Backtest five dividend growers with ChatGPT

TL;DR

ChatGPT can reproduce the five dividend growers test through a NexusTrade connection that permits portfolio authoring and backtests. The recorded January 2025 to October 2026 simulation returned 44.80% versus SPY's 33.99%, with 24.33% drawdown. The longer January 2014 to October 2026 run returned 340.90% versus 419.23% for SPY. Both reinvest dividends; the strategy pays 0.1% stock transaction fees and the stored SPY benchmark pays no candidate commissions.

Explore the completed 2025 through 2026 test

Strategy return

44.80%

SPY return

33.99%

Maximum drawdown

24.33%

Strategy fees

$24.15

Recorded historical simulation: Jan 02, 2025 to Oct 07, 2026. Initial capital was $10,000.

Recorded in NexusTrade on October 8, 2026. Strategy assumptions: fractional shares, reinvested dividends and a 0.1% stock transaction fee. The stored SPY benchmark pays no candidate commissions.

Try this example in ChatGPT

Connect ChatGPT, then copy the request below. Review the returned result before continuing.

The rules behind this result

Allocation trigger

Review allocations when portfolio value equals its initial value OR at least 30 days have passed since a filled buy. This is not an exact monthly schedule.

Equal weights

Each selected company has the same weight: five qualifying companies would target 20% each before execution effects.

Execution assumptions

Fractional shares and reinvested dividends; the original stock fee was 0.1% per transaction.

From the S&P 500 universe, keep companies meeting the ten-year annual dividend-increase filter, select the five largest by market value and weight them equally. The target is 100% invested.

Create and read back the exact strategy

Use the reviewed configuration at https://nexustrade-prod.nyc3.cdn.digitaloceanspaces.com/Explore/Data/existing-dividend-five-configuration-20261010-2468740bdaf783a8.json. Ask your client to pass that complete object to build_portfolio, inspect valid and issues, then show the canonical portfolio. After reviewing it, save the same JSON with create_portfolio and read the returned chat draft with get_portfolio. Keep the draft ID in your experiment record.

Check the SP500 universe, dividend filter, market-cap ranking, five selected names, equal weights, full deployment, 30-day trigger, fractional shares and reinvestment against the reviewed configuration. Keep the new draft undeployed while testing.

Reproduce the five-dividend-growers experiment in ChatGPT. Read the exact configuration at https://nexustrade-prod.nyc3.cdn.digitaloceanspaces.com/Explore/Data/existing-dividend-five-configuration-20261010-2468740bdaf783a8.json. Pass the complete configuration to build_portfolio. Show valid, issues and canonical rules. Preserve the universe, filter, ranking, weighting, rebalance trigger, fractional shares and dividend policy. After I approve the rules, save the same JSON with create_portfolio, then read the returned portfolio_id with get_portfolio. Record that chat draft ID. Stop before backtesting or deployment.

Submit the matching historical test

Continue only when the connected client has the required authoring and backtest tools. Review the dates, capital, fee and dividend assumptions, and research-token cost before approving one test.

Replace the placeholder with the returned chat draft ID. Review these inputs, then approve one backtest_portfolio call. They match the displayed 2025 run. Save the accepted backtest ID before leaving the conversation.

generate_events is false for the curve comparison. Enable it when you want to inspect individual signals and orders; detailed event capture uses more credits. Keep all other settings fixed when making that inspection run.

{
  "portfolio_id": "REPLACE_WITH_RETURNED_CHAT_ID",
  "start_date": "2025-01-02",
  "end_date": "2026-10-07",
  "interval": "Day",
  "initial_value": 10000,
  "baseline_symbol": "SPY",
  "generate_events": false,
  "dividend_policy": "reinvest",
  "fee_config": {
    "Stock": {
      "amount": 0.1,
      "type": "percent"
    }
  }
}

Tool reference

Inspect the completed curve and explain a rebalance

Poll query_backtest_status with the accepted backtest_id. PENDING or RUNNING means continue with that same ID. Resolve a reported ERROR before changing the experiment. Once complete, read query_backtest_history and follow pagination for the full period. Save the completed statistics and history with the rules and inputs.

Compare value with comparisonValue, then follow cash and positionValue around a rebalance or drawdown. With detailed events enabled, ask the client to read query_backtest_events and explain which condition fired, which stocks qualified, how weights changed and which orders filled. Keep the explanation tied to returned records.

1. Detailed signal traces

   The original example used generate_events:false. A new generate_events:true test costs 5× research tokens; review that request before submitting it and read query_backtest_events while its three-day traces are retained.

Use the accepted backtest_id to poll query_backtest_status until complete. Then read query_backtest_history with that same ID, page 1 and page_size 500; follow all pages. Show the strategy and SPY curves, return, maximum drawdown and fees for 2025-01-02 through 2026-10-07. Explain one rebalance or drawdown using cash, holdings and recorded order events when available. Preserve the returned IDs and source fields in the experiment record. Keep polling the existing test rather than submitting another.

Compare the other recorded periods

The same rules also have completed runs starting in 2022 and 2014. Compare the full period table before selecting a variant; the long 2014 run returned less than SPY.

Inspect the longer recorded period

Choose a ChatGPT connection that supports this task

On ChatGPT web, open Plugins, select Add, then Add custom MCP server. Enter NexusTrade as the name and https://nexustrade.io/api/mcp as the Server URL. Select OAuth, review the server notice, check I understand and want to continue, then choose Create as a plugin. Complete NexusTrade sign-in and review the requested permissions before allowing access.

