royy/chinese-equity-quant
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Rewrite reversal report as tutorial

Browse Source
This commit is contained in:
Yuxuan Yan
2026-06-12 17:12:29 +08:00
parent 2c0ca53bd6
commit 16b4988f16
2 changed files with 426 additions and 133 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/reversal_5d_all_universe_pipeline_report.md
+422 -129
View File
@@ -1,134 +1,368 @@
# 5-Day Reversal — End-to-End Pipeline Report # Tutorial: Testing a 5-Day Reversal Alpha
Generated: 2026-06-11T17:17:34 This document is a teaching walkthrough for someone who is new to this
research framework and only lightly familiar with quant research. We will use
one concrete experiment, a 5-day reversal alpha on the full downloaded Chinese
A-share universe, to learn how the framework defines an alpha, stores it, tests
it, turns it into a portfolio, and explains the gap between a research result
and simulated trading PnL.
This report runs the **5-day reversal** signal end to end through the decoupled The original experiment was generated on 2026-06-11. The important point is not
pipeline (`data → alpha → combo → portfolio build → portfolio simulate/eval`) on the timestamp; it is the research method.
the full downloaded A-share universe, and answers the seven review questions:
alpha storage, metric sanity, NaN/look-ahead handling, alpha↔portfolio
closeness, alpha↔PnL closeness, per-phase timing, and visualizations.
Per this repo's convention an **alpha is a signed cross-sectional position ## The Research Question
weight, not a return predictor**, so evaluation is return / Sharpe / turnover /
drawdown — there is deliberately **no IC/IR** anywhere.
## TL;DR A quant research project starts with a hypothesis:
The naive built-in `reversal` alpha (raw `-pct_change(5)` then cross-sectional > If a stock fell a lot over the last few trading days, it may rebound soon; if
**z-score**) loses **-87.45%** in costless research on the full > it rose a lot, it may cool off soon.
~5,200-name universe. That is **not** evidence the signal is bad — it is an
artifact of z-score weighting on A-shares: a handful of newly listed /
post-suspension / limit-up names produce huge `pct_change` outliers, and
z-scoring pours the book into exactly those names (stored weights reach
-52σ).
Switching only the **weighting** to a bounded cross-sectional **rank** This is called **short-horizon reversal**. It is a simple idea: recent losers
(`reversal_rank`) and restricting to a per-date **liquid, non-ST, tradable** are candidates to buy, and recent winners are candidates to sell or underweight.
universe recovers a genuine reversal edge: **72.24%** In this repo, the tested version looks back 5 trading days.
costless research cumulative return at Sharpe
**0.73** with a
54.31% daily hit rate.
The binding constraint is **cost, not signal**: at ~148×/year The central research question is:
turnover, a 10 bps one-way per-trade cost (5 bps commission + 5 bps slippage,
charged on each leg — so ~20 bps per round trip) erases the edge — every variant
is negative after costs. A tradable 5-day reversal needs
turnover control, not a different signal.
## Headline Metrics > Does this 5-day reversal rule create useful portfolio returns after the
> framework applies realistic storage, portfolio construction, execution
> constraints, and trading costs?
| run | weighting | research cum | research Sharpe | research turn/yr | exec before cost | exec net | exec net Sharpe | The answer from this run is nuanced:
| --- | --- | --- | --- | --- | --- | --- | --- |
| naive z-score (full) | z-score | -87.45% | -2.4515 | 160× | 18.39% | -111.94% | -1.4508 |
| rank (full) | rank | -3.48% | -0.0198 | 143× | 50.52% | -66.61% | -1.1839 |
| rank (liquid subset) | rank | 72.24% | 0.7310 | 148× | 110.18% | -17.16% | -0.2226 |
*Research = costless, no-look-ahead weights · next-day return. Execution = next-open - The naive built-in version loses badly on the full universe because raw
fills on the discretized integer book under suspension / price-limit / volume-cap z-score weighting is too sensitive to A-share outliers.
constraints, 5 bps commission + 5 bps slippage.* - A rank-weighted version on a liquid, non-ST, tradable universe has a positive
costless research result.
- The daily-traded implementation is still not tradable after costs because
turnover is too high.
