TradingAgents/docs/iterations/hypotheses/active.json

45 lines
1.8 KiB
JSON

{
"max_active": 5,
"hypotheses": [
{
"id": "insider_buying-min-txn-100k",
"scanner": "insider_buying",
"title": "Raise min_transaction_value to $100K",
"description": "Hypothesis: filtering to insider purchases \u2265$100K (vs. current $25K) produces higher-quality picks by excluding routine small-lot grants and focusing on high-conviction, out-of-pocket capital deployment. Research (Lakonishok & Lee 2001; Cohen et al. 2012) shows large-value insider buys predict forward returns; small ones do not.",
"branch": "hypothesis/insider_buying-min-txn-100k",
"pr_number": 529,
"status": "running",
"priority": 3,
"expected_impact": "medium",
"hypothesis_type": "implementation",
"created_at": "2026-04-10",
"min_days": 21,
"days_elapsed": 3,
"picks_log": [
"2026-04-10",
"2026-04-13",
"2026-04-14"
],
"baseline_scanner": "insider_buying",
"conclusion": null
},
{
"id": "social_dd-ranker-suppression",
"scanner": "social_dd",
"title": "Does ranker suppression cause us to miss social_dd 30d winners?",
"description": "social_dd shows 60% 30d win rate (+2.32% avg) but only 41.7% 7d (-1.92%). Hypothesis: the ranker and recommendation system evaluate at 7d horizon, unfairly penalizing a slow-win scanner. Most picks (22/25) already score >=65, so score suppression is not the primary issue \u2014 horizon mismatch is.",
"branch": null,
"pr_number": null,
"status": "concluded",
"priority": 0,
"expected_impact": "medium",
"hypothesis_type": "statistical",
"created_at": "2026-04-13",
"min_days": 0,
"days_elapsed": 0,
"picks_log": [],
"baseline_scanner": "social_dd",
"conclusion": "statistical"
}
]
}