Skip to content

Momentum & Trend

Version: 29 September 2026
Language: English
Purpose: Technical analysis / quantitative research / MarketLens design reference
Focus: Clearly distinguish Momentum, Trend, Cross-sectional Momentum, Time-series Momentum, and Trend Following


  1. Core Answer: Do Momentum and Trend Have to Be Combined?
  2. The Most Important Terminology
  3. The Foundation of All Calculations: Price, Return, Log Return, Excess Return
  4. Momentum: Major Academic Research and Formulas
  5. Trend: Major Academic Research and Formulas
  6. The Real Difference Between Momentum and Trend
  7. Summary Table of Common Indicator Formulas
  8. How to Normalize and Compare Different Signals
  9. Current MarketLens edges.py Approach
  10. MarketLens: Recommended Separation of Momentum and Trend
  11. Avoiding Double Counting
  12. Backtesting: How to Know Whether a Formula Really Has an Edge
  13. Simple Numerical Example
  14. Summary of Major Researchers and Studies
  15. Implementation Priority
  16. Important Limitations
  17. References

1. Core Answer: Do Momentum and Trend Have to Be Combined?

Section titled “1. Core Answer: Do Momentum and Trend Have to Be Combined?”

No.

Momentum and Trend are highly related, but they are not the same concept in financial research.

The simplest way to understand them is:

  • Trend asks: “Is price currently moving in a persistent upward or downward direction?”
  • Momentum asks: “Does past price performance tend to continue, and which stocks or assets have been strongest?”
  • Cross-sectional Momentum asks: “Relative to stocks B, C, and D, is stock A a winner or a loser?”
  • Time-series Momentum asks: “Was this asset’s own past return positive or negative, and does that direction tend to persist?”
  • Trend Following is a class of trading rules: go long in an uptrend and go short or stay out in a downtrend.

Therefore:

Time-Series Momentum can be viewed as one very important mathematical form of Trend Following; however, Cross-sectional Momentum is not the same as Trend.

CaseTrendMomentumInterpretation
AStrong uptrendStrongThe classic winner
BUptrendWeakeningPrice is still rising, but the rate of advance is slowing
CDowntrendShort-term strengtheningBear-market rally / early reversal
DSidewaysShort-term strongTemporary breakout, but long-term trend is not yet confirmed

There is an even more important example.

Suppose a stock fell 2% over the past 12 months while the market fell 20%.

  • Absolute trend / time-series momentum: it may still be negative.
  • Cross-sectional / relative momentum: it may already rank as one of the strongest winners in the market.

So the two should not be mechanically tied together.


English TermMeaningCore Question
MomentumMomentumDoes past strength tend to continue?
Price MomentumPrice MomentumWhat was the asset’s own past return?
Cross-sectional MomentumCross-sectional MomentumWhich stocks are stronger or weaker relative to others?
Time-series MomentumTime-series MomentumDoes an asset’s own positive / negative past return persist?
Relative StrengthRelative StrengthHow strong is the stock relative to other stocks or a benchmark?
TrendTrendIs price moving persistently in one direction?
Trend FollowingTrend FollowingTake positions in the direction of an established trend
Moving AverageMoving AverageSmooth price to estimate trend
BreakoutBreakoutPrice moves beyond a previous range or high
52-Week High52-Week HighHow close is the current price to its one-year high?
Residual MomentumResidual MomentumMomentum after removing market / size / value factor effects
Risk-Managed MomentumRisk-Managed MomentumAdjust momentum exposure according to volatility
Trend FilterTrend FilterTransform past prices / returns into a trend signal

3. The Foundation of All Calculations: Price, Return, Log Return, Excess Return

Section titled “3. The Foundation of All Calculations: Price, Return, Log Return, Excess Return”

If today’s price is (P_t) and the price (k) periods ago is (P_{t-k}):

Rt−k,t=PtPt−k−1R_{t-k,t} = \frac{P_t}{P_{t-k}} - 1

Example:

  • 63 trading days ago: 100
  • Today: 115

Then:

R63=115100−1=15%R_{63} = \frac{115}{100} - 1 = 15\%

This is similar to the concept behind MarketLens ROC63.


Log return is defined as:

rt=ln⁡(PtPt−1)r_t = \ln\left(\frac{P_t}{P_{t-1}}\right)

where:

  • (P_t): current-period price
  • (P_{t-1}): previous-period price
  • (\ln): natural logarithm

The most important logarithm rule is:

ln⁡(ab)=ln⁡(a)+ln⁡(b)\boxed{\ln(ab)=\ln(a)+\ln(b)}

In other words, multiplication becomes addition inside a logarithm.

But note:

ln⁡(a+b)≠ln⁡(a)+ln⁡(b)\ln(a+b) \neq \ln(a)+\ln(b)

The expression above is incorrect.

For example:

ln⁡(2×3)=ln⁡(6)\ln(2\times3)=\ln(6)

and:

ln⁡(2)+ln⁡(3)≈0.6931+1.0986≈1.7917\ln(2)+\ln(3) \approx 0.6931+1.0986 \approx 1.7917

while:

ln⁡(6)≈1.7918\ln(6)\approx1.7918

The difference is only rounding.

Formal Derivation of the Logarithm Addition Rule

Section titled “Formal Derivation of the Logarithm Addition Rule”

Suppose:

x=ln⁡(a)x=\ln(a)

Then:

ex=ae^x=a

Similarly:

y=ln⁡(b)y=\ln(b)

Then:

ey=be^y=b

Multiply the two:

ab=exeyab=e^xe^y

Using the exponential rule:

exey=ex+ye^xe^y=e^{x+y}

Therefore:

ab=ex+yab=e^{x+y}

Take the natural logarithm of both sides:

ln⁡(ab)=ln⁡(ex+y)\ln(ab)=\ln\left(e^{x+y}\right)

Because:

ln⁡(ez)=z\ln(e^z)=z

we obtain:

ln⁡(ab)=x+y\ln(ab)=x+y

Substituting:

x=ln⁡(a),y=ln⁡(b)x=\ln(a),\qquad y=\ln(b)

gives:

ln⁡(ab)=ln⁡(a)+ln⁡(b)\boxed{\ln(ab)=\ln(a)+\ln(b)}

Why Is This Useful for Multi-Period Returns?

Section titled “Why Is This Useful for Multi-Period Returns?”

Suppose the stock price is:

  • Day 0: (P_0=100)
  • Day 1: (P_1=110)
  • Day 2: (P_2=121)

The Day 1 log return is:

r1=ln⁡(110100)=ln⁡(1.1)r_1 = \ln\left(\frac{110}{100}\right) = \ln(1.1)

The Day 2 log return is:

r2=ln⁡(121110)=ln⁡(1.1)r_2 = \ln\left(\frac{121}{110}\right) = \ln(1.1)

The two daily log returns can be added directly:

r1+r2=ln⁡(110100)+ln⁡(121110)r_1+r_2 = \ln\left(\frac{110}{100}\right) + \ln\left(\frac{121}{110}\right)

Using:

ln⁡(a)+ln⁡(b)=ln⁡(ab)\ln(a)+\ln(b)=\ln(ab)

we obtain:

r1+r2=ln⁡(110100×121110)r_1+r_2 = \ln\left( \frac{110}{100} \times \frac{121}{110} \right)

The middle 110 cancels:

r1+r2=ln⁡(121100)r_1+r_2 = \ln\left(\frac{121}{100}\right)

Therefore:

r1+r2=ln⁡(P2P0)\boxed{ r_1+r_2 = \ln\left(\frac{P_2}{P_0}\right) }

Generalizing to (n) periods:

∑t=1nrt=ln⁡(P1P0)+ln⁡(P2P1)+⋯+ln⁡(PnPn−1)\sum_{t=1}^{n}r_t = \ln\left(\frac{P_1}{P_0}\right) + \ln\left(\frac{P_2}{P_1}\right) + \cdots + \ln\left(\frac{P_n}{P_{n-1}}\right)

Using the logarithm addition rule:

=ln⁡[P1P0P2P1⋯PnPn−1]= \ln\left[ \frac{P_1}{P_0} \frac{P_2}{P_1} \cdots \frac{P_n}{P_{n-1}} \right]

All intermediate prices cancel, leaving:

∑t=1nrt=ln⁡(PnP0)\boxed{ \sum_{t=1}^{n}r_t = \ln\left(\frac{P_n}{P_0}\right) }

So remember:

Multi-period Simple Returns must be compounded; multi-period Log Returns can be added directly.

Simple cumulative return:

1+Rtotal=(1+R1)(1+R2)⋯(1+Rn)1+R_{total} = (1+R_1)(1+R_2)\cdots(1+R_n)

Therefore:

Rtotal=∏j=1n(1+Rj)−1R_{total} = \prod_{j=1}^{n}(1+R_j)-1

For log returns:

rtotal=r1+r2+⋯+rn\boxed{ r_{total} = r_1+r_2+\cdots+r_n }

The conversion between the two is:

Rsimple=erlog−1\boxed{ R_{simple}=e^{r_{log}}-1 }

and:

rlog=ln⁡(1+Rsimple)\boxed{ r_{log}=\ln(1+R_{simple}) }

For example, if total log return is:

rlog=ln⁡(1.21)≈0.19062r_{log}=\ln(1.21)\approx0.19062

convert it back to simple return:

e0.19062−1≈0.21=21%e^{0.19062}-1 \approx 0.21 = 21\%

This is why log returns are especially convenient in volatility calculations, statistical analysis, and time-series models.


The most common meaning of excess return is:

Asset return minus the risk-free return.

Formula:

Rtexcess=Rt−Rf,t\boxed{ R_t^{excess}=R_t-R_{f,t} }

where:

  • (R_t): asset return
  • (R_{f,t}): risk-free return over the same period
  • (R_t^{excess}): how much more or less the asset earned relative to a risk-free investment

Example:

  • Stock return = (12%)
  • Risk-free return = (4%)

Then:

ExcessReturn=12%−4%=8%ExcessReturn = 12\%-4\% = 8\%

So the investment earned 8% more than the risk-free alternative.