That is a normal research outcome. Good research is not just asking "did the
backtest go up?" It is asking **which layer explains the result**: signal,
weighting, universe, construction, execution, or cost.
## How This Framework Defines An Alpha
In many quant textbooks, an alpha is described as a **prediction** of future
returns. This framework uses a stricter and more practical convention:
> An alpha is a signed cross-sectional position weight.
That sentence is the key to the whole repo.
- **Signed** means positive values are long exposure and negative values are
short exposure.
- **Cross-sectional** means the alpha compares stocks to other stocks on the
same date.
- **Position weight** means the output is already an instruction about what the
portfolio wants to own. It is not merely a score to correlate with future
returns.
The stored alpha file always has this schema:
| column | meaning |
| --- | --- |
| `symbol_id` | Stock identifier such as `sh600000` or `sz000001`. |
| `date` | The signal date. The alpha is formed using information known by this date's close. |
| `alpha_name` | A label for this particular run, such as `reversal_5d_all`. |
| `weight` | Signed desired exposure. Positive means long; negative means short. |
Because the framework treats alphas as position weights, it evaluates them with
portfolio metrics: return, Sharpe, turnover, drawdown, and hit rate. It does
**not** use IC/IR, because IC/IR would treat the alpha as a return predictor.
## The Pipeline In One Picture
Every phase reads parquet files and writes parquet files. That makes the system
easy to inspect and rerun one layer at a time.
```text
daily bars
-> alpha weights
-> combined weights
-> portfolio targets and integer positions
-> simulated fills and PnL
-> evaluation metrics
```
For this experiment, the important phases are:
| phase | command family | what it teaches you |
| --- | --- | --- |
| Data | `cli.py data download` | What market data is available. |
| Alpha compute | `cli.py alpha compute` | How a raw research idea becomes stored weights. |
| Alpha eval | `cli.py alpha eval` | How those weights perform in a clean costless research view. |
| Combo | `cli.py combo combine` | How one or more alphas become one combined book. |
| Portfolio build | `cli.py portfolio build` | How weights become target values and integer shares. |
| Portfolio simulate | `cli.py portfolio simulate` | How the integer book trades at next open with constraints and costs. |
| Portfolio eval | `cli.py portfolio eval` | How the continuous target portfolio behaves as a research portfolio. |
In a real research workflow, you should learn to pause after every phase and
inspect the parquet output. Most mistakes are easier to find at the interface
between two phases than at the final PnL line.
## Step 1: Define The Raw Reversal Signal
The built-in 5-day reversal alpha is implemented as:
```python
signal = -close.pct_change(5, fill_method=None)
```
For stock `i` on date `t`, this is approximately:
```text
signal[i, t] = -(close[i, t] / close[i, t-5] - 1)
```
So:
- If a stock rose by 10% over the last 5 trading days, the raw signal is `-10%`.
It becomes a candidate short or underweight.
- If a stock fell by 10% over the last 5 trading days, the raw signal is `+10%`.
It becomes a candidate long or overweight.
Notice the timing. The signal uses prices through date `t`. It must not use the
return from `t` to `t+1`, because that is the future. The costless alpha
evaluator tests the weight formed on date `t` against the next close-to-close
return; the later execution simulator is the separate layer that trades the
constructed integer book at the next open.
The code lives in `pipeline/alpha/library/reversal.py`:
```python
class ReversalAlpha(BaseAlpha):
name = "reversal"
def __init__(self, lookback: int = 5):
self.lookback = lookback
def signal(self, close: pd.DataFrame) -> pd.DataFrame:
return -close.pct_change(self.lookback, fill_method=None)
```
The alpha class only defines the raw signal. The base class then turns that
signal into weights.
## Step 2: Turn A Signal Into Cross-Sectional Weights
By default, `BaseAlpha.to_weights()` does a cross-sectional z-score each date:
```text
weight[i, t] = (signal[i, t] - mean_signal[t]) / std_signal[t]
```
This means the framework asks:
> On this date, which stocks have stronger reversal scores than the rest of the
> market, and by how much?
That is useful, but it has a weakness. If a few stocks have extreme trailing
returns because they are newly listed, suspended, illiquid, or limit-constrained,
z-scoring can put a very large amount of relative exposure into exactly those
names.
