add momentum indicator

momentum_indicator.ipynb

@@ -0,0 +1,279 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "<div style=\"float:right; width:100px; text-align: center; margin: 10px;\">\n",
+    "<img src=\"https://crypto-lake.com/assets/img/lake.png\" alt=\"Lake\"/>\n",
+    "</div>\n",
+    "\n",
+    "# Momentum indicator\n",
+    "\n",
+    "Backtest of a simple momentum indicator on 1m candle data. The logic was proposed by @BeatzXBT on twitter.\n",
+    "\n",
+    "We use [crypto-lake.com](https://crypto-lake.com/#data) sample/free market data, FTRB-USDT market on Ascendex.\n",
+    "\n",
+    "Quick links:\n",
+    "- [edit this notebook online](https://mybinder.org/v2/gh/crypto-lake/analysis-sharing/main?filepath=momentum_indicator.ipynb) using Binder\n",
+    "- [follow our activity on twitter](https://twitter.com/intent/user?screen_name=crypto_lake_com)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import datetime\n",
+    "\n",
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "import cufflinks as cf\n",
+    "import plotly.express as px\n",
+    "import statsmodels.api as sm\n",
+    "\n",
+    "import lakeapi\n",
+    "\n",
+    "cf.go_offline()\n",
+    "\n",
+    "# This runs on the paid data only at the moment\n",
+    "# lakeapi.use_sample_data(anonymous_access=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Parameters\n",
+    "symbol = 'ADA-USDT'\n",
+    "exchange = 'BINANCE'\n",
+    "\n",
+    "# Free sample data contain subset of the below time period\n",
+    "start = datetime.datetime(2023, 8, 1)\n",
+    "end = datetime.datetime(2023, 9, 1)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Data"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print('Loading candles')\n",
+    "candles = lakeapi.load_data(\n",
+    "    table = 'candles',\n",
+    "    start = start,\n",
+    "    end = end,\n",
+    "    symbols = [symbol],\n",
+    "    exchanges = [exchange],\n",
+    "    drop_partition_cols = True,\n",
+    ").sort_values('origin_time')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Minimize the data\n",
+    "df = candles[['origin_time', 'close']]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Logic"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "depths = np.asarray([200, 100, 50, 25, 10]) # in minutes\n",
+    "n = len(depths)\n",
+    "# depths *= 10 # try a longer time frame?\n",
+    "\n",
+    "# Signal logic\n",
+    "for depth in depths:\n",
+    "\tdf[f'ema_{depth}'] = np.log(df['close'] / df['close'].ewm(span=depth).mean()) * 100\n",
+    "df['trend_val'] = df[[f'ema_{depth}' for depth in depths]].ewm(span=n, axis = 1).mean().sum(axis=1)\n",
+    "df['pure_sum'] = df[[f'ema_{depth}' for depth in depths]].sum(axis=1)\n",
+    "\n",
+    "\n",
+    "# Future return for evaluation\n",
+    "df['future_return'] = df['close'].pct_change(10).shift(-10) * 100\n",
+    "df = df.dropna()\n",
+    "\n",
+    "df[::3600].head(20)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Evaluation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "df.set_index('origin_time')[['close', 'trend_val']].iplot(secondary_y='trend_val')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# df.sample(1000).iplot(kind='scatter', x='trend_val', y='future_return', mode='markers', size=3)\n",
+    "px.scatter(df.sample(10_000), x='trend_val', y='future_return', trendline='ols')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional: install statsmodels for the next cell to run\n",
+    "!pip install -q statsmodels"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Numerical linreg fit statistics\n",
+    "mod = sm.OLS(df['trend_val'], df['future_return'])\n",
+    "res = mod.fit()\n",
+    "print(res.summary())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Conclusion\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "It seems the trend value has a slight negative correlation with future trend on our data. I don't consider the correlation significant enough for this signal to be used in real trading.\n",
+    "\n",
+    "I also tried different settings, mostly longer ewm spans, but the results were similar."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "---\n",
+    "\n",
+    "### For reference: the original logic"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# This seems to be equivalent to pandas Series.ewm(span=window).mean()\n",
+    "def ema(arr_in: np.ndarray, window: int) -> np.ndarray:\n",
+    "    \"\"\"\n",
+    "    Hyper-fast EMA implementation\n",
+    "    \"\"\"\n",
+    "    \n",
+    "    n = arr_in.shape[0]\n",
+    "    ewma = np.empty(n, dtype=float)\n",
+    "    alpha = 2 / float(window + 1)\n",
+    "    w = 1\n",
+    "    ewma_old = arr_in[0]\n",
+    "    ewma[0] = ewma_old\n",
+    "\n",
+    "    for i in range(1, n):\n",
+    "        w += (1-alpha)**i\n",
+    "        ewma_old = ewma_old*(1-alpha) + arr_in[i]\n",
+    "        ewma[i] = ewma_old / w\n",
+    "        \n",
+    "    return ewma\n",
+    "\n",
+    "# This is replicated in the cell below in a batch/more-efficient way\n",
+    "def trend_feature(klines: np.ndarray, lengths: np.ndarray) -> float:\n",
+    "    \"\"\"\n",
+    "    Make sure lengths are fed in from longest to shortest\n",
+    "    \"\"\"\n",
+    "\n",
+    "    closes = klines[:, 4]\n",
+    "    curr_price = closes[-1]\n",
+    "    n = len(lengths)\n",
+    "    vals = np.empty(n, dtype=float)\n",
+    "\n",
+    "    for i in range(n):\n",
+    "        length = lengths[i]\n",
+    "        ema_val = ema(closes[-length:], length)[-1]\n",
+    "\n",
+    "        # Safety check\n",
+    "        if ema_val == 0:\n",
+    "            vals[i] = vals[i-1]   \n",
+    "\n",
+    "        else:\n",
+    "            vals[i] = np.log(curr_price / ema_val) * 100\n",
+    "\n",
+    "    trend_val = ema(vals, n)\n",
+    "\n",
+    "    return np.sum(trend_val)\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3.10.6 ('env': venv)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.4"
+  },
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "7291192e6945a3c8ed0f956f4124ad6de62ed7ea9152f7ae8e29637075883be3"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

requirements.txt

@@ -2,7 +2,7 @@ setuptools==59.6.0
 numpy==1.23.4
 pandas==1.5.1
 cufflinks==0.17.3
-lakeapi==0.5.0
+lakeapi==0.6.3
 notebook==6.5.4
 scipy==1.11.1
 kaleido==0.2.1