strongly insists it matters. You’re left with a short, trustworthy list of the dials that actually work — and a confidence level for each. A traditional approach would either guess which dials matter in advance (introducing human bias) or turn everything on (flooding the analysis with noise).

The analogy — the talent scout vs. the HR form. Traditional peer groups come from a form: “what industry is this company in?” But the companies that actually behave alike may not share an industry sticker. The talent scout watches how people actually perform and groups them by what they genuinely have in common — a young company with an aggressive investor behaves like another young company in a totally different industry. This tool finds those real peer groups in the data, so “comparable companies” means companies that actually compare.

Why it matters for investing. Deciding which IPOs to participate in is a one-off, high-uncertainty decision — precisely the territory where belief-probability shines. Knowing which few drivers actually matter (and how confident we are), and grouping deals by genuine similarity, gives your team an edge in allocation and underwriting that a sector-label analysis can’t provide.

Why it beats the traditional approach. Traditional regression keeps every candidate factor in the model, so noise factors muddy the signal and confidence statements are weak. The Bayesian version automatically selects the few real factors, attaches an honest probability to each, and builds peer groups from actual behavior rather than arbitrary labels — all without an analyst pre-committing to a model.

Part of the The Bayesian Investor — A Primer for the CIO series.

Use Case 3 — “When One Domino Falls”: Cross-Asset Stress Testing With Private-Asset Look-Through

What it does. This answers the single most important question for an investment committee: if one asset is hit by a shock, what happens to everything else — including the private positions we can’t directly observe or mark daily? It traces the direct transmission of stress across public markets and then “looks through” private-equity and real-estate funds to their public exposures, so the committee can see how an unlisted holding would react to a market event.

The method. (Bayesian graphical model with bootstrap posterior edge inclusion for the stress map; public-comp look-through for private positions; stale-mark detection via graph-shift monitoring.)

The analogy — the domino table. Set up a row of dominoes. Tapping one domino doesn’t knock down the whole table at random — it knocks down the ones actually touching it, which then knock down the ones touching them. The path of the fall follows the real connections, not the surface similarities. If two dominoes happen to be painted the same color, they don’t move together unless they’re physically linked. This tool maps the real physical links, then lets you tap any domino and watch the true path of the fall.

The analogy — the X-ray. A private fund’s value isn’t published daily; it’s a mark that gets updated periodically. But the fund holds investments that respond to public markets. The X-ray sees through the fund’s wrapper to its underlying public exposures, so a market shock can be translated into “what this private position is probably doing right now” — even before the next formal mark arrives.

The analogy — the freshness label. When the public mirror of a private position starts drifting away from the last private valuation, the tool flags it: “this private mark may be stale — the world has moved, but the number hasn’t.” That’s a quiet, valuable warning that the committee’s picture of the private book is out of date.

Why it matters for investing. For an institution with private assets, the hardest reporting problem is: “if markets fall 10%, what’s our total book doing, including the stuff we can’t mark daily?” This tool gives the committee a defensible, transparent answer — and flags when private valuations need refreshing.

Why it beats the traditional approach. Correlation-based stress testing confuses “in the same room” with “directly connected,” so it mis-attributes which shocks actually travel where. This tool traces direct links, attaches confidence to every claim, looks through private wrappers to their real exposures, and alerts when private marks go stale — none of which the traditional single-number stress test can do.

Part of the The Bayesian Investor — A Primer for the CIO series.

Use Case 4 — Asset Allocation: The Operating System

What it does. Asset allocation is the highest-level decision a firm makes: how to split capital across everything you can own. The way to think about a modern multi-asset allocation is not as a better version of a fixed 60/40 stock-and-bond split (60% equities, 40% bonds — the classic default portfolio) — it is an operating system. The product being sold isn’t the asset mix; it’s the owner’s objective and constraints. Every piece of the allocation exists for one reason only: it improves the chance of hitting that objective. And the way to build it is governance first, then the portfolio math, then the plumbing.

The method. (A suite, detailed below: Black-Litterman for the views channel, hierarchical models for private-asset premia, state-space unsmoothing for private NAVs, posterior predictive simulation for the objective, and Bayesian model averaging for the allocation engines.)