That is what happened in the naive full-universe run. Stored weights reached
about `-52` standard deviations. The research result collapsed:
| run | weighting | research cumulative return | research Sharpe | research turnover/year |
| --- | --- | --- | --- | --- |
| naive z-score, full universe | z-score | -87.45% | -2.4515 | 160x |
The lesson is not "reversal is bad." The lesson is:
> The same raw signal can become a bad portfolio if the weighting method reacts
> badly to outliers.
## Step 3: Make The Weighting More Robust
The repo also has a rank-weighted version, `reversal_rank`. It uses the same raw
5-day reversal signal, but converts the cross-section to ranks instead of
z-scores:
```python
ranks = signal.rank(axis=1)
weights = ranks.subtract(ranks.mean(axis=1), axis=0)
```
Rank weighting keeps the ordering of stocks but removes the importance of the
exact outlier magnitude. A stock can be "the worst recent loser" or "the best
recent winner," but it cannot become 52 standard deviations important just
because its raw percentage move is unusual.
The full-universe rank version was much less pathological, but still not a
clean signal:
| run | weighting | research cumulative return | research Sharpe | research turnover/year |
| --- | --- | --- | --- | --- |
| rank, full universe | rank | -3.48% | -0.0198 | 143x |
That tells us the weighting fix helped, but the universe still contains many
names that are poor candidates for a daily reversal strategy.
## Step 4: Define The Investable Universe
An alpha should be tested on stocks that could plausibly be traded. The liquid
run applies a per-date mask before weights are created. A stock must be:
- seasoned, with at least 60 observed closes;
- currently tradable, using `tradestatus == 1`;
- not ST, using `isST == 0`;
- inside the top 1000 names by trailing 20-day average traded amount.
This mask is applied to the signal, not to the price history used to compute the
5-day return. That distinction matters. We still compute `pct_change(5)` on the
full contiguous price history, then decide which names are eligible to hold on
each signal date.
The liquid rank result is the cleanest research result:
| run | weighting | universe | research cumulative return | research Sharpe | hit rate |
| --- | --- | --- | --- | --- | --- |
| rank, liquid subset | rank | top 1000 liquid, tradable, non-ST | 72.24% | 0.7310 | 54.31% |
This is the first point where a researcher can say:
> There appears to be a real 5-day reversal effect in a cleaner A-share
> universe, before trading costs.
That last phrase, **before trading costs**, is essential.
![Research equity](assets/reversal_5d_research_equity.png) ![Research equity](assets/reversal_5d_research_equity.png)
## 1. Are Alpha Values Properly Stored? When reading this chart, focus on the shape and relative behavior:
All alpha artifacts conform to `ALPHA_COLUMNS` (`symbol_id, date, alpha_name, - The naive z-score line shows the outlier problem.
weight`), carry no null / non-finite weights, no duplicate `(symbol_id, date)` - The rank full-universe line shows that robust weighting helps but does not
keys, and have numerically-zero daily cross-sectional means (weights are fully solve the universe problem.
demeaned per date). - The liquid rank line shows the signal-level edge before execution costs.
| run | schema ok | null w | non-finite w | dup keys | max |daily mean| | weight range | combo identity Δ | ## Step 5: Check That The Alpha File Is Sane
Before trusting any metric, inspect the stored alpha artifact. The run checked:
- The columns match `ALPHA_COLUMNS`.
- There are no null weights.
- There are no non-finite weights.
- There are no duplicate `(symbol_id, date)` rows.
- The daily cross-sectional mean is approximately zero.
- A one-alpha combo is an exact identity transform.
