Caution-Flag Overreaction
An Upbit caution flag hits the KRW price harder than the global one; when the gap is extreme and the flag is lifted, it closes fast.
- SignalFlag + discount ≤ −5%
- EntryLong KRW · short perp
- ExitAbove −1% or 14d
How it works
Two legs, one spread · holds days – 2 weeks- cex_symbol_cautions12/hWhich coins are flagged and since when
- cex_announcements12/hLift, extension or delisting notice
- premium30/hKRW discount on the flagged coin
- forex12/hKRW conversion
- SignalFlag + discount ≤ −5%Caution or warning flag on Upbit or Bithumb, with the KRW premium
- EntryLong KRW · short perpOnly after a lift or extension notice, never on the flag itself
- ExitAbove −1% or 14dWhichever comes first
- RiskOut on delisting noticeStop at −12% premium; 2% of book per coin
The signal
ARDR · Upbit premium · flag → lift · sample| Date (KST) | Premium |
|---|---|
| Oct 19 | 0.2% |
| Oct 20 | 0.08% |
| Oct 21 | 0.16% |
| Oct 22 | 0.14% |
| Oct 23 | -0.08% |
| Oct 24 | -0.15% |
| Oct 25 | -0% |
| Oct 26 | 0.03% |
| Oct 27 | -0.01% |
| Oct 28 | -7.6% |
| Oct 29 | -7.3% |
| Oct 30 | -7.21% |
| Oct 31 | -7.27% |
| Nov 1 | -7.05% |
| Nov 2 | -6.67% |
| Nov 3 | -6.52% |
| Nov 4 | -6.38% |
| Nov 5 | -5.96% |
| Nov 6 | -5.62% |
| Nov 7 | -5.61% |
| Nov 8 | -5.55% |
| Nov 9 | -6.14% |
| Nov 10 | -3.99% |
| Nov 11 | -2.64% |
| Nov 12 | -1.55% |
| Nov 13 | -0.82% |
| Nov 14 | -0.49% |
| Nov 15 | -0.16% |
| Nov 16 | 0.23% |
| Nov 17 | 0.36% |
Sample backtest
Sample · illustrative 1Y · not real performance| Date (UTC) | Equity |
|---|---|
| Jul 21 | 0.23% |
| Jul 22 | 0.52% |
| Jul 23 | 0.4% |
| Jul 24 | 0.35% |
| Jul 25 | 0.65% |
| Jul 26 | 0.8% |
| Jul 27 | 1.17% |
| Jul 28 | 1.37% |
| Jul 29 | 1.07% |
| Jul 30 | 0.9% |
| Jul 31 | 0.59% |
| Aug 1 | 0.36% |
| Aug 2 | 0.58% |
| Aug 3 | 0.56% |
| Aug 4 | 0.48% |
| Aug 5 | 0.34% |
| Aug 6 | -0.21% |
| Aug 7 | -0.43% |
| Aug 8 | -0.44% |
| Aug 9 | -0.44% |
| Aug 10 | -0.05% |
| Aug 11 | 0.04% |
| Aug 12 | -0.09% |
| Aug 13 | -0.05% |
| Aug 14 | -0.21% |
| Aug 15 | 0.02% |
| Aug 16 | 0.57% |
| Aug 17 | 0.89% |
| Aug 18 | 1.35% |
| Aug 19 | 1.51% |
| Aug 20 | 1.37% |
| Aug 21 | 1.58% |
| Aug 22 | 1.74% |
| Aug 23 | 2.07% |
| Aug 24 | 2.61% |
| Aug 25 | 2.66% |
| Aug 26 | 2.63% |
| Aug 27 | 2.49% |
| Aug 28 | 2.12% |
| Aug 29 | 2.2% |
| Aug 30 | 2.31% |
| Aug 31 | 2.3% |
| Sep 1 | 2.41% |
| Sep 2 | 2.02% |
| Sep 3 | 1.57% |
| Sep 4 | 1.39% |
| Sep 5 | 1.12% |
| Sep 6 | 1.29% |
| Sep 7 | 1.52% |
| Sep 8 | 1.39% |
| Sep 9 | 1.4% |
| Sep 10 | 1.16% |
| Sep 11 | 0.97% |
| Sep 12 | 1.29% |
| Sep 13 | 1.56% |
| Sep 14 | 1.99% |
| Sep 15 | 2.42% |
| Sep 16 | 2.36% |
| Sep 17 | 2.46% |
| Sep 18 | 2.59% |
| Sep 19 | 2.73% |
| Sep 20 | 3.34% |
| Sep 21 | 3.73% |
| Sep 22 | 3.9% |
| Sep 23 | 4.06% |
| Sep 24 | 3.77% |
| Sep 25 | 3.68% |
| Sep 26 | 3.85% |
| Sep 27 | 3.87% |
| Sep 28 | 4.15% |
| Sep 29 | 4.13% |
| Sep 30 | 3.72% |
| Oct 1 | 3.48% |
| Oct 2 | 3.08% |
| Oct 3 | 2.91% |
| Oct 4 | 3.13% |
| Oct 5 | 3.08% |
| Oct 6 | 3.08% |
| Oct 7 | 2.94% |
| Oct 8 | 2.48% |
| Oct 9 | 2.45% |
| Oct 10 | 2.54% |
| Oct 11 | 2.74% |
| Oct 12 | 3.26% |
| Oct 13 | 3.39% |
| Oct 14 | 3.41% |
| Oct 15 | 3.53% |
| Oct 16 | 3.46% |
| Oct 17 | 3.84% |
| Oct 18 | 4.39% |
| Oct 19 | 4.71% |
| Oct 20 | 5.15% |
| Oct 21 | 5.16% |
| Oct 22 | 4.99% |
| Oct 23 | 5.14% |
| Oct 24 | 5.17% |
| Oct 25 | 5.45% |
| Oct 26 | 5.81% |
| Oct 27 | 5.66% |
| Oct 28 | 5.53% |