Investing in stocks involves additional risk.

If a stock returns 6%, but a near-risk-free investment would have returned 5% over the same period, then the extra return received for taking the additional risk is only:

6%−5%=1%6\%-5\%=1\%

So excess return is closer to:

The return earned above a low-risk alternative after taking risk into account.

For example, an MSCI-style momentum measure can be written as:

MOM12=(PT−1PT−13−1)−RfMOM_{12} = \left( \frac{P_{T-1}}{P_{T-13}}-1 \right) - R_f

where:

PT−1PT−13−1\frac{P_{T-1}}{P_{T-13}}-1

is the stock’s price return over the period. Subtracting (R_f) gives the excess return relative to the risk-free rate.

Excess Return Can Also Be Defined Relative to Another Benchmark

Section titled “Excess Return Can Also Be Defined Relative to Another Benchmark”

In some contexts, excess return is used more broadly as:

Rstock−RbenchmarkR_{stock}-R_{benchmark}

For example:

Rstock−RS&P500R_{stock}-R_{S\&P500}

More precisely, this is a benchmark-relative return.

In asset-pricing models such as CAPM, excess return usually specifically means:

Ri−Rf\boxed{R_i-R_f}

while market excess return is:

Rm−Rf\boxed{R_m-R_f}

Time-series momentum research also often uses excess returns because researchers want to know whether an asset’s own past return exceeded the risk-free return and whether that direction subsequently persisted.


Annualized volatility is:

A way to convert return volatility observed over a shorter period, such as daily, weekly, or monthly, into an annual scale.

Volatility is usually the standard deviation of returns.

If daily return standard deviation is (\sigma_d):

σannual=σd252\boxed{ \sigma_{\text{annual}} = \sigma_d\sqrt{252} }

because the US equity market has approximately 252 trading days per year.

For example, if daily volatility is:

σd=2%=0.02\sigma_d=2\%=0.02

then:

σannual=0.02252\sigma_{annual} = 0.02\sqrt{252}

Since:

252≈15.87\sqrt{252}\approx15.87

we get:

σannual≈0.3174=31.74%\sigma_{annual} \approx 0.3174 = 31.74\%

So we can say:

The stock’s annualized volatility is approximately 31.7%.

Why Multiply by (\sqrt{252}) Instead of 252?

Section titled “Why Multiply by (\sqrt{252}) Instead of 252?”

Suppose daily return variance is:

Var(R)=σ2Var(R)=\sigma^2

If daily returns are approximately independent, then the variance of an (n)-day return is approximately:

Var(R1+R2+⋯+Rn)=nσ2Var(R_1+R_2+\cdots+R_n) = n\sigma^2

But volatility is standard deviation, which is the square root of variance:

σn=nσ2\sigma_n = \sqrt{n\sigma^2}

Therefore:

σn=σn\boxed{ \sigma_n = \sigma\sqrt{n} }

So converting daily volatility into annualized volatility gives:

σannual=σd252\sigma_{annual} = \sigma_d\sqrt{252}

not:

σd×252\sigma_d\times252

There are approximately 52 weeks in a year:

σannual=σweekly52\boxed{ \sigma_{annual} = \sigma_{weekly}\sqrt{52} }

There are 12 months in a year:

σannual=σmonthly12\boxed{ \sigma_{annual} = \sigma_{monthly}\sqrt{12} }

The general formula is:

Annualized Volatility=Period Volatility×Periods Per Year\boxed{ Annualized\ Volatility = Period\ Volatility \times \sqrt{Periods\ Per\ Year} }

Annualized Volatility Is Not an Annual Return Forecast

Section titled “Annualized Volatility Is Not an Annual Return Forecast”

If a stock has annualized volatility of 31.7%, that does not mean:

  • it will definitely rise 31.7% in one year; or
  • it will definitely fall 31.7% in one year.

It is a statistical measure of return dispersion / variability, not an expected-return forecast.

A common form of risk-adjusted momentum is:

RiskAdjustedMomentum=MomentumReturnAnnualizedVolatilityRiskAdjustedMomentum = \frac{MomentumReturn} {AnnualizedVolatility}

Suppose two stocks both rose 30%.

Stock A:

σA=15%\sigma_A=15\%

Stock B:

σB=40%\sigma_B=40\%

Then:

RAMA=30%15%=2.0RAM_A = \frac{30\%}{15\%} = 2.0

while:

RAMB=30%40%=0.75RAM_B = \frac{30\%}{40\%} = 0.75

Both stocks have the same raw return, but Stock A achieved the return with lower volatility, so its risk-adjusted momentum is higher.

Some studies use 252, 260, 261, or other trading-day constants.

For example:

σannual=σd261\sigma_{annual} = \sigma_d\sqrt{261}

The key point is not that one constant must always be used, but that:

The same annualization convention should be used consistently throughout the research, backtest, and production system.


4. Momentum: Major Academic Research and Formulas

Section titled “4. Momentum: Major Academic Research and Formulas”

There are far too many momentum papers to cover every study in one document. The sections below focus on the most representative and widely cited research streams that directly influence practical calculation methods.


4.1 Jegadeesh & Titman (1993): Cross-sectional Momentum

Section titled “4.1 Jegadeesh & Titman (1993): Cross-sectional Momentum”

Classic paper:

Narasimhan Jegadeesh & Sheridan Titman (1993), Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency.

The core idea is not to decide whether a single stock is “in an uptrend.” Instead:

  1. Consider many stocks.
  2. Calculate each stock’s return over the previous (J) months.
  3. Rank them.
  4. Buy the group with the highest past returns.
  5. Short the group with the lowest past returns.
  6. Hold for (K) months.

General form:

MOMi,t(J)=∏j=1J(1+Ri,t−j)−1MOM_{i,t}^{(J)} = \prod_{j=1}^{J}(1+R_{i,t-j})-1

Or approximately:

MOMi,t(J)=Pi,t−1Pi,t−J−1−1MOM_{i,t}^{(J)} = \frac{P_{i,t-1}}{P_{i,t-J-1}}-1

Then rank all stocks according to (MOM_{i,t}^{(J)}).

Let:

  • (W_t): the group of stocks with the highest past returns
  • (L_t): the group with the lowest past returns

Then the zero-cost momentum portfolio is:

RMOM,t+1=RW,t+1−RL,t+1R_{MOM,t+1} = R_{W,t+1} - R_{L,t+1}

Jegadeesh and Titman tested multiple (J/K) combinations, including 3-, 6-, 9-, and 12-month formation and holding periods.

This is relative / cross-sectional.

A stock does not need to be in a beautiful absolute uptrend. As long as it is stronger than other stocks, it may enter the winner group.


4.2 Carhart / Kenneth French: 12–2 Momentum Factor

Section titled “4.2 Carhart / Kenneth French: 12–2 Momentum Factor”

Carhart (1997) added a momentum factor to the Fama-French framework, producing the widely used four-factor model:

Ri−Rf=αi+βM(RM−Rf)+βSSMB+βHHML+βMOMMOM+ϵiR_i-R_f = \alpha_i + \beta_M(R_M-R_f) + \beta_S SMB + \beta_H HML + \beta_{MOM} MOM + \epsilon_i

The purpose of the Carhart Four-Factor Model is not to directly tell you which stock to buy. Instead, it attempts to explain:

How much of a stock, portfolio, or fund return comes from the overall market, company size, value/growth exposure, momentum exposure, and how much remains unexplained by these four factors.

Conceptually:

Carhart Four-Factor=Market+Size+Value+Momentum\boxed{ Carhart\ Four\text{-}Factor = Market + Size + Value + Momentum }

The original Fama-French Three-Factor Model included:

Market+SMB+HML\boxed{ Market + SMB + HML }

Carhart (1997) added:

MOM\boxed{ MOM }

creating the Four-Factor Model.

FactorFull NameCore Question
(R_M-R_f)Market Excess ReturnHow did the overall equity market perform relative to a risk-free asset?
SMBSmall Minus BigDid small-cap stocks outperform large-cap stocks?
HMLHigh Minus LowDid value stocks outperform growth stocks?
MOMMomentum / Winners Minus LosersDid past winners outperform past losers?

The market factor is:

RM−Rf\boxed{ R_M-R_f }

where:

  • (R_M): market return
  • (R_f): risk-free return

Example:

  • Market return = (10%)
  • Risk-free return = (4%)

Then:

RM−Rf=10%−4%=6%R_M-R_f = 10\%-4\% = 6\%

So:

Market Excess Return=6%\boxed{ Market\ Excess\ Return=6\% }

Meaning:

Investing in the broad equity market earned 6% more than a near-risk-free asset.

If a portfolio has:

βM=1.2\beta_M=1.2

you can loosely interpret this as:

The portfolio has greater market sensitivity than the market itself; for each 1 percentage point change in market excess return, portfolio excess return has, on average, about 1.2 percentage points of factor exposure.

SMB stands for:

Small Minus Big

It captures the Size Factor.

The core question is:

Did small-cap stocks outperform large-cap stocks during the period?

A highly simplified expression is:

SMB=Return(Small)−Return(Big)\boxed{ SMB = Return(Small) - Return(Big) }

Suppose:

  • Small-cap portfolio return = (12%)
  • Large-cap portfolio return = (8%)

Then:

SMB=12%−8%=4%SMB = 12\%-8\% = 4\%

So:

SMB=+4%\boxed{ SMB=+4\% }

Meaning:

Small-cap stocks earned about 4% more than large-cap stocks during that period.

If:

SMB<0SMB<0

then large-cap stocks were relatively stronger.

In the typical Fama-French 2 × 3 portfolio construction, stocks are first sorted by:

  1. Size: Small / Big
  2. Book-to-Market: Low / Medium / High

This creates six portfolios.

A simplified expression is:

SMB=13(S/L+S/M+S/H)−13(B/L+B/M+B/H)SMB = \frac{1}{3} (S/L+S/M+S/H) - \frac{1}{3} (B/L+B/M+B/H)

where:

  • (S) = Small
  • (B) = Big
  • (L) = Low book-to-market
  • (M) = Medium book-to-market
  • (H) = High book-to-market

The purpose is to isolate the size effect as much as possible from value/growth composition.

