Home › Articles

Quantitative Investing in US Stocks: Before Quant Trading, Quantify the Market Regime

2026-10-16 · Research note · Not investment advice; figures are as of the sources and dates cited in the text · 中文
Contents7 sections
  1. What quant is, and what it is not
  2. The four layers of a US stock quant framework
  3. The order of judgment matters more than the score
  4. How do you know whether this quant method works?
  5. Five common traps in quant trading US stocks
  6. Where to start
  7. Frequently asked questions

Many people search for "quantitative investing in US stocks" or "quant trading US stocks" hoping to find a strategy that places orders automatically and makes money reliably. This article does not sell that. It covers the upstream step of quant trading that is most often skipped: use rules and public data to calculate what state the US stock market is in right now, and make sure the method can be checked against what actually happened afterwards. This site (QuantLabs) is built on that idea, and below we use its actual design as an example of what a quant framework for the US stock market looks like and where each layer can go wrong.

This article is methodology and educational content. It is not investment advice and not a trading signal. All thresholds and weights in it are the settings this site actually uses, and you can find them on the methodology page and the individual indicator pages.

What quant is, and what it is not

"Quant" comes down to three things: fixed rules, verifiable data, testable results. The same inputs always produce the same output, with no reliance on how anyone feels that day; every input can be traced to a source and a date; every judgment can be checked against the real market path afterwards.

Quant trading is usually split into two layers:

Layer Question it answers Typical approach
Market state (regime) Are we in expansion, transition, recession, or tail risk? How much risk should we carry? Macro indicators such as credit spreads, volatility structure, valuation, and liquidity
Strategy and execution In this state, what to buy, when to enter and exit, and how large a position? Factor models, momentum, mean reversion, execution algorithms

Most introductory material jumps straight to the second layer and backtests a moving-average crossover strategy without asking "does this strategy still work in the years when credit spreads blow out?" This site only does the first layer, and deliberately does not do automated trading.

The four layers of a US stock quant framework

1. Data layer: only use data whose source you can point to. Every day this site pulls public sources including FRED (the Federal Reserve economic database), Yahoo Finance, CBOE, the CNN Fear & Greed Index, and SEC filings, for a total of 105 indicators. Every number passes a plausible-range check: a value that was parsed incorrectly (for example, a CAPE of 41.7 scraped as 4.17) is left blank rather than written in.

2. Indicator layer: turn raw data into meaningful quantities. For example:

  • HY OAS (high-yield credit spread): what it costs companies to borrow, the most direct thermometer of credit stress.
  • VIX term structure (VIX ÷ VIX3M): when short-term fear is higher than longer-term expectations (above 1), the market is paying up for near-term protection.
  • ECY (Excess CAPE Yield): 1/CAPE minus the real interest rate. It measures stock valuation relative to bonds and, unlike CAPE alone, takes interest rates into account.
  • Fed net liquidity: the Federal Reserve balance sheet minus the Treasury General Account (TGA) and reverse repo (RRP), an approximation of "money available to the market."

3. Scoring layer: condense many indicators into a comparable score. This site's Market Score (0–100) is a weighted combination of six components: liquidity 25%, valuation 20%, momentum 20%, macro 15%, credit 10%, volatility 10%. A component with missing data is not scored as 0; instead its weight is redistributed to the components that have data, because 0 means "very bad," and treating missing values as 0 would systematically distort the result.

4. Rule layer: beyond the score, you need non-negotiable hard rules. This site's Kill-Switch is decided by code, not by AI:

Condition Threshold
HY OAS Above 600 bps
HY OAS one-week widening More than 100 bps
VVIX (volatility of VIX) Above 140
Sahm rule ≥ 0.5 and HY OAS above 400 bps Both true at the same time

If any condition is met, the state is set directly to "tail risk," no matter how high the score is.

The order of judgment matters more than the score

Many quant dashboards average all their indicators into one score. The problem is that indicators are not equal: when credit spreads blow out, whether valuations are cheap hardly matters. This site combines its three dimensions in a fixed order:

  1. Credit sets the baseline (HY OAS below 400 bps is a precondition for expansion).
  2. Sentiment can shift the judgment by at most one level (if VIX jumps to 30 but credit is calm, the most it can produce is "transition," never a direct jump to "recession").
  3. Valuation does not affect the current week's state, only long-term expected returns.