Why Bayesian methods belong here at all. The payoff from this entire roadmap grows the moment the book stops being purely liquid public markets and starts including alternatives and real assets — private equity, real estate, infrastructure, private credit. That is exactly where the data are short, stale, smoothed, and not directly comparable to public markets. Bayesian thinking is the honest way to say: we have a prior about how the world works, the data are weak, and we will not let one lucky 12-year stretch of private-fund history dominate the whole allocation.

The owner’s contract — the voyage brief

Before any math, write down the destination as numbers, not slogans.

• The success metric. What actually has to happen? “We can keep paying the required spending,” or “the funded-status floor never breaks” (funded status = how well the pool’s assets cover what it owes to beneficiaries), or “real wealth holds after inflation” — over a defined horizon.
• The risks that actually bind. Not just “volatility this year” but the storms you must survive mid-voyage: the deepest drawdown you can tolerate (drawdown = how far the value falls from its peak), how much cash you need to cover a once-in-twenty-years bad year (an outflow, or a capital call — a request from a private fund asking you to pay in more committed money), and whether the board still needs a reference benchmark to measure against.
• Decision rights. What the board sets once a year (the destination, the risk budget, what’s out of bounds) versus what the team can move freely inside ranges.

If this layer is vague, every layer below becomes politics. The owner’s objective and constraints are the product; everything else is just how you try to hit them.

The exposure map — translate vehicles into risk ingredients

The next move is to stop thinking in asset-class names and start thinking in risk ingredients — the underlying forces that actually drive returns. Throughout, “exposure” simply means how much of your value is tied to a given force. The clean way to build a ladder of exposures:

• cash and interest-rate duration (how long your bonds take to mature)
• inflation and real-rate exposure
• credit (the premium for lending to borrowers who might fail) — the extra return you demand for the risk they won’t pay back
• equity — in excess of what the firm’s own debt and country already give it
• country and currency
• illiquidity and complexity (the premium for being locked up) — the extra return you demand for not being able to sell quickly
• real-asset specifics: occupancy, commodity prices, development, vintage (the year a private fund was launched, which shapes its economics)

The key trap this avoids. Private equity is not “another equity sleeve.” It is public-equity exposure plus leverage plus illiquidity plus control plus stale marks (stale marks = private valuations that lag reality, the freshness-label problem from Use Case 3). Real estate is duration plus inflation plus occupancy plus leverage (leverage = borrowing to amplify the position). If you allocate by vehicle name, you double-count the equity part and think you’re diversified when you aren’t — the two “different” sleeves are quietly the same ingredient twice.

Analogy — the recipe’s ingredients. A menu lists dishes, but the kitchen runs on ingredients. Two dishes that look different may share the same base stock. Allocate by ingredient, not by what the dish is called, or you end up serving the same thing twice and calling it variety.

The reference portfolio — the no-views recipe

Build a reference portfolio — the allocation you would hold if you had no tactical opinions and no scarce access to star managers. This is the “standard menu”: the market’s settled wisdom, in factor-ingredient form, with a conservative premium for illiquidity. Then set ranges around exposures, not around fund names.

This is the prior. Everything Bayesian later hangs on it: it is your starting belief about the world, before any team member’s view and before any fund pitch. It is not a target to defend — it’s the anchor you pull away from only when evidence earns the pull.

Several allocation engines, not one

Do not pick one allocation religion. Run a small committee of uncorrelated allocation methods, each with a different strength and a different typical failure:

The point: the solution’s target allocation is a blend of engines, not the output of one optimizer run. This is how you diversify the allocation decision itself — the same way you’d diversify a portfolio. (We’ll see in the Bayesian section how the engines’ votes get blended by how well they’ve actually performed.)

Cheap beta first, then implementation

For each target exposure, the implementation order matters (“beta” here is the market-exposure component of a return — the part that simply rides the market, as defined in Phase 2):

1. Buy the commodity first — commoditized exposure as cheaply as possible (index products, futures, pooled funds).
2. Add active or factor sleeves only where there’s reason to believe the remaining piece isn’t commoditized — where a skilled hand genuinely adds something the market isn’t already pricing.
3. Use overlays — separate derivative instruments layered on top of the book (derivatives are contracts whose value is tied to something else — here, to duration, equity, or currency) — so that illiquid holdings don’t freeze the total risk mix. When you can’t move the private part quickly, you can still move the public part to keep the whole book on target.