If a portfolio regression produces:

βS=+0.6\beta_S=+0.6

the portfolio has a relatively clear small-cap tilt.

If:

βS≈0\beta_S\approx0

its size-factor exposure is weak.

If:

βS<0\beta_S<0

the portfolio generally tilts toward large-cap stocks.

For example:

βS=0.5\beta_S=0.5

and:

SMB=4%SMB=4\%

then the model-implied size-factor contribution is approximately:

0.5×4%=2%0.5\times4\%=2\%

HML stands for:

High Minus Low

But here, High / Low does not refer to momentum high / low.

It refers to:

High Book-to-Market Minus Low Book-to-Market

Book-to-Market ratio:

B/M=Book EquityMarket EquityB/M = \frac{Book\ Equity}{Market\ Equity}

Generally:

  • High (B/M) → Value stocks
  • Low (B/M) → Growth stocks

So HML is a Value Factor.

Conceptually:

HML=Return(Value)−Return(Growth)\boxed{ HML = Return(Value) - Return(Growth) }

Company A:

  • Book equity = $50m
  • Market cap = $100m

Then:

B/MA=50100=0.50B/M_A = \frac{50}{100} = 0.50

Company B:

  • Book equity = $50m
  • Market cap = $500m

Then:

B/MB=50500=0.10B/M_B = \frac{50}{500} = 0.10

Company A has the higher book-to-market ratio, so it is relatively more “value-like.”

Company B has the lower book-to-market ratio, so it is relatively more “growth-like.”

Suppose:

  • High book-to-market / Value portfolio return = (11%)
  • Low book-to-market / Growth portfolio return = (7%)

Then:

HML=11%−7%=4%HML = 11\%-7\% = 4\%

So:

HML=+4%\boxed{ HML=+4\% }

Meaning:

Value stocks earned about 4% more than growth stocks during the period.

If:

HML<0HML<0

then growth stocks outperformed value stocks.

A typical simplified form is:

HML=12(S/H+B/H)−12(S/L+B/L)HML = \frac{1}{2}(S/H+B/H) - \frac{1}{2}(S/L+B/L)

where:

  • (H) = High book-to-market = Value
  • (L) = Low book-to-market = Growth

Therefore:

HML=Value−Growth\boxed{ HML = Value - Growth }

If:

βH>0\beta_H>0

the portfolio tilts toward value stocks.

If:

βH<0\beta_H<0

the portfolio tilts toward growth stocks.

For example:

βH=−0.4\beta_H=-0.4

and:

HML=+5%HML=+5\%

then the value-factor contribution is:

−0.4×5%=−2%-0.4\times5\% = -2\%

A negative (\beta_H) indicates a growth tilt, so the portfolio is negatively affected when value outperforms growth.

Very Important: HML High/Low Is Not Momentum High/Low

Section titled “Very Important: HML High/Low Is Not Momentum High/Low”

This is an easy point to confuse.

For HML:

High = High Book-to-Market = Value
Low = Low Book-to-Market = Growth

For Momentum:

High = High Prior Return = Winner
Low = Low Prior Return = Loser

Therefore:

HML High/Low≠Momentum High/Low\boxed{ HML\ High/Low \neq Momentum\ High/Low }

The words High and Low appear in both, but they refer to completely different sorting variables.

MOM captures:

The future return difference between past winners and past losers.

Simplified:

MOM=Return(Winners)−Return(Losers)\boxed{ MOM = Return(Winners) - Return(Losers) }

Example:

  • Winner portfolio = (13%)
  • Loser portfolio = (5%)

Then:

MOM=13%−5%=8%MOM = 13\%-5\% = 8\%

So:

MOM=+8%\boxed{ MOM=+8\% }

Meaning:

Past winners earned about 8% more than past losers during that period.

The 12–2 momentum ranking described later in this section is one important academic convention used to form these winner / loser portfolios.

The model is:

Ri−Rf=αi+βM(RM−Rf)+βSSMB+βHHML+βMOMMOM+ϵiR_i-R_f = \alpha_i + \beta_M(R_M-R_f) + \beta_SSMB + \beta_HHML + \beta_{MOM}MOM + \epsilon_i

After accounting for all four factor exposures, the remaining average intercept is:

α\boxed{\alpha}

Conceptually:

The average excess return not explained by the four known factors in the model.

If:

α>0\alpha>0

then average actual return was higher than what the four factor exposures explain.

But be careful:

Positive alpha should not automatically be interpreted as “manager skill.”

You also need to consider:

  • statistical significance
  • sample length
  • transaction costs
  • omitted factors
  • model specification
  • survivorship bias

and related issues.

Each (\beta) is the portfolio’s sensitivity / exposure to a factor.

For example:

βM=1.1\beta_M=1.1

means relatively high market exposure.

βS=0.3\beta_S=0.3

means a modest small-cap tilt.

βH=−0.2\beta_H=-0.2

means a modest growth tilt.

βMOM=0.5\beta_{MOM}=0.5

means positive momentum exposure.

A simple way to remember them:

βM = sensitivity to the market
βS = sensitivity to the size factor
βH = sensitivity to the value factor
βMOM = sensitivity to the momentum factor
ϵi\epsilon_i

is the regression residual / unexplained period-specific return.

Conceptually:

ActualReturn=ModelExplainedReturn+ResidualActualReturn = ModelExplainedReturn + Residual

Therefore:

ϵ\epsilon

represents the short-term deviation not explained by the model during a particular period.

It is not the same as (\alpha):

  • (\alpha): regression intercept / average component not explained by the factors
  • (\epsilon_t): period-specific residual shock

Suppose a portfolio has:

βM=1.1\beta_M=1.1 βS=0.3\beta_S=0.3 βH=−0.2\beta_H=-0.2 βMOM=0.5\beta_{MOM}=0.5

and the factor returns for the period are:

RM−Rf=6%R_M-R_f=6\% SMB=4%SMB=4\% HML=4%HML=4\% MOM=8%MOM=8\%

Assume:

α=1%\alpha=1\%

and, for simplicity:

ϵ=0\epsilon=0

Then portfolio excess return is explained by:

Ri−Rf=1%+1.1(6%)+0.3(4%)−0.2(4%)+0.5(8%)R_i-R_f = 1\% + 1.1(6\%) + 0.3(4\%) - 0.2(4\%) + 0.5(8\%)

Market contribution:

1.1×6%=6.6%1.1\times6\% = 6.6\%

Size contribution:

0.3×4%=1.2%0.3\times4\% = 1.2\%

Value contribution:

−0.2×4%=−0.8%-0.2\times4\% = -0.8\%

Momentum contribution:

0.5×8%=4.0%0.5\times8\% = 4.0\%

Plus alpha:

1.0%1.0\%

Therefore:

Ri−Rf=1.0%+6.6%+1.2%−0.8%+4.0%R_i-R_f = 1.0\% + 6.6\% + 1.2\% - 0.8\% + 4.0\% Ri−Rf=12.0%\boxed{ R_i-R_f=12.0\% }

If the risk-free return is:

Rf=4%R_f=4\%

then total portfolio return is:

Ri=12%+4%=16%R_i = 12\%+4\% = 16\%

So:

Total Portfolio Return=16%\boxed{ Total\ Portfolio\ Return=16\% }

The portfolio’s 16% total return is not all “alpha.”

The model decomposes it into:

Risk-free return 4.0%
Market exposure 6.6%
Size exposure 1.2%
Value exposure -0.8%
Momentum exposure 4.0%
Alpha 1.0%
--------------------------------
Total 16.0%

So although the portfolio appears to earn 16%, the model attributes most of that return to:

  • broad market exposure
  • small-cap exposure
  • growth/value exposure
  • momentum exposure

In this simplified example, the component unexplained by the four factors is only:

α=1%\alpha=1\%

Real research does not decompose just one month.

Researchers usually use many periods of historical data, such as monthly returns:

t=1,2,…,Tt=1,2,\ldots,T

and estimate a time-series regression:

Ri,t−Rf,t=αi+βM(RM,t−Rf,t)+βSSMBt+βHHMLt+βMOMMOMt+ϵi,tR_{i,t}-R_{f,t} = \alpha_i + \beta_M(R_{M,t}-R_{f,t}) + \beta_SSMB_t + \beta_HHML_t + \beta_{MOM}MOM_t + \epsilon_{i,t}

The regression estimates:

α^,β^M,β^S,β^H,β^MOM\hat\alpha,\quad \hat\beta_M,\quad \hat\beta_S,\quad \hat\beta_H,\quad \hat\beta_{MOM}

So the one-period example above is only an illustrative example of factor contributions. In practice, beta and alpha are estimated from multi-period regressions.

How Is the Four-Factor Model Useful for MarketLens?

Section titled “How Is the Four-Factor Model Useful for MarketLens?”

MarketLens does not necessarily need to use the Carhart model directly for everyday stock ranking.

It is more useful for:

For example, determine whether a MarketLens portfolio is effectively just:

High market beta
+ small-cap tilt
+ growth tilt
+ momentum tilt

rather than having an independent edge.

Decompose portfolio return:

Return=Market+Size+Value+Momentum+Alpha+ResidualReturn = Market + Size + Value + Momentum + Alpha + Residual

The Residual Momentum section later uses a similar idea:

Remove returns explained by market / size / value factors first, then study the remaining firm-specific momentum.

If a MarketLens strategy already has strong momentum-factor exposure, adding many similar momentum signals may simply repeat the same exposure.

Remember the Carhart Four-Factor Model as:

Market+Size+Value+Momentum\boxed{ Market + Size + Value + Momentum }

where:

Market = Rm − Rf
SMB = Small − Big
HML = Value − Growth
MOM = Winners − Losers

and:

Alpha = average return component not explained by the factors
Beta = exposure / sensitivity to each factor
Epsilon = period-specific residual not explained by the model

Most importantly:

HML High/Low=Book-to-Market High/Low\boxed{ HML\ High/Low = Book\text{-}to\text{-}Market\ High/Low }

while:

MOM High/Low=Prior Return High/Low\boxed{ MOM\ High/Low = Prior\ Return\ High/Low }

Today, a common academic momentum definition is prior 12-to-2 month return, which usually skips the most recent month to reduce short-term reversal / microstructure effects.