| Oct 29 | 5.21% |
| Oct 30 | 4.78% |
| Oct 31 | 4.85% |
| Nov 1 | 4.86% |
| Nov 2 | 4.85% |
| Nov 3 | 4.91% |
| Nov 4 | 4.47% |
| Nov 5 | 4.14% |
| Nov 6 | 4.02% |
| Nov 7 | 3.88% |
| Nov 8 | 4.24% |
| Nov 9 | 4.52% |
| Nov 10 | 4.52% |
| Nov 11 | 4.65% |
| Nov 12 | 4.49% |
| Nov 13 | 4.5% |
| Nov 14 | 4.95% |
| Nov 15 | 5.28% |
| Nov 16 | 5.82% |
| Nov 17 | 6.19% |
Generated to show the shape of the strategy, not measured on real fills. Ask your agent to backtest on DataMaxi+ history before trading.
Backtest this strategy
One prompt: your agent pulls the history below, tests these rules with fees and slippage, and charts the equity curve and drawdown. Run it before you trust the sample above.
- Period
- The last year of flags, 1-hour candles from the flag to 30 days after
- History calls
- 50
- Credits
- 210
- Cost per run
- ≈ $0.021
Backtest the "Caution-Flag Overreaction" strategy on DataMaxi+ history. Thesis: An Upbit caution flag hits the KRW price harder than the global one; when the gap is extreme and the flag is lifted, it closes fast. Use the DataMaxi+ MCP server at https://mcp.datamaxiplus.com/mcp. Pull the history once, cache it, and run the backtest in your own code sandbox (Python is fine). ## History to pull (split the period with from/to; never refetch a range you have) 1. cex_announcements(exchange="upbit,bithumb", category="notice", limit="100") — exchange notices, newest first; about 10 calls 2. cex_candle(exchange="upbit", market="spot", symbol="ARDR-KRW", interval="1h") — OHLCV, up to 1,000 candles per call, history from Jan 2024; about 20 calls 3. cex_candle(exchange="binance", market="futures", symbol="ARDR-USDT", currency="KRW", interval="1h") — OHLCV, up to 1,000 candles per call, history from Jan 2024; about 20 calls Universe: Upbit and Bithumb coins flagged caution or warning in the last year that also have a Binance perp Period: the last year of flags, 1-hour candles from the flag to 30 days after ## Rules to test Signal: New caution or warning flag on Upbit or Bithumb, plus the coin's KRW premium and the venue's notices. Entry: Flag active ≥ 24h, KRW premium ≤ −5%, and a notice that the flag is lifted or the review extended (not a delisting): long KRW spot, short the global perp. Exit: Premium back above −1%, or 14 days. Risk: Exit both legs at once on any delisting notice. Stop at −12% premium. Max 2% of book per coin. ## Assumptions Costs: 0.05% Upbit spot and 0.05% Binance perp taker per leg, 0.3% slippage per leg (flagged coins are thin) Data: Caution flags keep no history (cex_symbol_cautions shows today's), so rebuild flag, lift and extension dates from exchange notices, and the premium from candles (cex_candle with currency=KRW). Count every flag that ended in a delisting, not only the winners. Use only data stamped before each decision (no look-ahead). Say plainly where the data cannot test a rule. ## Report Net return, annualised return, Sharpe, max drawdown, number of trades, win rate and average hold, next to buy-and-hold over the same period. Chart the equity curve and the drawdown, and list the 5 worst trades. ## Budget About 50 history calls ≈ 210 credits ≈ $0.021 from my prepaid DataMaxi+ balance. Tell me how many calls you made. If DataMaxi+ is not connected, ask me to [connect DataMaxi+](https://claude.ai/customize/connectors?modal=add-custom-connector) using https://mcp.datamaxiplus.com/mcp before continuing.