Analogy — buy the commodity, not the story. If you want the exposure to copper, you don’t need to own a copper mine and its management drama; you can buy copper itself. Save the mine for when you genuinely believe you can out-pick other buyers of copper.

Alternatives and real assets as a cash-flow system

This is where most solutions go wrong. Alternatives are not a weight; they are a commitment, pacing, and liquidity problem (liquidity = how easily and how quickly something can be turned into cash).

• Target the economic exposure (e.g. 15% of the book as private-equity value), not “15% committed” — because money gets committed years before it’s deployed.
• Run a pacing model: track unfunded commitments, expected calls and distributions, and the path of net asset value over time (net asset value, or NAV = the current estimated worth of the fund’s holdings).
• Size a liquidity sleeve — a dedicated portion of the book held in cash-like form precisely so it can answer short-notice demands — to a joint stress: public drawdown plus elevated capital calls plus required spending, all at once.
• Measure the risk in unsmoothed space (see the Bayesian section) — don’t be fooled by the calm-looking appraisal numbers. (An appraisal is an expert’s estimate of a property’s value, typically refreshed only periodically — hence “appraisal numbers” for real-estate and private-fund valuations.)
• Rebalance the public assets so total factor exposure stays on target while the private NAV drifts on its own.

Why this matters. Without overlays and pacing, a 20% private-equity target quietly becomes 12% after a public rally and 28% after a public crash — the exact opposite of what you wanted. The private tap turns slowly; the public dial must do the steering.

Analogy — the slow tap, not the faucet. A private commitment is like a tap you turn and wait for water: money is called slowly, over years, in unpredictable amounts. You can’t turn it off fast. So you keep a reservoir (the liquidity sleeve) and use the fast public dial to compensate while the slow tap does its thing.

Intra-horizon risk and stops

Risk must be monitored on the path, not just at year-end.

• Watch intra-horizon risk — the worst the book can look between measurement dates — not only the final number.
• Pre-commit de-risk rules at the exposure level (e.g. “if this exposure breaches its band, we trim it”) rather than reacting deal-by-deal.
• Run total-portfolio factor risk, including the lagged sensitivity of private positions (they react to markets with a delay, because marks lag).
• Attribute performance properly: what came from the allocation engine, what from the implementation vehicle, and what is just stale-mark noise.

Analogy — the autopilot’s alarm. You don’t wait until the plane has drifted far off course to notice; the alarm sounds the moment it leaves the corridor, and the correction is automatic. Pre-committed rules beat heroic last-minute judgment.

Where Bayesian methods belong — the disciplined referee

Here is how the whole Bayesian toolkit slots into the operating system.

The master identity (in plain words). Bayesian updating always has the same shape: what you end up believing is proportional to what the data say, multiplied by what you believed before. In allocation language: shrink noisy sample estimates toward an economically meaningful reference, with the amount of shrinking equal to how much you trust the data versus the reference. When data are rich (US large-cap, Treasuries), the answer hugs the data. When data are poor (secondaries, timber, emerging-market private equity), the answer hugs the economic prior. That is the whole point — and it prevents a clean backtest (a strategy’s glowing record on historical data) from bullying the rest of the book.

1. Black-Litterman as the views channel. The reference portfolio is the prior; the allocation engines’ forecasts are the views. The Bayesian referee blends the two, weighting each view by how much the engine that produced it has actually been right — not by how loud or confident it sounds. Weak views barely move the book; tight, proven views move it meaningfully. This is exactly how you combine several allocation engines without letting the loudest forecast grab the whole risk budget.

Two discipline notes:

• The optimizer must be fed the uncertainty about the forecasts, not just the forecasts themselves — otherwise you pretend you know the future mean when you don’t.
• Do not feed private-asset “views” into this channel as if quarterly appraisal numbers were daily traded prices. That is how you manufacture fake confidence.