Suppose we are about to form a portfolio in month (t).

12–2 momentum does not use the entire most recent 12 months. Instead, it uses:

t−12tot−2t-12 \quad\text{to}\quad t-2

and deliberately skips:

t−1t-1

which is the most recent completed month.

Timeline:

Past Portfolio month
t-12 t-11 ... t-3 t-2 t-1 t
|---------------------------|
USED FOR MOMENTUM SKIP INVEST

Therefore:

12–2 Momentum looks at roughly 11 months of cumulative return and skips the most recent month.

The Kenneth French Data Library often writes this as:

prior (2–12) return

which refers to the same general 12–2 lookback convention.

If the current month is (t):

MOM12−2=∏j=212(1+Rt−j)−1MOM_{12-2} = \prod_{j=2}^{12}(1+R_{t-j})-1

Because these are simple returns, multi-period returns cannot simply be added; they must be compounded:

(1+R1)(1+R2)⋯(1+Rn)−1(1+R_1)(1+R_2)\cdots(1+R_n)-1

For example, if several monthly returns are:

+5%,+3%,−2%,+4%+5\%,\quad +3\%,\quad -2\%,\quad +4\%

you should not simply write:

5+3−2+4=10%5+3-2+4=10\%

Instead:

(1.05)(1.03)(0.98)(1.04)−1(1.05)(1.03)(0.98)(1.04)-1

If monthly simple return is:

Rs=PsPs−1−1R_s = \frac{P_s}{P_{s-1}}-1

then:

1+Rs=PsPs−11+R_s = \frac{P_s}{P_{s-1}}

Therefore:

MOM12−2=∏j=212(1+Rt−j)−1MOM_{12-2} = \prod_{j=2}^{12}(1+R_{t-j})-1

can be expanded as:

=(Pt−2Pt−3)(Pt−3Pt−4)⋯(Pt−12Pt−13)−1= \left( \frac{P_{t-2}}{P_{t-3}} \right) \left( \frac{P_{t-3}}{P_{t-4}} \right) \cdots \left( \frac{P_{t-12}}{P_{t-13}} \right) -1

All intermediate prices cancel:

=Pt−2Pt−13−1= \frac{P_{t-2}}{P_{t-13}}-1

So:

MOM12−2=Pt−2Pt−13−1\boxed{ MOM_{12-2} = \frac{P_{t-2}}{P_{t-13}}-1 }

This price form makes the structure clear:

  • endpoint is (t-2)
  • the most recent month (t-1) is skipped
  • starting price is (t-13), because we need the returns from (t-12) through (t-2)

Suppose you form a portfolio in October 2026.

The most recent completed month is:

September 2026

12–2 momentum uses:

Oct 2025 → Aug 2026 Sep 2026 Oct 2026
|--------------------| |--------| |-------|
Momentum window Skip month Hold

That means:

  • approximately October 2025 through August 2026
  • September 2026 is excluded from the momentum ranking
  • the portfolio is formed / held in October 2026

Suppose:

PSep 2025 end=100P_{\text{Sep 2025 end}}=100

and:

PAug 2026 end=140P_{\text{Aug 2026 end}}=140

Then:

MOM12−2=140100−1=40%MOM_{12-2} = \frac{140}{100}-1 = 40\%

So the 12–2 momentum signal is:

40%\boxed{40\%}

Even if the price rose from 140 to 170 in September 2026, that most recent monthly gain would not be included in this 12–2 signal.

Why Deliberately Skip the Most Recent Month?

Section titled “Why Deliberately Skip the Most Recent Month?”

This is one of the most important design choices in 12–2 momentum.

Medium-term return continuation and very-short-term price behavior are not exactly the same.

The most recent month may be affected by:

  • short-term reversal
  • temporary overreaction
  • bid-ask / microstructure effects
  • news shock
  • very short-term mean reversion

Academic momentum therefore often separates out the most recent month so it does not dominate the medium-term momentum signal.

For example:

Longer-term path:
100 → 105 → 112 → 120 → 130 → 140
Most recent month:
140 → 125

If you use a simple trailing return, the most recent month’s sharp decline reduces momentum significantly.

12–2 momentum asks:

Before this most recent short-term fluctuation, was the stock a strong medium-term winner?

That is not exactly the same as asking:

How much did the stock rise over the most recent 12 months in total?

12–2 Momentum vs Ordinary Trailing 12-Month Return

Section titled “12–2 Momentum vs Ordinary Trailing 12-Month Return”

Ordinary trailing 12-month return may be written as:

R12M=PtPt−12−1R_{12M} = \frac{P_t}{P_{t-12}}-1

It includes the most recent month.

But 12–2 momentum is:

MOM12−2=Pt−2Pt−13−1MOM_{12-2} = \frac{P_{t-2}}{P_{t-13}}-1

which deliberately excludes the most recent month.

For example:

100 → 150 → 120

Suppose 150 was the price one month ago and 120 is the current price.

Ordinary trailing return may be only:

120100−1=20%\frac{120}{100}-1 = 20\%

while skipped-month momentum may still be close to:

150100−1=50%\frac{150}{100}-1 = 50\%

The two measures capture different information.

Important: A Stock’s 12–2 Return Is Not Yet a Momentum Factor

Section titled “Important: A Stock’s 12–2 Return Is Not Yet a Momentum Factor”

Suppose Oracle has:

MOM12−2=35%MOM_{12-2}=35\%

This only means:

Oracle’s own 12–2 momentum signal is 35%.

It is not yet the academic momentum factor.

To construct Cross-sectional Momentum, you need to calculate the same signal across the entire stock universe.

For example:

Stock12–2 Return
A+75%
B+48%
ORCL+35%
D+10%
E-8%
F-40%

Then rank all stocks.

You can further convert the ranking into:

MomentumPercentilei=PercentileRank(MOM12−2,i)MomentumPercentile_i = PercentileRank(MOM_{12-2,i})

For example, if ORCL is stronger than 88% of the universe:

MomentumPercentileORCL=88\boxed{ MomentumPercentile_{ORCL}=88 }

So for MarketLens, these two numbers should be stored separately:

MOM12_2 = 35%
MomentumPercentile = 88

How Does Kenneth French Turn It Into a True MOM Factor?

Section titled “How Does Kenneth French Turn It Into a True MOM Factor?”

The Kenneth French momentum factor does not simply rank stocks by momentum; it also considers size.

Stocks are first split into:

Small
Big

and then, based on prior (2–12) return:

Low
Neutral
High

This forms six portfolios:

SL = Small Low
SN = Small Neutral
SH = Small High
BL = Big Low
BN = Big Neutral
BH = Big High

where:

  • High = winners
  • Low = losers

A simplified momentum-factor expression is:

MOM=12(SH+BH)−12(SL+BL)\boxed{ MOM = \frac{1}{2}(SH+BH) - \frac{1}{2}(SL+BL) }

That is:

MomentumFactor=Return(Winners)−Return(Losers)\boxed{ MomentumFactor = Return(Winners) - Return(Losers) }

Some literature uses:

WML = Winners Minus Losers

to describe the same basic concept.

12–2 Signal and MOM Factor Return Occur at Different Times

Section titled “12–2 Signal and MOM Factor Return Occur at Different Times”

This is crucial in backtesting.

12–2 return is the formation signal.

It uses past data to determine:

Who is a winner?
Who is a loser?

The actual MOM factor return is:

The subsequent holding-period return difference between the winner and loser portfolios.

Therefore:

Past 12–2 Return
↓
Cross-sectional Ranking
↓
Winner / Loser Groups
↓
Hold Portfolios
↓
Future Winner Return − Future Loser Return
↓
Momentum Factor Return

So:

12 ⁣− ⁣2 Return≠MOM Factor Return\boxed{ 12\!-\!2\ Return \neq MOM\ Factor\ Return }

Suppose next month:

Small Winners:

SH=+4%SH=+4\%

Big Winners:

BH=+2%BH=+2\%

Small Losers:

SL=−1%SL=-1\%

Big Losers:

BL=0%BL=0\%

Winner portfolio:

RW=4%+2%2=3%R_W = \frac{4\%+2\%}{2} = 3\%

Loser portfolio:

RL=−1%+0%2=−0.5%R_L = \frac{-1\%+0\%}{2} = -0.5\%

Momentum factor:

MOM=3%−(−0.5%)MOM = 3\%-(-0.5\%)

So:

MOM=3.5%\boxed{ MOM=3.5\% }

Note:

This 3.5% is not the past 12–2 return. It is the next-period realized winner-minus-loser return after portfolios were formed using the past 12–2 ranking.

MarketLens should avoid compressing all momentum information into one number immediately.

Instead, retain:

Ticker: ORCL
MOM12_2 = 31.4%
ROC63 = 8.2%
MOM12_2 Percentile = 91

They represent:

Academic-style medium-term momentum:

Pt−2Pt−13−1\frac{P_{t-2}}{P_{t-13}}-1

More recent, roughly three-month momentum:

ROC63=PtPt−63−1ROC63 = \frac{P_t}{P_{t-63}}-1

ORCL’s 12–2 momentum ranking relative to the entire universe:

PercentileRank(MOM12−2)PercentileRank(MOM_{12-2})

For example:

MOM12_2 = 31.4%
ROC63 = 8.2%
MOM12_2 Percentile = 91

can be interpreted as:

ORCL rose about 31.4% over the academic 12–2 window; its more recent 63-session momentum is +8.2%; and its 12–2 momentum is stronger than roughly 91% of the stock universe.

This is much clearer than immediately adding all the numbers together and is better suited to testing each signal independently.