| run | schema ok | null weights | non-finite weights | duplicate keys | max abs daily mean | weight range | combo identity diff |
| --- | --- | --- | --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- | --- | --- | --- |
| naive z-score (full) | True | 0 | 0 | 0 | 3.32e-16 | [-52.2, 19.2] | 0.00e+00 | | naive z-score (full) | True | 0 | 0 | 0 | 3.32e-16 | [-52.2, 19.2] | 0.00e+00 |
| rank (full) | True | 0 | 0 | 0 | 0.00e+00 | [-2603.0, 2603.0] | 0.00e+00 | | rank (full) | True | 0 | 0 | 0 | 0.00e+00 | [-2603.0, 2603.0] | 0.00e+00 |
| rank (liquid subset) | True | 0 | 0 | 0 | 0.00e+00 | [-498.5, 498.5] | 0.00e+00 | | rank (liquid subset) | True | 0 | 0 | 0 | 0.00e+00 | [-498.5, 498.5] | 0.00e+00 |
The decisive storage signal is the **weight range**. The naive z-score alpha The rank ranges look numerically large because rank weights scale with the
stores weights as extreme as number of names. That is fine: later evaluation divides by gross exposure, and
`[-52, 19]` — portfolio construction normalizes by `sum(abs(weight))`. The important
single names tens of sigma from the cross-section. Rank weighting is bounded by difference is that rank weights are bounded by cross-sectional rank, not by the
construction, so its stored weights are well-behaved. Same signal, completely raw size of an abnormal stock move.
different book.
![Weight distributions](assets/reversal_5d_weight_distributions.png) ![Weight distributions](assets/reversal_5d_weight_distributions.png)
## 2. Do The Alpha Metrics Make Sense? This is a good habit: when a backtest looks strange, plot the weights before
debugging the PnL. A broken or concentrated weight distribution often explains
the result.
Yes, and they tell a coherent story: ## Step 6: Understand The Alpha Evaluation Formula
- The **z-score full** run is dominated by a few outlier names; its research The costless alpha evaluator asks:
Sharpe of -2.45 reflects a
book that is effectively long/short a tiny set of extreme movers, which in
A-shares keep trending — so the reversal bet loses.
- **Rank full** (-3.48%) is roughly flat:
the direction is right (hit rate
51.18%) but
the long tail of illiquid / ST / freshly listed names adds noise.
- **Rank liquid** is the clean result: a positive, monotone reversal premium
(72.24%, Sharpe
0.73) once the
investable universe is sane.
This matches the prior literature that short-horizon reversal is a real but > If we held the alpha weights from date `t`, what close-to-close return would
liquidity- and cost-sensitive A-share effect. > we earn from `t` to `t+1`?
## 3. NaN And Look-Ahead Handling This is intentionally a **research-layer approximation**, not the trading
simulator. At this stage the framework has only an alpha weight file. It has not
yet rounded shares, checked limits, clipped trades, or paid costs. The purpose
is to answer a clean signal question: "Do these close-formed weights line up
with the next day's returns?"
- The raw signal uses `close.pct_change(5, fill_method=None)` — missing prices The actual trading layer comes later. `portfolio simulate` takes the integer
are **not** forward-filled, so a suspended name does not silently inherit a `position_shares` from the portfolio builder, executes the target from signal
stale price. date `t` at `open[t+1]`, then marks PnL as overnight movement on the old book
- Weights are formed at close `t` and earn the **next** close-to-close return plus intraday movement on the newly filled book, minus trading cost.
`t → t+1`. Forward returns are computed on the full market calendar *before*
selecting signal dates, so a sparse signal grid still earns the next
*available* return rather than the next signal date. The final signal date,
which has no forward return, is dropped from metrics (that is why the
research day count is one less than the stored signal-date count).
- The liquid-universe mask is applied to the **signal**, not to the price
history: `pct_change(5)` is always computed on contiguous prices, and the mask
only decides what is *held*. It uses `tradestatus`, `isST`, a ≥60-session
seasoning rule, and a trailing-20-day liquidity rank — all backward-looking.
## 4. How Close Are Alpha And Constructed Portfolio? The daily research return is:
`portfolio build` normalizes the alpha to `target_weight = w / Σ|w|` and scales ```text
by booksize. The continuous target portfolio is an exact normalization of the R[t] = sum_i(weight[i, t] * return[i, t+1]) / sum_i(abs(weight[i, t]))
stored alpha (research return correlation ≈ 1.0); the **integer** book then ```
diverges because small per-name targets are rounded away under A-share lot
rules.
| run | target_value identity max|Δ| | alpha→target max|Δ| | research corr(alpha,portfolio) | mean integer gross | mean L1 tracking | This has three important consequences:
- The alpha is normalized by its gross exposure, so the scale of raw weights
does not by itself create a higher return.
