Deleveraging as a US Stock Signal: A Regime-Dependent Strategy That

We backtested companies actively paying down debt on US stocks over 25 years. Pre-2013, the signal added real value. Post-2013, it became a drag of nearly 10 percentage points per year. Understanding the switch matters more than the headline CAGR.

Contents

1. Executive Summary
2. Method
3. The Signal: Why Deleveraging Should Work
4. Results
5. Two Regimes
6. When It Worked: 2000-2007
7. The Breaking Point: 2013 Onward
8. The 2020 Exception
9. Full Annual Returns
10. The Screen
11. Limitations
12. Global Results
13. Run It Yourself
14. Takeaway

Executive Summary

A D/E reduction screen with a quality filter produces a 3.78% CAGR against SPY's 8.01% over 2001-2025. But that headline conceals a clean structural break: the signal worked consistently during credit-tightening cycles and broke down once zero-interest-rate policy made debt cheap. This isn't a bad signal. It's a rate-regime signal.

Method

We screen NYSE, NASDAQ, and AMEX stocks quarterly for companies that actively reduced their debt-to-equity ratio by at least 10% year-over-year compared to the prior annual filing, while maintaining a return on equity above 8% and a market cap above $1B. From the qualifying universe, we hold the top 30 stocks ranked by the magnitude of D/E reduction. Equal weight, quarterly rebalance in January, April, July, and October.

Financial data uses a 45-day lag on current annual filings and a 410-day lookback window for the prior-year comparison, matching realistic signal availability at each rebalancing date.

Methodology

The Signal: Why Deleveraging Should Work

The theory isn't subtle. Companies reducing debt are doing several things at once: improving their interest coverage, reducing default risk, freeing future cash flows from debt service obligations, and signaling that management prioritizes financial conservatism over leverage-driven growth. The ROE filter adds a quality requirement, excluding distressed firms shedding debt because creditors are forcing them to.

In classical corporate finance, reducing leverage lowers the weighted average cost of capital and raises the present value of future earnings. That should show up in stock returns.

For a time, it did.

Results

The full 25-year result is a 3.78% CAGR. $10,000 invested grows to $26,000. SPY turns the same $10,000 into $73,000 over the same period.

The risk metrics make the result worse than the CAGR suggests. A down capture of 125.82% means the portfolio amplifies market losses rather than cushioning them. Beta of 1.14, max drawdown of -57.1% versus SPY's -45.5%. The strategy wasn't defensive. It was a concentrated small-universe bet that underperformed in both directions.

Key Performance Metrics

The portfolio was always invested. No quarters where the screen produced fewer than 30 stocks. That rules out empty-universe explanations for the underperformance.

Two Regimes

The 25-year headline conceals what is actually two different strategies operating in two different market environments.

Pre-2013 (2000-2012): Signal works

Thirteen years. Average excess return of approximately +2.7% per year. Win rate over SPY of 8 out of 13 years. This covers the dotcom crash, the slow recovery, and the 2008 crisis (with mixed but explainable results).

The mechanism: in periods of credit tightening or genuine financial stress, markets reward balance sheet conservatism. When debt is expensive or risky, companies reducing leverage are doing the right thing. The signal captures real quality differentiation.

Post-2013 (2013-2025): Signal breaks

Thirteen years. Average excess return of approximately -9.8% per year. Win rate of 2 out of 13 years. Consistent, deepening underperformance across bull markets, rate hikes, and everything in between.

The mechanism flips: when debt is cheap and monetary policy supports borrowing, companies reducing leverage look conservative in a negative sense. They're declining to use a cheap input. Capital-efficient growth companies, tech platforms, buyback-driven industrials, all of them used leverage strategically while this screen was selling it.

When It Worked: 2000-2007

The strongest stretch runs from 2000 through 2007, with one bad year in 2005 and a partial stumble in 2003-2004 before recovering.

Six out of eight years beat SPY. The dotcom bust punished leveraged tech companies selectively while older-economy businesses with declining debt loads looked genuinely different. In 2003-2004, the recovery was broad enough that the signal still captured upside. In 2006-2007, credit conditions were still tight enough that balance sheet quality mattered.