2. Hierarchical models for alternatives. This is the piece genuinely designed for private assets — the multi-office firm analogy from Phase 2, applied to funds. Build the hierarchy: asset class → strategy/geography/sector → vintage/fund → deal. Each fund’s expected premium is pulled toward its strategy’s average; each strategy toward its asset class; each asset class toward a long-run economic prior (the premium for being locked up over the public equivalent — not over cash).

This does three things a raw sample average cannot:

• A six-year-old infrastructure fund does not get an 18% expected return just because its first few deals were lucky.
• Information is shared across funds with similar economics — a young fund borrows strength from its older, better-measured cousins.
• You get a whole distribution of expected premia, so you can allocate by “probability of helping the objective,” not by a single number.

This is the right tool for “we own forty funds and none of them has a clean thirty-year history.”

3. Unsmoothing — remove the filter that hides the true swing. Appraisal-based real assets and private NAVs are smoothed: this quarter’s reported number is partly last quarter’s number, partly the hidden true move, plus noise. A Bayesian tracking model (the GPS from Phase 3) treats the true return as the hidden target, estimates how much smoothing is happening, and produces a posterior for the true volatility and the true correlation with public markets. You then optimize on the unsmoothed posterior — not the brochure numbers.

Why you must. Without this, every real-asset sleeve looks like a free lunch: high return, low volatility, low correlation. The solution over-allocates to illiquids until the first period when the marks catch up or the distributions stop — and the free lunch disappears at exactly the worst time.

4. Predictive returns, not in-sample fits. What you actually care about is “what happens next,” with all the parameter uncertainty included. Run many possible future paths — public returns, private NAVs and cash flows, spending and calls — and choose the allocation weight that maximizes the probability of meeting the objective (e.g. “the chance that wealth never breaches the floor over ten years is at least 90%”). The classic optimizer that only balances expected return against risk cannot even express that sentence; the predictive path approach can.

5. Bayesian model averaging of the engines. Instead of picking risk-parity or valuation or cycle, let the engines vote — with votes weighted by how well each has predicted on data it was not trained on (out-of-sample prediction, the honest test), shrinking toward equal weight when evidence is weak. This is the multi-engine committee with an explicit scorekeeper. A cruder, honest version used in practice: fixed engine weights with wide uncertainty, updated slowly (once a year), never on a single quarter.

6. Views you can defend. Good views are stated as beliefs with ranges: “Buyout should earn a few points over leveraged small-cap public equity, not eight points over cash.” “Core real-estate duration is closer to long real bonds plus a modest illiquidity premium.” “This vintage is crowded; scale the premium down.” Bad views confuse data with belief: “This fund did 2.4x, so private-equity expected return is 16%” — treating a fund’s history (or a consultant’s table) as data rather than as another prior to be weighted.

The quarterly loop — how the machine runs

Annually (the board): confirm the objective, the risk budget, the liquidity floor (the minimum cash the book must always keep available), the allowed universe, and the engine weights.

Quarterly (the investment team):

1. Refresh public-market exposures back to target — cheaply.
2. Update private NAVs, unsmooth them, re-estimate their lagged sensitivity to markets.
3. Re-run the pacing model: are next-year’s capital calls fundable inside the liquidity floor?
4. Let the allocation engines vote; the Bayesian referee turns the votes into a target allocation.
5. Implement the changes with futures first, cash second, manager changes last. (Futures are standardized contracts to buy or sell an index or asset at a set price later — the fastest, cheapest way to move a public exposure.)

Continuously: cash, capital calls, currency hedging (protecting against currency swings), and compliance against the ranges.

Analogy — the ship’s cadence. The board sets the voyage and the storm limits once a year. The bridge adjusts course continuously, re-checks the fuel and the cargo each quarter, and only ever calls for a major crew change when nothing smaller can do the job.