Remember:

12 ⁣− ⁣2 Return→Cross-Sectional Ranking→Winner/Loser Portfolios→Future MOM Factor Return\boxed{ 12\!-\!2\ Return \rightarrow Cross\text{-}Sectional\ Ranking \rightarrow Winner/Loser\ Portfolios \rightarrow Future\ MOM\ Factor\ Return }

where:

  • 12–2 Return = signal
  • Percentile / Rank = relative strength in the universe
  • Winner / Loser grouping = portfolio formation
  • Future WML return = academic momentum factor realization

Do not mix these four stages into one number.

Simplified:

MOM=12(SmallHigh+BigHigh)−12(SmallLow+BigLow)MOM = \frac{1}{2}(SmallHigh+BigHigh) - \frac{1}{2}(SmallLow+BigLow)

This definition is very useful as an academic baseline for MarketLens.


4.3 Moskowitz & Grinblatt (1999): Industry Momentum

Section titled “4.3 Moskowitz & Grinblatt (1999): Industry Momentum”

Paper:

Tobias Moskowitz & Mark Grinblatt (1999), Do Industries Explain Momentum?

Core idea:

A portion of stock-level momentum may come from industry momentum.

Suppose industry (g) contains (N_g) stocks:

Rg,t=∑i∈gwi,tRi,tR_{g,t} = \sum_{i \in g} w_{i,t}R_{i,t}

where (w_i) can be equal weight or value weight.

Then calculate the industry’s formation-period return:

MOMg,t=∏j=1J(1+Rg,t−j)−1MOM_{g,t} = \prod_{j=1}^{J}(1+R_{g,t-j})-1

Then:

  • Buy winning industries
  • Short losing industries

If Oracle has strong stock-level momentum and the software / technology industry also has strong momentum, the signal is more complete than looking at the stock alone.


4.4 Lee & Swaminathan (2000): Momentum + Trading Volume

Section titled “4.4 Lee & Swaminathan (2000): Momentum + Trading Volume”

Paper:

Charles Lee & Bhaskaran Swaminathan (2000), Price Momentum and Trading Volume.

They show that past trading volume contains information about the magnitude and persistence of momentum.

A common turnover definition is:

Turnovert=VolumetSharesOutstandingtTurnover_t = \frac{Volume_t}{SharesOutstanding_t}

Multi-period average:

Turnover‾N=1N∑j=0N−1Turnovert−j\overline{Turnover}_{N} = \frac{1}{N} \sum_{j=0}^{N-1}Turnover_{t-j}

Research may double-sort stocks by:

  • past return momentum
  • past turnover / volume

For example:

VolumeRatio=VolumetSMA50(Volume)tVolumeRatio = \frac{Volume_t}{SMA_{50}(Volume)_t}

But note:

This VolumeRatio is a practical simplification, not the only standard formula used in the Lee-Swaminathan paper.


4.5 George & Hwang (2004): 52-Week High Momentum

Section titled “4.5 George & Hwang (2004): 52-Week High Momentum”

Paper:

Thomas George & Chuan-Yang Hwang (2004), The 52-Week High and Momentum Investing.

Core measure:

High52Ratioi,t=Pi,tmax⁡(Pi,t−251,…,Pi,t)High52Ratio_{i,t} = \frac{P_{i,t}} {\max(P_{i,t-251},\ldots,P_{i,t})}

Usually:

0<High52Ratio≤10 < High52Ratio \le 1

The closer the ratio is to 1:

The closer the current price is to its 52-week high.

Example:

  • Current price: 95
  • 52-week high: 100
High52Ratio=0.95High52Ratio=0.95

George and Hwang rank stocks using this ratio and examine whether proximity to the 52-week high helps explain future returns.

Relationship to the Current MarketLens Approach

Section titled “Relationship to the Current MarketLens Approach”

Your current logic is:

close >= 0.85 * high_252

which is equivalent to:

High52Ratio≥0.85High52Ratio \ge 0.85

This is directionally consistent with the 52-week-high literature, but the 85% threshold is a MarketLens-specific cutoff, not a standard academic cutoff from George & Hwang.


4.6 Blitz, Huij & Martens (2011): Residual Momentum

Section titled “4.6 Blitz, Huij & Martens (2011): Residual Momentum”

Paper:

David Blitz, Joop Huij & Martin Martens (2011), Residual Momentum.

First, use Fama-French factors to explain stock returns:

Ri,t−Rf,t=αi+βi,MMKTt+βi,SSMBt+βi,HHMLt+ϵi,tR_{i,t}-R_{f,t} = \alpha_i + \beta_{i,M}MKT_t + \beta_{i,S}SMB_t + \beta_{i,H}HML_t + \epsilon_{i,t}

where:

ϵi,t=Residual Return\epsilon_{i,t} = \text{Residual Return}

Meaning:

Remove the portion of returns explained by market, size, and value exposure, leaving the firm-specific return.

Research typically uses rolling regressions, for example over 36 months.

Then calculate residual momentum over the previous 12 months, often skipping the most recent month:

RMOMi,t=∑j=212ϵi,t−jσ(ϵi,t−12:t−2)RMOM_{i,t} = \frac{ \sum_{j=2}^{12}\epsilon_{i,t-j} }{ \sigma(\epsilon_{i,t-12:t-2}) }

Some studies express this as average residual return divided by residual volatility.

Residual momentum attempts to isolate momentum that is less contaminated by common factor exposures.

Conceptually:

Raw Stock Return
|
Remove Market / Size / Value
|
Residual Return
|
Residual Momentum

This may be useful for MarketLens if you later want to distinguish between:

  • “The stock is strong because its entire factor group is strong.”
  • “The stock itself has unusually persistent firm-specific strength.”

4.7 Novy-Marx (2012): Intermediate-Horizon Momentum

Section titled “4.7 Novy-Marx (2012): Intermediate-Horizon Momentum”

Paper:

Robert Novy-Marx (2012), Is Momentum Really Momentum?

The paper shows that the intermediate portion of past return history contains particularly strong predictive information.

A key finding is that returns roughly 12 to 7 months ago can carry substantial predictive power.

A simplified intermediate-horizon signal may be written as:

MOM12−7=Pt−7Pt−13−1MOM_{12-7} = \frac{P_{t-7}}{P_{t-13}}-1

while a more recent component could be:

MOM6−1=Pt−1Pt−7−1MOM_{6-1} = \frac{P_{t-1}}{P_{t-7}}-1

The surprising empirical result is that the intermediate component can sometimes be more important than the most recent component.

This suggests MarketLens should not assume:

“The more recent the return, the more useful it must be.”

Your existing Weighted RS Raw assigns the largest weight to the recent 63-day return.

That may be a sensible trading heuristic, but it is not the same as the academic intermediate-horizon momentum result.

This makes it worth testing separate components such as:

12–7 month return
6–2 month return
3 month return

rather than forcing a fixed weighting from the start.


4.8 Moskowitz, Ooi & Pedersen (2012): Time-Series Momentum

Section titled “4.8 Moskowitz, Ooi & Pedersen (2012): Time-Series Momentum”

Paper:

Tobias Moskowitz, Yao Hua Ooi & Lasse Heje Pedersen (2012), Time Series Momentum.

Time-Series Momentum asks:

If an asset’s own past excess return was positive or negative, does that direction tend to continue?

A simplified signal is:

Si,t=sign(Ri,t−h:texcess)S_{i,t} = sign(R^{excess}_{i,t-h:t})

where (h) might be 12 months.

Then the position is conceptually:

Positioni,t=Si,tPosition_{i,t} = S_{i,t}

so:

  • positive past return → long
  • negative past return → short

This differs from cross-sectional momentum because no ranking against other assets is required.

The strategy commonly scales position size inversely with volatility.

A simplified form:

wi,t=Si,tσ∗σ^i,tw_{i,t} = S_{i,t} \frac{\sigma^*}{\hat\sigma_{i,t}}

where:

  • (\sigma^*): target volatility
  • (\hat\sigma_{i,t}): estimated asset volatility

This prevents a highly volatile asset from dominating portfolio risk.

One common volatility estimate is an exponentially weighted moving average:

σt2=(1−λ)rt−12+λσt−12\sigma_t^2 = (1-\lambda)r_{t-1}^2 + \lambda\sigma_{t-1}^2

or equivalently, as a weighted sum of historical squared returns.

Time-Series Momentum is very close to trend following because it asks whether an asset’s own return direction persists.

Unlike cross-sectional momentum:

TSMOM:
Compare the asset with itself over time.
Cross-sectional MOM:
Compare the asset with other assets.

4.9 Asness, Moskowitz & Pedersen (2013): Value and Momentum Everywhere

Section titled “4.9 Asness, Moskowitz & Pedersen (2013): Value and Momentum Everywhere”

Paper:

Cliff Asness, Tobias Moskowitz & Lasse Heje Pedersen (2013), Value and Momentum Everywhere.

The study documents momentum across multiple asset classes and markets.

The important idea is:

Momentum is not only a US-stock anomaly.

Evidence appears across:

  • equities
  • equity indices
  • currencies
  • government bonds
  • commodities

This matters for MarketLens conceptually because it suggests momentum may reflect a broad behavioral / risk phenomenon rather than a single-market technical trick.

The study also examines the interaction between value and momentum.

Value and momentum can be negatively correlated in some settings, which means they can provide diversification when combined.


4.10 Fama & French (2012): International Momentum

Section titled “4.10 Fama & French (2012): International Momentum”

Paper:

Eugene Fama & Kenneth French (2012), Size, Value, and Momentum in International Stock Returns.

They investigate size, value, and momentum across international markets.

The important implication is:

Momentum effects are not confined to the US market, but their strength can vary by region and market structure.

For MarketLens, this is especially relevant if you scan:

  • US equities
  • Hong Kong equities
  • other international universes

The model should not assume that a threshold calibrated on the US market will work identically in Hong Kong.


4.11 Barroso & Santa-Clara (2015): Risk-Managed Momentum

Section titled “4.11 Barroso & Santa-Clara (2015): Risk-Managed Momentum”

Paper:

Pedro Barroso & Pedro Santa-Clara (2015), Momentum Has Its Moments.

A major problem with momentum strategies is crash risk.

The paper proposes scaling momentum exposure according to estimated momentum volatility.