Open the new plugin’s details and choose Try in chat. The composer should show NexusTrade before you send the connection check. Select or mention the plugin again when a later message needs another tool call. A regular chat without the plugin cannot use the connection.

If your workspace shows the legacy Apps controls, use the official developer-mode instructions for its account roles. Available tools and write actions depend on the plan and workspace policy. A public read does not establish permission to run paid research or change a portfolio.

Hosted MCP endpoint

https://nexustrade.io/api/mcp

Desktop and plugins use different connections

ChatGPT web and local Codex tasks use separate connections. For the custom MCP workflow described here, use ChatGPT web. In the Codex desktop app, open Settings, MCP servers and Add server; choose Streamable HTTP, enter the NexusTrade endpoint, Save, Restart and Authenticate. Check /mcp in that local task. Use the separate Codex guide for its CLI setup. Native ChatGPT desktop plugin availability must be checked in that app; its ordinary chats do not inherit local Codex MCP settings.

Check the returned account tools

The result is a paginated object with portfolios, total and totalPages. Walk pages when necessary. A new account may return an empty list. Keep authentication errors and missing results separate.

Use fetch_portfolios with the JSON below. Read a portfolioId returned by that call with get_portfolio, setting its portfolio_id argument to the same value. Compare the name, type and strategies with the intended NexusTrade account. If no portfolio exists, keep the empty result and continue to the new draft preview below. Do not create anything during this check.

{
  "include_paper": true,
  "include_live": false,
  "include_chat_portfolios": true,
  "include_positions": false,
  "limit": 20,
  "page": 1
}

Confirm permission for this task

Discover build_portfolio, create_portfolio, get_portfolio and backtest_portfolio, query_backtest_status and query_backtest_history in the current connection. Check that the granted app actions allow this task before submitting anything. A plugin listing or successful read does not grant write access. On ChatGPT web, Pro custom MCP is read/fetch only; use authorized Business or Enterprise/Edu write actions for these steps.

Create one variant, then test unseen dates

Save a separate variant that changes only SelectTop.limit from five to ten. Keep the universe, dividend filter, weights, capital, fees, interval and development dates fixed. Record both configurations and their completed backtest IDs before comparing return, drawdown, holdings and turnover.

After choosing a candidate, freeze its configuration and decision criteria. The three displayed historical periods have already been inspected. Select another chronological period you have not used in the example, corpus search or variant selection, then evaluate the frozen candidate once. Keep further revisions in a new research cycle. An optional final lockbox follows the same rule.

Keep the recorded experiment

Save the exact configuration, draft ID, accepted backtest ID, inputs and completed result together. Reopen that record after a connection interruption and continue polling the accepted run. Before paper trading, review a separate paper draft and observe its actual deployed history.

Exact strategy configuration

{
  "initialValue": 10000,
  "supportsFractionalShares": true,
  "dividendPolicy": "reinvest",
  "strategies": [
    {
      "active": true,
      "condition": {
        "lhs": null,
        "rhs": null,
        "comparison": null,
        "type": "Or",
        "conditions": [
          {
            "lhs": {
              "compound": false,
              "targetAssets": [],
              "type": "PortfolioValue"
            },
            "rhs": {
              "compound": false,
              "targetAssets": [],
              "type": "InitialValue"
            },
            "comparison": "equal",
            "type": "Base"
          },
          {
            "lhs": {
              "compound": false,
              "targetAssets": [],
              "side": "Buy",
              "type": "DaysSinceOrder",
              "orderStatus": "Filled"
            },
            "rhs": {
              "compound": false,
              "targetAssets": [],
              "value": 30,
              "type": "Value"
            },
            "comparison": "greaterThanOrEqual",
            "type": "Base"
          }
        ]
      },
      "action": {
        "type": "DynamicRebalance",
        "universe": {
          "source": "SP500"
        },
        "pipeline": [
          {
            "type": "Filter",
            "condition": {
              "lhs": {
                "compound": true,
                "targetAssets": [],
                "type": "ConsecutiveIncrease",
                "maxLength": 1,
                "period": "Year",
                "lookback": 10,
                "indicators": [
                  {
                    "compound": false,
                    "targetAsset": {
                      "type": "",
                      "symbol": ""
                    },
                    "targetAssets": [],
                    "type": "LastDividend"
                  }
                ]
              },
              "rhs": {
                "compound": false,
                "targetAssets": [],
                "value": 10,
                "type": "Value"
              },
              "comparison": "greaterThanOrEqual",
              "type": "Base"
            }
          },
          {
            "type": "SelectTop",
            "metric": {
              "compound": false,
              "targetAsset": {
                "type": "",
                "symbol": ""
              },
              "targetAssets": [],
              "type": "Fundamental",
              "metric": "marketCap"
            },
            "limit": 5,
            "direction": "Highest"
          }
        ],
        "weightIndicator": {
          "compound": false,
          "targetAssets": [],
          "value": 1,
          "type": "Value"
        },
        "deploymentPercent": {
          "compound": false,
          "targetAssets": [],
          "value": 100,
          "type": "Value"
        }
      },
      "orderExecution": {
        "type": "Market"
      },
      "automaticOrderApproval": false
    }
  ],
  "name": "Five largest dividend growers",
  "cash": 10000,
  "buyingPower": 10000,
  "supportsCrypto": false,
  "alertsEnabled": false
}

Chart range diagnostics

Continue exploring