- The next day's return is used, so the test avoids look-ahead.
- The last signal date is dropped from performance metrics because there is no
next return for it.
Turnover is also measured from the weights:
```text
turnover[t] = sum_i(abs(weight[i, t] - weight[i, t-1])) / sum_i(abs(weight[i, t-1]))
```
The annualized turnover numbers around 143x to 160x are a warning. Even a
positive signal can be hard to monetize if it asks the portfolio to trade too
much every day.
## Step 7: Build A Portfolio From The Alpha
The alpha file is still an abstract research book. `portfolio build` turns it
into target exposures and integer shares.
The main normalization is:
```text
target_weight[i, t] = weight[i, t] / sum_i(abs(weight[i, t]))
target_value[i, t] = booksize * target_weight[i, t]
target_shares[i, t] = target_value[i, t] / construction_price[i, t]
```
Then the framework creates an integer A-share book using lot rules and repair
logic. This is where a research portfolio starts to become a tradable portfolio.
The continuous target portfolio matched the stored alpha almost exactly:
| run | target value identity max abs diff | alpha to target max abs diff | research correlation alpha vs portfolio | mean integer gross | mean L1 tracking |
| --- | --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- | --- |
| naive z-score (full) | 0.0000 | 0.00e+00 | 1.000000 | 9,138,331 | 2,542,655 | | naive z-score (full) | 0.0000 | 0.00e+00 | 1.000000 | 9,138,331 | 2,542,655 |
| rank (full) | 0.0000 | 0.00e+00 | 1.000000 | 8,984,098 | 2,678,278 | | rank (full) | 0.0000 | 0.00e+00 | 1.000000 | 8,984,098 | 2,678,278 |
| rank (liquid subset) | 0.0000 | 0.00e+00 | 1.000000 | 9,810,256 | 862,303 | | rank (liquid subset) | 0.0000 | 0.00e+00 | 1.000000 | 9,810,256 | 862,303 |
The integer book is not exact because small target positions can be rounded
away. The liquid subset has lower tracking error because it spreads the book
over fewer and more tradable names.
![Portfolio tracking](assets/reversal_5d_portfolio_tracking.png) ![Portfolio tracking](assets/reversal_5d_portfolio_tracking.png)
## 5. How Close Are Alpha Metrics And Final PnL? When you research a new alpha, ask two separate questions:
The costless research metric and the simulated net PnL diverge for two - Does the continuous target portfolio match the alpha? It should.
mechanical reasons, both quantified below: (a) **execution friction** — next-open - Does the integer tradable portfolio still resemble the target? It may not,
fills, integer shares, and constraints; and (b) **cost** — the dominant term especially for small books or very broad universes.
here.
## Step 8: Simulate Execution And Costs
Research returns are not the same as tradable PnL. The simulator executes the
integer `position_shares` at the next available open and applies constraints:
- suspension;
- price limit;
- volume cap;
- proportional trading cost.
The cost model is:
```text
cost = abs(traded_shares * open) * (cost_bps + slippage_bps) / 10000
```
For this run, cost is 5 bps commission plus 5 bps slippage. Slippage is treated
as cash cost, not as an additional execution price adjustment.
The execution results explain the final research conclusion:
| run | corr(alpha, exec net) | PnL before cost | total cost | net PnL | mean daily turnover | | run | corr(alpha, exec net) | PnL before cost | total cost | net PnL | mean daily turnover |
| --- | --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- | --- |
@@ -136,59 +370,118 @@ here.
| rank (full) | 0.9613 | 5,052,067 | 11,713,451 | -6,661,383 | 0.5133 | | rank (full) | 0.9613 | 5,052,067 | 11,713,451 | -6,661,383 | 0.5133 |
| rank (liquid subset) | 0.9762 | 11,017,842 | 12,733,803 | -1,715,960 | 0.5715 | | rank (liquid subset) | 0.9762 | 11,017,842 | 12,733,803 | -1,715,960 | 0.5715 |
The research↔execution-net daily-return correlation stays high (the book *does* The liquid rank run made about 11.0 million before cost, but paid about 12.7
track the signal), but the level collapses after cost. For the liquid run, gross million in cost. That is why the final net PnL is negative.
costless edge is real yet total cost
(**12,733,803**) This is not a contradiction. It is exactly what a research pipeline should show:
swamps it. This is the central finding: 5-day reversal is a signal you must trade
*slowly* to monetize. > The signal exists in the costless layer, but the daily implementation trades
> too much to keep the edge.