A simplified form is:

wt=σ∗σ^MOM,tw_t = \frac{\sigma^*}{\hat\sigma_{MOM,t}}

and:

RMOM,t+1scaled=wtRMOM,t+1R^{scaled}_{MOM,t+1} = w_t R_{MOM,t+1}

When momentum volatility rises sharply:

σ^MOM,t↑\hat\sigma_{MOM,t}\uparrow

then:

wt↓w_t\downarrow

So the strategy reduces exposure in high-risk periods.

This is important because a strong raw momentum score does not necessarily imply that full-size exposure is appropriate.


4.12 Daniel & Moskowitz (2016): Momentum Crashes

Section titled “4.12 Daniel & Moskowitz (2016): Momentum Crashes”

Paper:

Kent Daniel & Tobias Moskowitz (2016), Momentum Crashes.

Momentum strategies can experience severe losses during specific market regimes, especially when:

  • the market has recently fallen substantially;
  • volatility is elevated;
  • losers rebound sharply during market recovery.

This often creates a violent reversal:

Previous Losers
↓
Market Rebound
↓
Losers rally sharply
↓
Winner-minus-Loser Momentum crashes

This matters for MarketLens because:

A momentum signal should not be treated as independent from market regime and risk conditions.

This supports keeping a separate:

Market Regime / Risk

layer rather than simply giving more points to every momentum-like signal.


5. Trend: Major Academic Research and Formulas

Section titled “5. Trend: Major Academic Research and Formulas”

The (N)-period Simple Moving Average is:

SMAN(t)=1N∑j=0N−1Pt−jSMA_N(t) = \frac{1}{N} \sum_{j=0}^{N-1}P_{t-j}

For example:

SMA50=Pt+Pt−1+⋯+Pt−4950SMA_{50} = \frac{P_t+P_{t-1}+\cdots+P_{t-49}}{50}

SMA smooths price and estimates the underlying direction.

A simple trend condition is:

Pt>SMAN(t)P_t>SMA_N(t)

This says current price is above its recent average.

It does not by itself measure relative performance against other stocks.


EMA gives more weight to recent prices.

Formula:

EMAt=αPt+(1−α)EMAt−1EMA_t = \alpha P_t + (1-\alpha)EMA_{t-1}

where:

α=2N+1\alpha = \frac{2}{N+1}

For an EMA with (N=10):

α=211\alpha=\frac{2}{11}

Compared with SMA, EMA responds more quickly to recent price changes.


A common trend rule:

Signalt=sign(MAshort,t−MAlong,t)Signal_t = sign(MA_{short,t}-MA_{long,t})

For example:

BuyifSMA50>SMA200Buy \quad\text{if}\quad SMA_{50}>SMA_{200}

and:

Sell/ExitifSMA50<SMA200Sell/Exit \quad\text{if}\quad SMA_{50}<SMA_{200}

A stricter condition is:

Pt>SMA50>SMA150>SMA200P_t>SMA_{50}>SMA_{150}>SMA_{200}

This is a trend structure / alignment, not a classic cross-sectional momentum formula.


5.4 Brock, Lakonishok & LeBaron (1992): Moving Average / Trading Range Break

Section titled “5.4 Brock, Lakonishok & LeBaron (1992): Moving Average / Trading Range Break”

Paper:

William Brock, Josef Lakonishok & Blake LeBaron (1992), Simple Technical Trading Rules and the Stochastic Properties of Stock Returns.

They tested two classic types of technical rule:

  1. Moving-average rules
  2. Trading-range breakout rules

A general moving-average rule is:

BuyifMAshort>MAlongBuy \quad\text{if}\quad MA_{short}>MA_{long} SellifMAshort<MAlongSell \quad\text{if}\quad MA_{short}<MA_{long}

The paper examined several MA combinations, including:

  • 1–50
  • 1–150
  • 5–150
  • 1–200
  • 2–200

It also tested banded versions to reduce whipsaw.

If a band (b) is used:

BuyifMAS>(1+b)MALBuy \quad\text{if}\quad MA_S>(1+b)MA_L SellifMAS<(1−b)MALSell \quad\text{if}\quad MA_S<(1-b)MA_L

5.5 Time-Series Momentum as Trend Following

Section titled “5.5 Time-Series Momentum as Trend Following”

Moskowitz-Ooi-Pedersen TSMOM is essentially:

Past rise → long
Past decline → short

Therefore:

TrendSignal12m=sign(Rt−12,t)TrendSignal_{12m} = sign(R_{t-12,t})

This is a return-based trend filter.

A moving average is a price-level-based trend filter.

They are not identical at the surface level, but they can be related within a more general linear-filter framework.


5.6 Hurst, Ooi & Pedersen (2017): Multi-Horizon Trend

Section titled “5.6 Hurst, Ooi & Pedersen (2017): Multi-Horizon Trend”

Paper:

Brian Hurst, Yao Hua Ooi & Lasse Heje Pedersen (2017), A Century of Evidence on Trend-Following Investing.

They use:

  • 1-month trend
  • 3-month trend
  • 12-month trend

and combine them with equal weights.

For each horizon:

Sh(t)=sign(Rt−h,texcess)S_h(t) = sign(R^{excess}_{t-h,t})

where:

h∈{1,3,12} monthsh \in \{1,3,12\}\text{ months}

A simplified combined signal is:

Scombined=13(S1+S3+S12)S_{combined} = \frac{1}{3} (S_1+S_3+S_{12})

The actual strategy also uses volatility targeting.

Do not rely on only one horizon.

A stock may be:

  • 1m positive
  • 3m positive
  • 12m negative

This may indicate a short-term rebound while the long-term trend has not yet recovered.


5.7 Levine & Pedersen (2016): A Unified Framework for Trend Filters

Section titled “5.7 Levine & Pedersen (2016): A Unified Framework for Trend Filters”

Paper:

Ari Levine & Lasse Heje Pedersen (2016), Which Trend Is Your Friend?

They show that:

  • Time-series momentum
  • Moving-average crossover
  • other linear trend filters

can be connected within a generalized framework.

A generic linear trend signal is:

St=∑k=1Kwkrt−kS_t = \sum_{k=1}^{K}w_k r_{t-k}

where:

  • (r_{t-k}): past return
  • (w_k): weight on each lag

Many trend methods differ mainly in:

The weight pattern applied to returns from different points in the past.

The simplest TSMOM can be represented approximately as:

wk=1Kw_k=\frac{1}{K}

meaning returns over the lookback period are aggregated with roughly equal directional contribution.

A moving-average crossover implicitly applies a different pattern of positive / negative lag weights.

This framework is especially useful for MarketLens because it allows:

  • ROC
  • MA crossover
  • EMA
  • multi-horizon trend

to be viewed as different filters rather than completely independent “edges.”


Paper:

Yufeng Han, Guofu Zhou & Yingzi Zhu (2016), A Trend Factor: Any Economic Gains from Using Information over Investment Horizons?

They attempt to capture:

  • short-term trend
  • intermediate-term trend
  • long-term trend

The research uses multiple moving-average price signals with horizons ranging from a few days to roughly 1,000 days. Cross-sectional regression is then used to combine these signals into forecasted returns, which are used to build a long-short trend factor.

Abstractly:

R^i,t+1=at+∑h∈Hbh,tXi,t(h)\hat R_{i,t+1} = a_t + \sum_{h \in H} b_{h,t}X_{i,t}^{(h)}

where:

  • (X^{(h)}): moving-average / trend signal at horizon (h)
  • (b_h): cross-sectional regression coefficient
  • (\hat R): forecasted return

Then:

  • Long highest forecast group
  • Short lowest forecast group

A Trend Score does not have to be based on only SMA200. It can be a multi-horizon ensemble.


This is a common quantitative technical-analysis method.

For the most recent (N) days of log price:

yt=ln⁡(Pt)y_t=\ln(P_t)

fit:

yt=a+bt+ϵty_t=a+bt+\epsilon_t

OLS slope:

b=∑t=1N(t−tˉ)(yt−yˉ)∑t=1N(t−tˉ)2b = \frac{ \sum_{t=1}^{N}(t-\bar t)(y_t-\bar y) }{ \sum_{t=1}^{N}(t-\bar t)^2 }
  • (b>0): uptrend
  • (b<0): downtrend
  • larger (|b|): steeper direction

The slope can be annualized:

TrendRate=e252b−1TrendRate = e^{252b}-1

You can also incorporate (R^2):

TrendQuality=TrendRate×R2TrendQuality = TrendRate \times R^2

This helps distinguish:

  • price rising quickly but erratically
  • price rising steadily and persistently

This is a common quantitative implementation and should not be mislabeled as an original Jegadeesh-Titman or Moskowitz formula.


Previous (N)-day high:

HHN(t)=max⁡(Ht−1,…,Ht−N)HH_N(t) = \max(H_{t-1},\ldots,H_{t-N})

Previous (N)-day low:

LLN(t)=min⁡(Lt−1,…,Lt−N)LL_N(t) = \min(L_{t-1},\ldots,L_{t-N})

Breakout signal:

BuyifPt>HHN(t)Buy \quad\text{if}\quad P_t>HH_N(t) SellifPt<LLN(t)Sell \quad\text{if}\quad P_t<LL_N(t)

This belongs to the trend-following / breakout family.


6. The Real Difference Between Momentum and Trend

Section titled “6. The Real Difference Between Momentum and Trend”
QuestionMomentumTrend
What does it look at?Past return / relative returnPrice direction and structure
Must it compare other stocks?Cross-sectional momentum: yesNot necessarily
Does it only use its own history?TSMOM: yesUsually
Common formulasROC, 12–2 return, rank, TSMOMMA, EMA, slope, breakout
Emphasiscontinuation / relative winnerpersistent direction
Can one be up while the other is down?YesYes
Typical horizonSeveral months to one yearDays to years

Momentum is about continuation of past performance; Trend is about the directional structure of price.

They can be highly correlated, but should not be treated as conceptually identical.