2005 is the outlier. The portfolio lost 3.1% in a year SPY gained 7.2%. Energy and material sector rotation away from quality that year hurt. But one bad year in an otherwise productive run.

The Breaking Point: 2013 Onward

The Federal Reserve held rates near zero from 2009 through 2015, then raised gently before cutting back in 2019. In that environment, debt became the cheapest input a company could get. Leverage-driven growth, share buybacks financed by cheap debt, and tech expansion funded by zero-cost capital all benefited directly.

Companies on this screen were doing the opposite. They were reducing leverage in a world where leverage was rewarded. The market priced them as conservative in a period where conservative was wrong.

2018 stands out beyond the ZIRP explanation: a year where SPY fell only 5.2% and the strategy dropped 25.1%. Sector concentration in industrials and energy, which saw the sharpest drawdowns that year, accounts for most of that gap. 2022's rate-hike environment should have been friendly to a deleveraging signal. It wasn't. The portfolio fell 37.1% against SPY's 19.0%. By then, the companies meeting the screen criteria were disproportionately value traps in cyclical sectors, not genuine quality.

The 2020 Exception

One year in the post-2013 run stands apart. In 2020, the portfolio returned +42.9% against SPY's +15.6%. A +27.3% spread. The only clean win since 2012.

The COVID shock created a brief environment similar to the dotcom era: markets differentiated sharply by balance sheet quality. Companies entering the crisis with declining debt loads were genuinely better positioned. They had more financial flexibility, lower refinancing risk, and weren't scrambling for liquidity. For one year, the signal reverted to its pre-2013 behavior.

That it was one year, not a sustained recovery, is instructive. Once the stimulus wave hit, the same dynamic resumed: cheap money, growth premia, leverage rewarded.

Full Annual Returns

The Screen

Simple version (D/E reduction only, no quality filter):

WITH current_fy AS (
    SELECT symbol, debtToEquityRatio AS de_current, date AS current_date
    FROM financial_ratios
    WHERE period = 'FY'
      AND debtToEquityRatio IS NOT NULL
      AND date >= CURRENT_DATE - INTERVAL 18 MONTH
    QUALIFY ROW_NUMBER() OVER (PARTITION BY symbol ORDER BY date DESC) = 1
),
prior_fy AS (
    SELECT symbol, debtToEquityRatio AS de_prior
    FROM financial_ratios
    WHERE period = 'FY'
      AND debtToEquityRatio IS NOT NULL
      AND date >= CURRENT_DATE - INTERVAL 30 MONTH
      AND date < CURRENT_DATE - INTERVAL 12 MONTH
    QUALIFY ROW_NUMBER() OVER (PARTITION BY symbol ORDER BY date DESC) = 1
)
SELECT
    c.symbol,
    ROUND(c.de_current, 2) AS de_current,
    ROUND(p.de_prior, 2) AS de_prior,
    ROUND((c.de_current - p.de_prior) / p.de_prior * 100, 1) AS de_change_pct
FROM current_fy c
JOIN prior_fy p ON c.symbol = p.symbol
WHERE p.de_prior > 0.1
  AND c.de_current >= 0
  AND (c.de_current - p.de_prior) / p.de_prior < -0.10
ORDER BY de_change_pct ASC
LIMIT 50

Advanced version (full backtest criteria with ROE, market cap, and exchange filters):

