Momentum Investing: Time-Series and Cross-Sectional Strategies
Momentum investing uses past returns to decide which assets to hold, avoid or sell short. The definition matters: ranking stocks against one another is different from following each asset's own trend. Both have a research literature, but neither establishes that an arbitrary RSI/MACD combination will outperform a benchmark.
Two Distinct Momentum Approaches
| Approach | Signal | Consequence |
|---|---|---|
| Cross-sectional momentum | Rank assets by past performance relative to a defined universe. | The highest-ranked asset can still have a negative absolute return. |
| Time-series momentum | Compare an asset's own past return with a specified reference, such as zero excess return. | Different assets can all receive positive or all receive negative signals. |
A long-only stock ranking, a long-short equity portfolio and a futures trend strategy have different exposures. Do not compare their headline returns without accounting for leverage, cash, financing, shorting and volatility.
What the Research Supports
Jegadeesh and Titman's 1993 study documented historical returns from buying past stock winners and selling past losers over intermediate holding periods. It is evidence about specified portfolios and samples, not a forecast for the next stock selected from a chart.
Moskowitz, Ooi and Pedersen's Time Series Momentum examined an asset's own past returns across futures and forward contracts. This distinction is more useful than defining momentum and trend-following solely by short versus long timeframes; time-series momentum is one form of trend-following.
Daniel and Moskowitz's Momentum Crashes documents severe momentum losses associated with market rebounds after declines and high volatility. A strategy that shorts past losers can be exposed when those losers rebound sharply. A long-only portfolio has a different loss mechanism and should not inherit the same numerical crash estimates without analysis.
Research findings need their original universe, formation period, holding period, weighting, sample dates and cost assumptions. This article presents no newly measured return, ranking-persistence table or claimed improvement from combining signals.
Specify a Portfolio Before Testing It
An illustrative research design could rank a predefined group of liquid assets once a month by their prior six-month total return, hold a fixed number of leaders at equal weights and rebalance at the next available trading opportunity. These choices are unvalidated examples, not optimized recommendations.
Write down how the universe is constructed at each historical date. Using only today's surviving stocks introduces survivorship bias. Use prices adjusted consistently for splits and distributions, and prevent future index membership or other later information from entering the historical selection.
Define tie-breaking, missing history, delistings, cash treatment and position limits. If a rank drops below the cutoff, decide whether to sell immediately at the next rebalance or apply a predetermined buffer. A buffer can reduce turnover but also retains some weaker-ranked holdings; measure the tradeoff rather than assuming it improves returns.
Indicators Answer Different Questions
RSI summarizes the balance of gains and losses over its lookback. A reading above 70 does not require a reversal. MACD compares moving averages; a positive histogram means MACD is above its signal line, not necessarily that price is rising or that the histogram is increasing.
A moving average can define a trend filter, while percentage rate of change can provide a ranking score. Combining these calculations often reuses the same price history. Agreement is not independent confirmation and does not by itself establish higher accuracy.
If adding a filter, compare the filtered and unfiltered rules on identical data and costs. Inspect trades excluded by the filter as well as those retained.
Compare Momentum with a Relevant Benchmark
| Question | What to keep consistent |
|---|---|
| Did selection add value? | Compare with a portfolio drawn from the same eligible universe. |
| Did lower exposure explain the result? | Report time invested, cash return and leverage. |
| Did turnover consume the benefit? | Include spreads, commissions, market impact and financing where applicable. |
| Was the result concentrated? | Inspect contribution by asset, sector and period. |
| Did risk change? | Compare volatility, drawdown, recovery time and tail losses. |
A momentum strategy can remain fully invested during a broad decline if it simply holds the least weak assets. Exiting before a reversal is an aspiration, not an inherent capability. Stops can execute beyond their intended price after gaps.
Buy-and-hold also requires a defined asset allocation and rebalancing policy. Neither approach guarantees a positive long-term result, and there is no universal portfolio percentage suitable for a momentum allocation.
Separate Development from Evaluation
Choose formation periods and rebalancing rules on a development sample. Keep a later evaluation period untouched. Record every variant tried, including failed ones, so a selected winner is not presented as the only experiment.
Report net returns, turnover, drawdown and benchmark-relative results together. Repeat the analysis with plausible higher costs and small parameter changes. Forward paper observation can expose timing and implementation errors, but a few months of paper trading does not establish a durable investment advantage.
Pineify can help draft chart-based momentum rules for TradingView. A single-symbol strategy test is not automatically a point-in-time portfolio backtest with historical constituents and synchronized rebalancing. Use a suitable portfolio dataset and engine when the question requires them.