7. Summary Table of Common Indicator Formulas

Section titled “7. Summary Table of Common Indicator Formulas”
IndicatorFormula / DefinitionType
ROC(P_t/P_{t-n}-1)Momentum
Log Return(\ln(P_t/P_{t-n}))Return
12–2 Momentum(R_{t-12:t-2})Cross-sectional Momentum
Rank / Percentilecross-sectional rankRelative Momentum
52w High Ratio(P_t/\max(P_{t-251:t}))Momentum / Anchor
RS Line(P_{stock}/P_{benchmark})Relative Strength
TSMOM(sign(R_{past}))Momentum + Trend
SMA(\frac1N\sum P)Trend
EMA(\alpha P_t+(1-\alpha)EMA_{t-1})Trend
MA Crossover(sign(MA_S-MA_L))Trend
SMA Slope((SMA_t-SMA_{t-k})/SMA_{t-k})Trend
Regression SlopeOLS slope of log priceTrend
Breakout(P_t>HH_N)Trend
Volatilitystd(return)(\sqrt{252})Risk
Volume Ratio(Volume/SMA(Volume))Confirmation
Residual MOMresidual return / residual volMomentum
Vol-Scaled MOM(\sigma^*/\hat\sigma \times MOM)Risk-managed Momentum

8. How to Normalize and Compare Different Signals

Section titled “8. How to Normalize and Compare Different Signals”

Directly adding values such as:

  • ROC63 = 18%
  • High52Ratio = 0.94
  • MA slope = 0.03
  • RS percentile = 91

is statistically inconsistent because they are on different scales.

Better methods include:

Across the entire stock universe:

PctRanki=ranki−1N−1PctRank_i = \frac{rank_i-1}{N-1}

Range:

0≤PctRank≤10 \le PctRank \le 1

For example:

  • ROC percentile = 0.92
  • 52w-high percentile = 0.85
  • regression-trend percentile = 0.88

These are directly comparable.


zi=xi−μxσxz_i = \frac{x_i-\mu_x}{\sigma_x}

Then winsorize, for example:

zi∗=min⁡(3,max⁡(−3,zi))z_i^* = \min(3,\max(-3,z_i))

to reduce the impact of outliers.


For example:

MomentumScore=0.35Z(MOM12−2)+0.25Z(ROC63)+0.20Z(High52Ratio)+0.20Z(RS)MomentumScore = 0.35Z(MOM_{12-2}) + 0.25Z(ROC_{63}) + 0.20Z(High52Ratio) + 0.20Z(RS)

But these weights are model-design choices. They must be validated through backtesting and should not be described as academic laws.


For example:

TrendScore=0.25SMAStack+0.25Z(Slope200)+0.25STSMOM+0.25SMultiTFTrendScore = 0.25S_{MAStack} + 0.25Z(Slope_{200}) + 0.25S_{TSMOM} + 0.25S_{MultiTF}

Again, this requires out-of-sample validation.


Your current edge_momentum_trend() roughly includes:

Pt>SMA50P_t>SMA_{50}

+3 points.

SMA50>SMA150SMA_{50}>SMA_{150}

+3.

SMA150>SMA200SMA_{150}>SMA_{200}

+3.

The code compares:

SMA200,t>SMA200,t−31SMA_{200,t}>SMA_{200,t-31}

+3.

This is a Trend signal.


Formula:

ROC63=PtPt−63−1ROC_{63} = \frac{P_t}{P_{t-63}}-1

The actual indexing uses 64 closing-price positions to obtain prices 63 sessions apart.

Scoring:

  • (ROC63\ge15%): +4
  • (0<ROC63<15%): +2

This is Momentum.


Pt≥0.85×High252P_t \ge 0.85 \times High_{252}

+2.

This is conceptually related to George-Hwang, but the 85% threshold is MarketLens-specific.


Your code also contains:

RSweighted=0.4R63+0.2R126+0.2R189+0.2R252RS_{weighted} = 0.4R_{63} + 0.2R_{126} + 0.2R_{189} + 0.2R_{252}

roughly:

  • 40% × 3-month return
  • 20% × 6-month
  • 20% × 9-month
  • 20% × 12-month

This is a custom multi-horizon momentum measure.

It is not the original formula from Jegadeesh-Titman, Carhart, or Moskowitz-Ooi-Pedersen.


Another edge uses:

Weekly:

Priceweekly>EMA10Price_{weekly}>EMA_{10} EMA10>EMA30EMA_{10}>EMA_{30}

Daily:

Price>SMA20Price>SMA_{20} SMA20>SMA50SMA_{20}>SMA_{50}

This is fundamentally Multi-Timeframe Trend.


10. MarketLens: Recommended Separation of Momentum and Trend

Section titled “10. MarketLens: Recommended Separation of Momentum and Trend”

Your current:

Momentum & Trend

can be separated into:

A. Momentum Edge
B. Trend Edge
C. Relative Strength Edge
D. Multi-Timeframe Trend Edge

This is conceptually cleaner.


Could include:

MOM12−2MOM_{12-2} ROC63ROC_{63} ROC126ROC_{126} High52RatioHigh52Ratio ResidualMomentumResidualMomentum

Could include:

P>SMA200P>SMA_{200} SMA50>SMA150>SMA200SMA_{50}>SMA_{150}>SMA_{200} Slope(SMA200)>0Slope(SMA_{200})>0 sign(R12m)sign(R_{12m}) sign(R1m)+sign(R3m)+sign(R12m)sign(R_{1m})+ sign(R_{3m})+ sign(R_{12m})

Your existing RS line is:

RSLinet=Pstock,tPbenchmark,tRSLine_t = \frac{P_{stock,t}}{P_{benchmark,t}}

If:

RSLinet≥max⁡(RSLinet−251:t−1)RSLine_t \ge \max(RSLine_{t-251:t-1})

then the relative-strength line has reached a 52-week high.

Using a market-wide percentile:

RSPercentile=PercentileRank(Momentumi)RSPercentile = PercentileRank(Momentum_i)

is closer to cross-sectional academic logic.


This is important for MarketLens.

If you score all of the following separately:

  • Price > SMA50
  • SMA50 > SMA150
  • SMA150 > SMA200
  • SMA20 > SMA50
  • EMA10 > EMA30
  • ROC63 positive
  • 52-week high
  • 12m TSMOM positive

they may all be measuring the same latent variable:

prices have been going up

So you may appear to have “8 edges,” while in reality you may only have 2–3 independent information sources.

Backtest all signals:

Corr(Signali,Signalj)Corr(Signal_i,Signal_j)

If:

Corr>0.8Corr>0.8

you should suspect double counting.


Possible buckets:

  1. Momentum
  2. Trend
  3. Relative Strength
  4. Volume / Participation
  5. Volatility / Contraction
  6. Market Regime
  7. Risk / Reward

Calculate a score inside each bucket first, then combine the bucket scores.


Method 3: Regression / Incremental Information

Section titled “Method 3: Regression / Incremental Information”

For example:

FutureReturn=α+β1MOM+β2TREND+β3RS+β4VOL+ϵFutureReturn = \alpha + \beta_1 MOM + \beta_2 TREND + \beta_3 RS + \beta_4 VOL + \epsilon

Then check whether adding a new signal improves:

  • (t)-stat
  • out-of-sample IC
  • Sharpe
  • drawdown

12. Backtesting: How to Know Whether a Formula Really Has an Edge

Section titled “12. Backtesting: How to Know Whether a Formula Really Has an Edge”

You cannot assume a formula will work in your current universe simply because it appeared in an academic paper.

Use today’s signal to predict future return:

IC=Corr(Signalt,FutureReturnt+h)IC = Corr(Signal_t,FutureReturn_{t+h})

You can use:

  • Pearson IC
  • Spearman Rank IC

Cross-sectional signals are often especially suitable for Spearman IC.


Sort stocks by signal into:

  • Q1: weakest
  • Q2
  • Q3
  • Q4
  • Q5: strongest

Then test whether:

Return(Q5)−Return(Q1)Return(Q5)-Return(Q1)

is consistently positive.


HitRate=#{FutureReturn>0}NHitRate = \frac{ \#\{FutureReturn>0\} }{ N }

But hit rate alone is not enough.


E=P(win)×AvgWin−P(loss)×AvgLossE = P(win)\times AvgWin - P(loss)\times AvgLoss

Ultimately, a strategy edge must translate into expectancy.


Sharpe=E[Rp−Rf]σ(Rp−Rf)Sharpe = \frac{ E[R_p-R_f] }{ \sigma(R_p-R_f) }

Equity peak:

Peakt=max⁡s≤tEquitysPeak_t=\max_{s\le t}Equity_s

Drawdown:

DDt=EquitytPeakt−1DD_t = \frac{Equity_t}{Peak_t}-1

Maximum Drawdown:

MDD=min⁡tDDtMDD=\min_t DD_t

Do not use the same dataset to:

  1. find a threshold
  2. test the threshold
  3. declare success

Instead:

Train period
↓
Parameter selection
↓
Validation period
↓
Lock model
↓
Out-of-sample test

Real return is:

NetReturn=GrossReturn−Commission−Spread−Slippage−MarketImpactNetReturn = GrossReturn - Commission - Spread - Slippage - MarketImpact

Transaction costs are especially important for high-turnover momentum / trend strategies.


Suppose a stock has:

MetricValue
Current Price120
63d ago100
126d ago92
189d ago88
252d ago80
SMA50112
SMA150101
SMA20095
SMA200 31d ago92
52w high125

ROC63=120100−1=20%ROC63 = \frac{120}{100}-1 = 20\%

Momentum is strong.


High52Ratio=120125=0.96High52Ratio = \frac{120}{125} = 0.96

So the stock is only 4% below its 52-week high.


120>112>101>95120>112>101>95

which satisfies:

Price>SMA50>SMA150>SMA200Price>SMA50>SMA150>SMA200

Simplified:

Slope200,31=9592−1=3.26%Slope_{200,31} = \frac{95}{92}-1 = 3.26\%

The long-term trend is rising.