Limits — where Bayes does not save you

• A bad prior, tightly held, is just ideology with better notation. For alternatives the prior is most of the result — so audit it. If the starting belief is wrong and confident, no amount of clever updating fixes it.
• Governance still dominates. If the board cannot hold the book through a three-year factor winter, no hierarchical model matters. The best posterior in the world is useless if the owner abandons it at the first bad stretch.
• Bayes does not create liquidity. It only stops you from pretending that appraisal volatility is tradeable volatility. The reservoir still has to be real cash.
• Don’t feed private NAVs into the views channel as if they were traded prices — you manufacture fake Sharpe ratios and fake confidence. (A Sharpe ratio is a single summary number: the return earned per unit of risk taken.)
• Watch the scoreboard incentive. If the manager is paid for beating a 60/40 benchmark while the real mandate is “don’t breach the spending floor,” the whole system will quietly game the posterior toward the scoreboard instead of the objective. Align the fee, not just the model.

Why it beats the traditional approach. Traditional allocation picks one method, runs it once, and lets a 12-year private-fund stretch or a clean factor backtest set the whole book. The operating system instead: starts from the owner’s contract, allocates by risk ingredients rather than vehicle names, runs a committee of engines, and lets a Bayesian referee — which respects how much the data can and cannot tell you — turn those engines into a target allocation. When data are strong, the data lead; when data are weak, the economic prior leads; and the owner’s liquidity and path constraints always have the final veto.

Part of the The Bayesian Investor — A Primer for the CIO series.

Use Case 5 — Performance & Risk Measurement and Attribution: The Scorekeeper

What it does. Every firm must answer two questions at the end of each period: where did our return come from? and where does our risk hide? This use case decomposes the portfolio’s return into its sources (which decisions added value, which just rode the market) and decomposes its risk into its contributors (which positions are the hazards), each with an honest confidence level.

The method. (A family of Bayesian attribution methods — see the catalog below for the specific method attached to each question, from Bayesian linear regression for return attribution through Bayesian risk decomposition for risk attribution.)

The analogy — the kitchen audit. After a great dinner service, the owner wants to know: was the success the menu (asset allocation), the cooking (security selection), or just the crowd (the market being in a good mood)? A naive look at the takings can’t separate these. The audit traces each dish back to its ingredients and each decision to its contribution — and, crucially, says how sure it is. “The menu added value” is one thing. “The menu added value, but we can’t actually be sure the crowd didn’t do it all” is a different, more honest, and more useful sentence.

The analogy — the fire-safety inspection. Risk is like fire hazard: you need to know where the risk lives, not just that “there is risk.” The inspection walks the building (the portfolio) and reports which rooms (positions) hold the most combustible material and how much of the total hazard each contributes. Some hazards are obvious (a big concentrated bet); some are sneaky (a cluster of positions that all catch fire together — a correlation risk that no single position reveals).

Why it matters for investing. This is the accountability engine of the firm. It tells you whether your process genuinely adds value or just rides the market’s tide, and it tells you where to concentrate risk-reduction effort. With confidence attached, it prevents both overconfidence (“we’re geniuses”) and false modesty (“we’re lucky”).

Why it beats the traditional approach. Traditional attribution gives point estimates — a single “alpha” number — with no sense of how much that number should be trusted. The Bayesian version separates skill from luck with an honest uncertainty band, decomposes risk into its true contributors (including hidden correlation risk), and knows when the honest answer is “too uncertain to say.”

Performance & Risk Attribution and Decomposition: A Focused Catalog

Here is a catalog of the scorekeeper’s tools — each answering one specific question, each with its analogy and its payoff. All of these are variations on the same Bayesian move: estimate, decompose, and attach an honest confidence level. The first group covers the core measurement questions; the second extends into timing, style, costs, and the deeper decomposition of both return and risk.

The common thread. Every row is the same bookmaker’s move, pointed at the scorekeeper’s job: take a number that looks like a single fact, pull it apart into its honest components, and attach a confidence to every piece — estimate the whole distribution, decompose it, and never report a single number without its band. That is the discipline this roadmap builds — and it applies just as cleanly to deciding how to split your portfolio as to explaining what last quarter actually earned.