WITH current_fy AS (
    SELECT symbol, debtToEquityRatio AS de_current, date AS current_date
    FROM financial_ratios
    WHERE period = 'FY'
      AND debtToEquityRatio IS NOT NULL
      AND date >= CURRENT_DATE - INTERVAL 18 MONTH
    QUALIFY ROW_NUMBER() OVER (PARTITION BY symbol ORDER BY date DESC) = 1
),
prior_fy AS (
    SELECT symbol, debtToEquityRatio AS de_prior
    FROM financial_ratios
    WHERE period = 'FY'
      AND debtToEquityRatio IS NOT NULL
      AND date >= CURRENT_DATE - INTERVAL 30 MONTH
      AND date < CURRENT_DATE - INTERVAL 12 MONTH
    QUALIFY ROW_NUMBER() OVER (PARTITION BY symbol ORDER BY date DESC) = 1
),
km AS (
    SELECT symbol, returnOnEquityTTM AS roe, marketCap
    FROM key_metrics_ttm
)
SELECT
    c.symbol,
    p.companyName,
    p.exchange,
    p.sector,
    ROUND(c.de_current, 2) AS de_current,
    ROUND(pr.de_prior, 2) AS de_prior,
    ROUND((c.de_current - pr.de_prior) / pr.de_prior * 100, 1) AS de_change_pct,
    ROUND(k.roe * 100, 1) AS roe_pct,
    ROUND(k.marketCap / 1e9, 2) AS market_cap_bn
FROM current_fy c
JOIN prior_fy pr ON c.symbol = pr.symbol
JOIN km k ON c.symbol = k.symbol
JOIN profile p ON c.symbol = p.symbol
WHERE pr.de_prior > 0.1
  AND c.de_current >= 0
  AND (c.de_current - pr.de_prior) / pr.de_prior < -0.10
  AND k.roe > 0.08
  AND k.marketCap > 1000000000
  AND p.exchange IN ('NYSE', 'NASDAQ', 'AMEX')
ORDER BY (c.de_current - pr.de_prior) / pr.de_prior ASC
LIMIT 30

Run the advanced screen live: cetaresearch.com/data-explorer?q=jc1D3eHgRx

Limitations

Annual data lag. The D/E signal comes from annual filings with a 45-day lag. Companies can deteriorate substantially between filings. In an efficient market with dense analyst coverage, that lag means buying into a signal the market already priced.

ROE filter imprecision. ROE above 8% passes many companies where equity has been written down, temporarily inflating the return metric. It's a rough quality filter, not a precise one.

Sector concentration. Companies actively reducing debt cluster in industrials, energy, and materials. The portfolio carries implicit sector bets that explain part of the dotcom-era outperformance and the cyclical drawdowns in 2018 and 2022.

Small universe. 24.4 average stocks is a concentrated portfolio. Individual position blowups can move annual returns materially.

Survivorship and point-in-time fidelity. The backtest uses annual filing dates for the D/E comparison. In a real portfolio, you'd also need to track restatements and filing amendments that change the historical D/E ratios.

Global Results

The same strategy produces different outcomes in markets with lower information efficiency and thinner analyst coverage. India (BSE + NSE) generates a +4.39% annual excess with a down capture below 41%. Canada adds modest outperformance with down capture around 49%. Both markets are slower to price annual balance sheet data than the US, which means the 45-day lag is less of a handicap.

Details on the full multi-exchange results are in the comparison blog.

Part of a Series: Canada | Global | India

Run It Yourself

The full backtest code is on GitHub: github.com/ceta-research/backtests under risk-03-deleveraging/.

git clone https://github.com/ceta-research/backtests.git
cd backtests/risk-03-deleveraging
pip install -r requirements.txt
python backtest.py --preset us

Takeaway

The deleveraging signal works when credit conditions make leverage expensive or risky. It fails when monetary policy makes leverage cheap and capital markets reward companies for using it.

Pre-2013: eight of thirteen years beating SPY, average excess of about +2.7% per year. Post-2013: two of thirteen years beating SPY, average drag of about -9.8% per year. The 2020 COVID exception is the one brief reversion to the original pattern.

This isn't a failed signal. It's a regime-conditional signal that happens to be in an unfavorable regime for most of the past decade. Whether that changes depends more on interest rate cycles than on balance sheet analysis.

The 25-year US result is 3.78% CAGR against 8.01% for SPY. In the US, with US data, over the full period tested, buying an index fund beats this screen by more than 4 percentage points per year.

Data: Ceta Research data warehouse (FMP financial data). Backtest period 2000-2025, 103 quarters. Returns in USD. Transaction costs not included. Not investment advice.