Estimate from the call counts in the prompt. Your agent may pull less or more; it reports the calls it made.
Rules
| Signal | New caution or warning flag on Upbit or Bithumb, plus the coin's KRW premium and the venue's notices. |
|---|---|
| Entry | Flag active ≥ 24h, KRW premium ≤ −5%, and a notice that the flag is lifted or the review extended (not a delisting): long KRW spot, short the global perp. |
| Exit | Premium back above −1%, or 14 days. |
| Risk | Exit both legs at once on any delisting notice. Stop at −12% premium. Max 2% of book per coin. |
Data used
≈ $0.1584 / day| MCP tool · REST route | Dataset | Tier | Calls / h | Credits / day | USD / day |
|---|---|---|---|---|---|
cex_symbol_cautions GET /cex/symbol/cautions | Listings | — · 1 cr | 12 | 288 | $0.0288 |
cex_announcements GET /cex/announcements | Listings | Latest · 1 cr | 12 | 288 | $0.0288 |
premium GET /premium | Premium | Latest · 1 cr | 30 | 720 | $0.072 |
forex GET /forex | Index price · forex | Latest · 1 cr | 12 | 288 | $0.0288 |
| Total | 66 | 1,584 | $0.1584 |
1 credit = $0.0001. Only successful calls are billed, from one balance shared by the API and MCP. Calls per hour are the suggested cadence; scale them to the symbols you watch.
Build it with your agent
Connect the MCP server →Open it in Claude with one click (it asks you to add the DataMaxi+ connector first if you haven't), or paste it into Cursor or any MCP client. It backtests first, then runs in paper mode and never places orders on its own.
Build the "Caution-Flag Overreaction" strategy for me with DataMaxi+ data. Thesis: An Upbit caution flag hits the KRW price harder than the global one; when the gap is extreme and the flag is lifted, it closes fast. Connect to the DataMaxi+ MCP server at https://mcp.datamaxiplus.com/mcp (header X-DTMX-APIKEY, read from env DATAMAXI_API_KEY). ## Data (call these tools; arguments are examples — loop over the symbols and venues you need) 1. cex_symbol_cautions(exchange="upbit,bithumb", min_level="caution", active_only="true") — Which coins are flagged and since when; about 12×/hour 2. cex_announcements(exchange="upbit,bithumb", limit="20") — Lift, extension or delisting notice; about 12×/hour 3. premium(asset="ARDR", source_exchange="binance", target_exchange="upbit") — KRW discount on the flagged coin; about 30×/hour 4. forex(symbol="USD-KRW") — KRW conversion; about 12×/hour ## Rules Signal: New caution or warning flag on Upbit or Bithumb, plus the coin's KRW premium and the venue's notices. Entry: Flag active ≥ 24h, KRW premium ≤ −5%, and a notice that the flag is lifted or the review extended (not a delisting): long KRW spot, short the global perp. Exit: Premium back above −1%, or 14 days. Risk: Exit both legs at once on any delisting notice. Stop at −12% premium. Max 2% of book per coin. ## What to do 1. Build the signal from the tools above and backtest the rules on the history you can pull; report trades, win rate, average trade and max drawdown. 2. Then run it live in paper mode and show me flagged coins with days under flag, KRW discount and the latest notice, refreshed on the cadence above. 3. Never place real orders. Propose trades and wait for my confirmation. ## Budget Every call is billed from my prepaid DataMaxi+ balance. Expected load: ~66 calls/hour ≈ 1,584 credits/day (≈ $0.16/day). Cache results, never repeat a call inside its refresh window, and print the number of calls each refresh made. If DataMaxi+ is not connected, ask me to [connect DataMaxi+](https://claude.ai/customize/connectors?modal=add-custom-connector) using https://mcp.datamaxiplus.com/mcp before continuing.
Risk notes
- Some flags end in a delisting; then the KRW leg can gap 30%+ and the perp hedge does not cover it.
- Flagged coins often lose deposit access, so the premium cannot be arbitraged closed by others — or by you.
- Very few signals a year; one delisting can erase every other win.
Educational example, not investment advice. Sample charts and figures are illustrative.
Keep exploring
Live screens, dashboards and the data behind this strategy.