Where the Use Cases Plug Into the Roadmap

• Use Cases 1 and 4 are portfolio-construction applications of Phases 1 and 2 (beliefs, models, and sampling) — building and allocating the book. Use Case 4 also draws on Phase 3 (the GPS/unsmoothing machinery for private assets) and Phase 4 (regime-awareness and the views channel) to run the full allocation operating system.
• Use Case 3 is the Phase 4 sensor-fusion and regime-switching machinery, applied to the hardest reporting problem: private-asset stress.
• Use Case 2 is a focused, high-conviction application of the Phase 2 modeling workflow to a one-off decision (IPO participation).
• Use Case 5 and the catalog are the Phase 2 alpha/beta discipline scaled into the firm’s full performance- and risk-reporting engine.

Every use case is one idea — start with a belief, update it relentlessly, and never lose sight of how sure you are — wearing a different hat. Your team now has both the machinery and the map of where to point it.

Factor Rotation — Regime-Aware Systematic Portfolio

Executive Summary

A systematic factor rotation strategy that combines two independent regime signals — one, that is slow-moving & lagged and the other more latent derived micro-structure signal that is contemporaneous — to dynamically allocate across 9 uncorrelated factor ETFs. The ensemble achieves 2.6× SPY’s Sharpe ratio with less than one-third the drawdown of SPY over a 10-year out-of-sample period.

Performance (Jul 2016 – Aug 2026)

Growth of $1 (with drawdown):

How It Works

1. Dual-Signal Regime Detection

Two independent signals identify market regime with no look-ahead bias:

• Slow — Lagged smoothed signal processed through a Gaussian Process + Dirichlet Process mixture. Detects macro regime shifts (expansionary vs contractionary).

• Real-Time — Latent variable derived from SPX prices detects stress before (in theory) it propagates to broader markets.

The two signals are near-independent (correlation 0.15), providing complementary coverage: One catches slow-moving liquidity cycles, while the other catches fast-moving stress events.

2. Factor Rotation via Bayesian Optimization

Monthly allocation across 9 uncorrelated factor ETFs:

The Bayesian optimizer maximizes risk-adjusted returns while penalizing turnover. The regime signal modulates risk aversion in real-time — when stress is detected, the optimizer naturally reduces exposure to risky factors.

3. Downside Protection

During stress periods, the strategy’s risk aversion parameter increases by 2.5×, causing the optimizer to shift toward defensive factors (low-vol, bonds, gold) without explicit override rules. This is thus more structural protection, than tactical timing.

Transaction Costs

The strategy trades liquid, highly-quoted ETFs with tight spreads. Transaction costs reduce Sharpe by approximately 3bps..

The optimizer includes a built-in turnover penalty that limits unnecessary trading. Rebalancing occurs only when weight changes exceed 5%.

Key Takeaways

• Superior risk-adjusted returns: Sharpe 2.04 vs SPY 0.90 (2.6× improvement)
• Structural downside protection: Max DD -7.89% vs SPY -25.48%
• Regime-aware allocation: Automatically reduces exposure during microstructure stress
• Low cost drag: Transaction costs reduce Sharpe by only 0.03
• Fully systematic: No discretionary decisions, no parameter overfitting

Important Disclosures

• Out-of-sample results: All performance shown is out-of-sample (Jul 2016 – Aug 2026). In-sample training period used for model calibration only.
• No guarantees: Past performance does not guarantee future results.
• Model risk: Regime detection models may produce false signals. Strategy relies on two independent signals to mitigate this risk.
• Leverage risk: The strategy uses leveraged ETFs (SSO) which carry additional risks including decay in choppy markets.

General

Essays, notes, and work samples on markets, probability, and AI.

Dashboard Samples & Artifacts

Selected screenshots from work across performance measurement, agent-based simulation, and multi-family operations.

Real Assets

Notes on commercial real estate, infrastructure, and private markets. Content coming soon.

License

Copyright (C) 2026 Pramod Thaz. All rights reserved.

This material may not be reproduced, displayed, modified, or distributed without the express prior written permission of the copyright holder.

🔒 Usage & Licensing

This repository does not have an open-source license. All rights are reserved by the copyright owner.

• Allowed: Viewing the code, forking the repository on GitHub.
• Not Allowed: Downloading, modifying, or using this code in other software.

If you would like to use this code for a specific project, please open an Issue or contact me directly to request a license.