The corresponding reference allocation:

State Stocks Short-term bonds Long-term bonds
Expansion 70% 20% 10%
Transition 50% 30% 20%
Recession 30% 30% 40%
Tail-Risk 20% 60% 20%

This table is part of the framework, not personal allocation advice; everyone's age, cash flow, and risk tolerance are different.

How do you know whether this quant method works?

What separates quant from gut feeling is that it can be proven wrong. This site has two validation mechanisms:

Predictive power (IC) backtest. For each indicator, we compute its Spearman rank correlation (Information Coefficient) with forward 4-, 13-, and 26-week returns of the S&P 500, the Nasdaq 100, and the Philadelphia Semiconductor Index. The results are on the predictive power research page (Chinese). There are three traps here that are easy to overlook:

  • Overlapping windows: computing the "forward 13-week return" every day means adjacent days have almost identical windows. 1,000 days of data do not give 1,000 independent samples; the effective sample size is roughly the number of days divided by the holding period, so this site only shows direction when the effective sample is below 20.
  • Direction is chosen in-sample: if you look at the sign of the IC first and then compute a "hit rate," the hit rate is structurally biased upward. Pure noise can reach about 57% in small samples.
  • Revised data: macro data such as CPI, M2, and the Sahm rule get revised, so a backfill picks up the revised values seen today, not the numbers the market saw at the time. This makes backtests of these indicators look better than they really are.

Signal track record. Every day's state judgment is logged, and the real S&P 500 path is later used to fill in the 4-week and 13-week returns, with statistics grouped by state. The results are on the market regime backtest page. Until there are enough samples, they are only a directional reference.

Five common traps in quant trading US stocks

  1. Look-ahead bias: the backtest uses data that did not exist at the time, for example using today's revised CPI to backtest last year's decisions.
  2. Overfitting: tuning parameters until the historical performance looks good. The more parameters and the fewer samples, the more likely you are just fitting noise.
  3. Treating a single signal as a bottom-picking tool: signals such as a VIX spike or extreme breadth often show up near lows, but in a multi-leg bear market they can come several months early. Further reading: "Signal Combinations Can Frame a Bottom, but Cannot Pinpoint the Low".
  4. Ignoring trading costs and taxes: for non-US residents, dividend withholding tax and estate tax risk on US-domiciled ETFs can eat a meaningful share of returns. This site's ETF tax page (Chinese) compares the long-term net value of VOO / QQQM against Irish-domiciled UCITS (CSPX / CNDX).
  5. Looking only at price, not credit: a falling stock price is not the same as a crisis. Historically, declines without a matching widening in credit spreads have mostly been corrections; only when both deteriorate together is it a signal to reduce risk.

Where to start

The value of quant research is not in predicting the next candlestick, but in making judgments rule-based, recorded, and testable. Quantify the market regime first, then talk about strategy, so you do not end up running the right strategy in the wrong environment.

Frequently asked questions

What is quantitative investing in US stocks?

Quantitative investing in US stocks means studying the US market with fixed rules, verifiable public data, and testable results, so the same inputs always produce the same output. This site only does the upstream part of quant trading: first calculate whether the US market is in expansion, transition, recession, or tail risk, and only then talk about strategy.

Do you have to write code to do quant trading in US stocks?

This article describes a framework, not code: what matters is that the rules are fixed, the data can be traced to a source and a date, and judgments can be checked afterwards. This site only does the market-state layer, and deliberately does not do automated trading.

How is the Market Score calculated?

The Market Score (0–100) is a weighted combination of six components: liquidity 25%, valuation 20%, momentum 20%, macro 15%, credit 10%, volatility 10%. A component with missing data is not scored as 0; its weight is redistributed to the components that have data. If any Kill-Switch condition is met, the state is set directly to tail risk, no matter how high the score is.

What are the common backtesting traps?

This article lists three IC backtest traps: overlapping windows make the effective sample far smaller than the number of days, choosing direction in-sample biases the hit rate structurally upward, and revised macro data make backtests look better than they really are. Other common problems include look-ahead bias, overfitting, and ignoring trading costs and taxes.

Which comes first: credit, sentiment, or valuation?

Credit sets the baseline, and HY OAS below 400 bps is a precondition for expansion; sentiment can shift the judgment by at most one level; valuation does not affect the current week's state, only long-term expected returns.

More articles

← All articles Open today's market dashboard →

Contents