01
Factor research
FACTOR RESEARCHStart from an economic mechanism, then test whether the signal is robust, explainable, and still worth trading after costs.
OUT-OF-SAMPLE QUANTITATIVE RESEARCH
Test hypotheses
beyond the
sample.
Ziju Quant studies signals that could be traded. A hypothesis has to be stated, then tested on data that was not used to build it, after costs, and against rules that would retire it. One backtest is not a conclusion.
- HypothesisState the question
- PerturbPerturb inputs
- Out-of-sampleHold out data
- ReviewReview the record
RESEARCH MODEMultiple tests
RISK LAYERInvalidation first
REVIEW STATEOpen to reversal
Research topics: FACTOR RESEARCH, ROBUSTNESS, RISK BUDGET, PORTFOLIO CONSTRUCTION, EXECUTION RESEARCH, MODEL REVIEW
01
OUR METHOD
Our method
ON BOOTSTRAP
One backtest is not a conclusion.
We examine where signals may fail.
Here, bootstrap is a research metaphor. Resampling can help assess estimation uncertainty, but it cannot replace a time-ordered out-of-sample test. Candidate signals also need a separate check for costs and invalidation rules.
Fitting history is not enough. The evidence that would retire a signal has to be written down first. Parameters are not retuned to suit a result. Signals held together are checked for the same underlying risk.
02
CORE CAPABILITIES
Capabilities
FOUR RESEARCH AREAS
State the question before managing the risk.
02
Portfolio design
PORTFOLIO DESIGNCheck whether the signals share the same risk, then set weights under correlation and risk-budget constraints.
03
Execution research
EXECUTION RESEARCHCosts, capacity, and liquidity are included before anyone treats a frictionless backtest as a result.
04
Risk framework
RISK FRAMEWORKData, models, the portfolio, and execution each have monitors. The action after a breach is written down in advance.
03
RESEARCH LOOP
Research process
A RECORD FOR EVERY CHANGE
Each change
needs a record.
Hypothesis, out-of-sample testing, portfolio construction, and review are fixed steps. The discussion follows the record, not a single equity curve.
01→
Frame a hypothesis
The question has to be falsifiable, and it has to come from market structure or behavioral bias.
02→
Test out of sample
Hold out time-ordered data that was not used in development, then check whether the result remains under rolling windows and stress scenarios.
03→
Set portfolio weights
Control correlation and shared exposure so several signals are not just the same risk counted again.
04↺
Review continuously
Record drift, structural change, decay, and invalidation rules when they appear, not after the fact.
04
RESEARCH CHECKS
Research checks
SIX QUESTIONS BEFORE A CONCLUSION
Before the conclusion,
answer six questions.
We do not look only for evidence that supports a conclusion. Disconfirming evidence, the boundary of use, and the exit rule are written down before the conclusion.
01MECHANISMLogic
Logic
Why might the signal exist? Could returns reflect risk premia, behavioral bias, or market constraints?
REVIEW FOCUSFalsifiable premise
02DATA INTEGRITYData
Data
Can data versions, true availability times, and missing-value treatment be reproduced independently?
REVIEW FOCUSPoint-in-time integrity
03ROBUSTNESSValidation
Validation
Does the result survive out-of-sample, rolling-window, parameter, and alternative-sample tests?
REVIEW FOCUSCross-sample stability
04IMPLEMENTATIONTrading
Trading
After costs, slippage, market impact, and capacity constraints, does the premise still hold?
REVIEW FOCUSTradability assumptions
05EXPOSURERisk
Risk
Which exposures drive returns, and what happens under stress and extreme scenarios?
REVIEW FOCUSBounded exposure
06INVALIDATIONInvalidation
Invalidation
Which observable evidence triggers a lower weight, renewed validation, or retirement?
REVIEW FOCUSExplicit exit rules
05
RISK FIRST
Risk first
LIMITS, THEN A RESPONSE
Risk needs measurement,
and clear response rules.
Data quality, model stability, portfolio exposure, liquidity, and execution drift each need a boundary and a pre-written action for when that boundary is crossed.
Diversify return sourcesInspect tail exposureDetect model driftPreserve liquidity headroom
RISK REVIEW LAYERSILLUSTRATIVE FRAMEWORK · 04 LAYERS
- 01DataPoint-in-time integrity
- 02ModelCross-sample stability
- 03PortfolioBounded exposure
- 04ExecutionTradability assumptions
Illustrative · Not a rating
06
RESEARCH NOTES
Research notes
LATEST NOTES
Research notes
Notes and discussions about research method.
Combinatorial Purged Cross-Validation (CPCV) and Label-Overlap Leakage Defense
Financial time series exhibit serial correlation and overlapping event labels. Standard k-fold cross-validation causes severe information leakage; purging overlapping samples and embargoing post-test autocorrelation are required to construct unbiased evaluation distributions.
Point-in-time data integrity and the defense against lookahead bias
Leaking future information into backtests is the most covert cause of live trading failure. Maintaining point-in-time data and tracking information availability timestamps is a non-negotiable quantitative baseline.
From backtest curves to live execution: slippage and capacity limits
Theoretical backtest gains quickly vanish in the frictions of live execution. Embedding market impact, turnover constraints, and liquidity ceilings directly into the research model is what keeps strategies viable.
RESEARCH INQUIRIES
Write the question
before we schedule a call.
To discuss research, data, engineering, or a collaboration, leave your name, email, and the specific question. We use that information only to reply to this inquiry.
contact@ziju.cn