![Execution vs research](assets/reversal_5d_exec_vs_research.png) ![Execution vs research](assets/reversal_5d_exec_vs_research.png)
## 6. Time Consumption By Phase ## Step 9: Read The Headline Metrics Like A Researcher
| phase | rank full (s) | rank liquid (s) | The complete summary is:
| --- | --- | --- |
| alpha compute | 108.3 | 116.1 |
| alpha eval | 98.0 | 118.7 |
| combo combine | 22.9 | 22.5 |
| portfolio build | 599.8 | 254.3 |
| portfolio eval | 94.2 | 90.0 |
| portfolio simulate | 168.7 | 163.3 |
| total | 1091.8 | 764.9 |
![Phase timings](assets/reversal_5d_phase_timings.png) | run | weighting | research cumulative return | research Sharpe | research turnover/year | exec before cost | exec net | exec net Sharpe |
| --- | --- | --- | --- | --- | --- | --- | --- |
| naive z-score (full) | z-score | -87.45% | -2.4515 | 160x | 18.39% | -111.94% | -1.4508 |
| rank (full) | rank | -3.48% | -0.0198 | 143x | 50.52% | -66.61% | -1.1839 |
| rank (liquid subset) | rank | 72.24% | 0.7310 | 148x | 110.18% | -17.16% | -0.2226 |
`portfolio build` dominates because it iterates per signal date and repairs a Here is the interpretation:
multi-thousand-name integer book under lot rules. The liquid run is faster
across the board because it carries far fewer non-zero names per date.
## 7. Reproduce The Run - **Naive z-score full universe**: not a useful test of the reversal idea,
because the weighting scheme lets outliers dominate the book.
- **Rank full universe**: a better test of the same idea, but still noisy
because the universe includes too many problematic names.
- **Rank liquid subset**: the best signal-level test; it finds a positive
costless reversal effect.
- **Execution net**: all variants lose after cost at daily rebalance frequency,
so the implementation is not yet tradable.
A beginner might look only at the final net PnL and say "the alpha failed." A
researcher should be more precise:
> The raw 5-day reversal idea has signal value in a liquid universe, but the
> current daily trading rule has too much turnover for the assumed cost model.
That distinction tells you what to try next.
## Step 10: Reproduce The Experiment
These commands reproduce the important artifacts, assuming the full daily-bar
dataset already exists at `data/daily_bars/all`.
```bash ```bash
# naive z-score baseline (full universe) — the built-in alpha, unchanged # Naive z-score baseline: built-in reversal alpha, full universe.
uv run python cli.py alpha compute --data-path data/daily_bars/all \ uv run python cli.py alpha compute --data-path data/daily_bars/all \
--alpha-name reversal_5d_all --alpha-type reversal --lookback 5 --output-dir alphas --alpha-name reversal_5d_all --alpha-type reversal --lookback 5 \
--output-dir alphas
# robust rank weighting, full + liquid universe (one script, both runs) # Rank-weighted full and liquid runs.
bash scripts/run_reversal_rank_e2e.sh bash scripts/run_reversal_rank_e2e.sh
# regenerate this report + figures # Regenerate figures, diagnostics, and the older auto-generated report.
# This command rewrites this markdown file, so run it only when you want
# generated output to replace the tutorial.
uv run python scripts/generate_reversal_5d_report.py uv run python scripts/generate_reversal_5d_report.py
``` ```
## Interpretation & Next Steps If you are learning the framework, do not run the whole pipeline blindly. Run
one phase, inspect the output parquet, then continue.
The pipeline is internally consistent end to end: storage validates, the trivial ## How To Research Your Own Alpha
one-alpha combo is an exact identity, the continuous target portfolio matches the
alpha, and the execution layer cleanly explains the gap to net PnL via friction
and cost. The premise that 5-day reversal "produces not-bad PnL" holds **at the
signal level** once weighting and universe are sane (rank + liquid), but **fails
net of cost** at daily rebalance frequency.