3m:

R63=20%R_{63}=20\%

6m:

R126=120/92−1=30.43%R_{126} = 120/92-1 = 30.43\%

9m:

R189=120/88−1=36.36%R_{189} = 120/88-1 = 36.36\%

12m:

R252=120/80−1=50%R_{252} = 120/80-1 = 50\%

Therefore:

WeightedRS=0.4(0.20)+0.2(0.3043)+0.2(0.3636)+0.2(0.50)WeightedRS = 0.4(0.20) + 0.2(0.3043) + 0.2(0.3636) + 0.2(0.50) WeightedRS≈31.35%WeightedRS \approx31.35\%

Note:

This 31.35% is a custom weighted momentum number. It is not an annual return or a percentile unless you compare it against the full stock universe.


14. Summary of Major Researchers and Studies

Section titled “14. Summary of Major Researchers and Studies”
YearResearchersStudyCore Method
1992Brock, Lakonishok, LeBaronSimple Technical Trading RulesMA crossover, trading-range breakout
1993Jegadeesh, TitmanWinners vs LosersCross-sectional J/K momentum
1997CarhartFour-Factor ModelAdds momentum factor
1999Moskowitz, GrinblattIndustry MomentumIndustry winners / losers
1999Hong, SteinUnderreaction / OverreactionMomentum theory mechanism
2000Lee, SwaminathanMomentum + VolumeVolume explains momentum persistence
2004George, Hwang52-Week HighPrice / 52-week-high ratio
2011Blitz, Huij, MartensResidual MomentumFactor-adjusted residual momentum
2012Novy-MarxIntermediate HorizonEmphasizes information from months 12–7
2012Moskowitz, Ooi, PedersenTime-Series Momentumsign(past return) + vol scaling
2012Fama, FrenchInternational MomentumMomentum across regions
2013Asness, Moskowitz, PedersenMomentum EverywhereAcross asset classes
2015Barroso, Santa-ClaraRisk-Managed MomentumVolatility-managed exposure
2016Daniel, MoskowitzMomentum CrashesRegime-dependent crash risk
2016Levine, PedersenWhich Trend Is Your Friend?TSMOM / MA / linear filters
2016Han, Zhou, ZhuTrend FactorMulti-horizon moving-average information
2017Hurst, Ooi, PedersenCentury of Trend Following1m/3m/12m TSMOM

If MarketLens is to evolve gradually from its current heuristic model into a more research-driven system, I would use this order:

Phase 1 — Keep the Current Model but Separate the Categories

Section titled “Phase 1 — Keep the Current Model but Separate the Categories”
Momentum
Trend
Relative Strength
MTF
Volume / Price Action
Market
Risk

Phase 2 — Add an Academic Momentum Baseline

Section titled “Phase 2 — Add an Academic Momentum Baseline”

First add:

MOM12−2MOM_{12-2}

and:

CrossSectionalPercentile(MOM12−2)CrossSectionalPercentile(MOM_{12-2})

because these are important academic benchmarks.


Add:

sign(R1m),sign(R3m),sign(R12m)sign(R_{1m}), \quad sign(R_{3m}), \quad sign(R_{12m})

to create a multi-horizon trend measure.


Phase 4 — Convert Trend Into a Continuous Score

Section titled “Phase 4 — Convert Trend Into a Continuous Score”

Move beyond:

pass / fail

to something like:

TrendStrength=f(MA slope, regression slope, R2, MA alignment)TrendStrength = f(MA\ slope,\ regression\ slope,\ R^2,\ MA\ alignment)

Test correlations among:

ROC63
MOM12-2
High52
RS percentile
SMA slope
TSMOM12
MTF

Remove highly redundant signals.


Do not simply ask:

Which weight is best?

Instead ask:

Which weight remains stable out of sample across different years and different market regimes?


  1. Academic evidence does not guarantee future profitability.
  2. Results can change across markets, time periods, and transaction-cost environments.
  3. Momentum has crash risk.
  4. Trend signals have lag.
  5. Short-term moving averages can whipsaw easily.
  6. Cross-sectional momentum requires a clearly defined universe.
  7. Survivorship bias can materially exaggerate results.
  8. Look-ahead bias must be completely avoided.
  9. Delisted stocks should not disappear from the historical universe.
  10. Adjusted prices are usually preferable for splits / dividends, but the treatment must match the research objective.
  11. The same period should not be used for both tuning and evaluation.
  12. Multiple highly correlated signals should not be counted as multiple independent edges.
  13. Thresholds such as ROC63 >= 15% and near high >= 85% should be labeled as model parameters, not natural laws.
  14. Academic long-short portfolio results do not imply that a long-only stock-picking strategy will produce identical results.

  1. Jegadeesh, N. & Titman, S. (1993). Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. The Journal of Finance, 48(1), 65–91.
    https://doi.org/10.1111/j.1540-6261.1993.tb04702.x
    PDF mirror: https://www.bauer.uh.edu/rsusmel/phd/jegadeesh-titman93.pdf

  2. Carhart, M. M. (1997). On Persistence in Mutual Fund Performance. The Journal of Finance, 52(1), 57–82.
    https://doi.org/10.1111/j.1540-6261.1997.tb03808.x

  3. Kenneth R. French Data Library. Momentum Factor (Mom) / Prior (2–12) Return.
    https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
    Momentum construction: https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_Library/det_mom_factor.html

  4. Moskowitz, T. J. & Grinblatt, M. (1999). Do Industries Explain Momentum? The Journal of Finance, 54(4), 1249–1290.
    https://doi.org/10.1111/0022-1082.00146

  5. Lee, C. M. C. & Swaminathan, B. (2000). Price Momentum and Trading Volume. The Journal of Finance, 55(5), 2017–2069.
    https://doi.org/10.1111/0022-1082.00280

  6. George, T. J. & Hwang, C.-Y. (2004). The 52-Week High and Momentum Investing. The Journal of Finance, 59(5), 2145–2176.
    https://doi.org/10.1111/j.1540-6261.2004.00695.x
    PDF: https://www.bauer.uh.edu/tgeorge/papers/gh4-paper.pdf

  7. Blitz, D., Huij, J. & Martens, M. (2011). Residual Momentum. Journal of Empirical Finance, 18(3), 506–521.
    https://doi.org/10.1016/j.jempfin.2011.01.003

  8. Novy-Marx, R. (2012). Is Momentum Really Momentum? Journal of Financial Economics, 103(3), 429–453.
    https://doi.org/10.1016/j.jfineco.2011.05.003

  9. Fama, E. F. & French, K. R. (2012). Size, Value, and Momentum in International Stock Returns. Journal of Financial Economics, 105(3), 457–472.
    https://doi.org/10.1016/j.jfineco.2012.05.011

  10. Asness, C. S., Moskowitz, T. J. & Pedersen, L. H. (2013). Value and Momentum Everywhere. The Journal of Finance, 68(3), 929–985.
    https://doi.org/10.1111/jofi.12021
    PDF: https://pages.stern.nyu.edu/~lpederse/papers/ValMomEverywhere.pdf


  1. Moskowitz, T. J., Ooi, Y. H. & Pedersen, L. H. (2012). Time Series Momentum. Journal of Financial Economics, 104(2), 228–250.
    https://doi.org/10.1016/j.jfineco.2011.11.003
    PDF: https://w4.stern.nyu.edu/facdir/lpederse/papers/TimeSeriesMomentum.pdf

  2. Levine, A. & Pedersen, L. H. (2016). Which Trend Is Your Friend? Financial Analysts Journal, 72(3).
    https://doi.org/10.2469/faj.v72.n3.3

  3. Han, Y., Zhou, G. & Zhu, Y. (2016). A Trend Factor: Any Economic Gains from Using Information over Investment Horizons? Journal of Financial Economics, 122(2), 352–375.
    https://doi.org/10.1016/j.jfineco.2016.01.029

  4. Hurst, B., Ooi, Y. H. & Pedersen, L. H. (2017). A Century of Evidence on Trend-Following Investing. The Journal of Portfolio Management, 44(1).
    SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2993026
    PDF: https://fairmodel.econ.yale.edu/ec439/hurst.pdf

  5. Brock, W., Lakonishok, J. & LeBaron, B. (1992). Simple Technical Trading Rules and the Stochastic Properties of Stock Returns. The Journal of Finance, 47(5), 1731–1764.
    https://doi.org/10.1111/j.1540-6261.1992.tb04681.x


  1. Barroso, P. & Santa-Clara, P. (2015). Momentum Has Its Moments. Journal of Financial Economics, 116(1), 111–120.
    https://doi.org/10.1016/j.jfineco.2014.11.010

  2. Daniel, K. & Moskowitz, T. J. (2016). Momentum Crashes. Journal of Financial Economics, 122(2), 221–247.
    https://doi.org/10.1016/j.jfineco.2015.12.002

  3. Hong, H. & Stein, J. C. (1999). A Unified Theory of Underreaction, Momentum Trading, and Overreaction in Asset Markets. The Journal of Finance, 54(6), 2143–2184.
    https://doi.org/10.1111/0022-1082.00184


The most important conclusions can be condensed into four statements:

  1. Momentum ≠ Trend.
  2. Cross-sectional momentum is about relative ranking across stocks.
  3. Time-series momentum is about persistence in an asset’s own past direction, so it is closely related to trend following.
  4. MarketLens should calculate Momentum, Trend, Relative Strength, and MTF separately and combine them only at the composite layer, rather than treating every upward-price signal as an independent edge and repeatedly awarding points for the same information.

For MarketLens, a cleaner research architecture could be:

Price Data
│
├── Momentum
│ ├── MOM 12–2
│ ├── ROC63 / ROC126
│ └── 52-Week High
│
├── Trend
│ ├── MA Stack
│ ├── MA Slope
│ ├── Regression Slope
│ └── TSMOM 1m / 3m / 12m
│
├── Relative Strength
│ ├── Stock / Benchmark RS Line
│ └── Cross-sectional Percentile
│
├── Confirmation
│ ├── Volume
│ ├── Breakout
│ └── Multi-Timeframe
│
└── Risk / Market Regime
├── Volatility
├── Broad Market
├── Drawdown
└── Position Sizing

This makes the mathematical meaning of each edge much clearer and makes it easier to backtest each signal independently to determine whether it contributes genuinely new information or simply measures the same underlying trend again.