Recommended next diagnostics: Use this checklist for a new idea.
- **Turnover control** — the highest-leverage lever: hold bands / no-trade zones, 1. State the hypothesis in plain language.
weight smoothing, or longer rebalance spacing to cut the ~150×/yr turnover. Example: "Stocks with poor 5-day returns may rebound over the next day."
- Volatility-scaled or decayed reversal to reduce churn.
- Sweep the liquidity cutoff and lookback to map the cost/edge frontier. 2. Write the raw signal.
Implement `signal(close) -> wide DataFrame` in an alpha class. Higher values
should mean stronger long preference.
3. Choose the weighting method.
The default z-score is useful, but it can be fragile. Consider rank weights,
caps, neutralization, or liquidity-aware filters if outliers dominate.
4. Define the investable universe before trusting results.
Make sure the strategy is not depending on suspended, ST, newly listed, or
illiquid names.
5. Evaluate the alpha as a portfolio, not as a prediction.
Check cumulative return, Sharpe, drawdown, hit rate, and turnover. Do not add
IC/IR unless the framework's alpha convention changes.
6. Build the portfolio and inspect tracking.
Confirm that target weights match the alpha, then check whether integer
shares still track the target book.
7. Simulate execution with costs.
The final research question is not only "is there a signal?" It is "is there
enough signal left after realistic trading?"
8. Diagnose the failure layer.
If results are bad, identify whether the problem is the raw signal, weighting,
universe, construction, execution constraints, turnover, or cost.
For this 5-day reversal study, the diagnosis is clear: **the signal-level result
is promising only after robust weighting and a liquid universe filter, but the
current implementation needs turnover control before it can be considered
tradable.**
## Next Research Directions
The natural next experiments are:
- Add turnover control: no-trade bands, slower rebalancing, or weight smoothing.
- Sweep the lookback window: compare 3-day, 5-day, 10-day, and 20-day reversal.
- Sweep liquidity filters: top 500, top 1000, top 1500 by traded amount.
- Add position caps so no single name can dominate after normalization.
- Compare rank weighting with volatility-scaled reversal.
The most important habit is to keep the layers separate. A good alpha research
workflow does not stop at a single performance number; it explains how the idea
travels from hypothesis, to signal, to weights, to portfolio, to executable PnL.
scripts/generate_reversal_5d_report.py
+4 -4
View File
@@ -527,7 +527,7 @@ def render_report(results: dict, data_summary: dict, timings: dict,
]) ])
storage = _md_table( storage = _md_table(
["run", "schema ok", "null w", "non-finite w", "dup keys", ["run", "schema ok", "null w", "non-finite w", "dup keys",
"max |daily mean|", "weight range", "combo identity Δ"], "max \\|daily mean\\|", "weight range", "combo identity Δ"],
storage_rows, storage_rows,
) )
@@ -547,7 +547,7 @@ def render_report(results: dict, data_summary: dict, timings: dict,
_money(p["l1_tracking_mean"]), _money(p["l1_tracking_mean"]),
]) ])
closeness = _md_table( closeness = _md_table(
["run", "target_value identity max|Δ|", "alpha→target max|Δ|", ["run", "target_value identity max\\|Δ\\|", "alpha→target max\\|Δ\\|",
"research corr(alpha,portfolio)", "mean integer gross", "mean L1 tracking"], "research corr(alpha,portfolio)", "mean integer gross", "mean L1 tracking"],
close_rows, close_rows,
) )
@@ -647,8 +647,8 @@ constraints, 5 bps commission + 5 bps slippage.*
## 1. Are Alpha Values Properly Stored? ## 1. Are Alpha Values Properly Stored?
All alpha artifacts conform to `ALPHA_COLUMNS` (`symbol_id, date, alpha_name, All alpha artifacts conform to `ALPHA_COLUMNS` (`symbol_id`, `date`, `alpha_name`,
weight`), carry no null / non-finite weights, no duplicate `(symbol_id, date)` `weight`), carry no null / non-finite weights, no duplicate `(symbol_id, date)`
keys, and have numerically-zero daily cross-sectional means (weights are keys, and have numerically-zero daily cross-sectional means (weights are
demeaned per date). demeaned per date).