benchmark.py

from datetime import datetime
from tqdm import tqdm

from analysis import Analysis
from trading import Trading
from twitter import Twitter

# The initial amount in dollars for the fund simulation.
FUND_DOLLARS = 100000

# The fee in dollars per trade (https://www.tradeking.com/rates).
TRADE_FEE = 0


def format_ratio(ratio):
    """Converts a ratio to a readable percentage gain."""

    return '%.3f%%' % (100 * (ratio - 1))


def format_dollar(amount):
    """Converts a dollar amount into a readable string."""

    return '${:,.2f}'.format(amount)


def format_timestamp(timestamp, weekday=False):
    """Converts a timestamp into a readable string."""

    date_format = '%-m/%-d/%Y %-I:%M %p'
    if weekday:
        date_format += ' (%A)'
    return timestamp.strftime(date_format)


def get_ratio(strategy):
    """Calculates the profit ratio of a strategy."""

    price_at = strategy['price_at']
    price_eod = strategy['price_eod']
    if price_at and price_eod:
        action = strategy['action']
        if action == 'bull':
            return price_eod / price_at
        elif action == 'bear':
            return price_at / price_eod
        else:
            return 1.0
    else:
        return 1.0


def get_sentiment_emoji(sentiment):
    """Returns an emoji representing the sentiment score."""

    if sentiment == 0:
        return ':neutral_face:'
    elif sentiment > 0:
        return ':thumbsup:'
    else:  # sentiment < 0:
        return ':thumbsdown:'


def get_market_status(timestamp):
    """Tries to infer the market status from a timestamp."""

    if not trading.is_trading_day(timestamp):
        return 'closed'

    # Calculate the market hours for the given day. These are the same for NYSE
    # and NASDAQ and include TradeKing's extended hours.
    pre_time = timestamp.replace(hour=8)
    open_time = timestamp.replace(hour=9, minute=30)
    close_time = timestamp.replace(hour=16)
    after_time = timestamp.replace(hour=17)

    # Return the market status for each bucket.
    if timestamp >= pre_time and timestamp < open_time:
        return 'pre'
    elif timestamp >= open_time and timestamp < close_time:
        return 'open'
    elif timestamp >= close_time and timestamp < after_time:
        return 'after'
    else:
        return 'closed'


# TODO: Refactor trading so this logic can live there.
def should_trade(strategy, date, previous_trade_date):
    """Determines whether a trade is happening for the strategy."""

    # We invest the whole value, so we can only trade once a day.
    if (previous_trade_date and
        previous_trade_date.replace(hour=0, minute=0, second=0) ==
            date.replace(hour=0, minute=0, second=0)):
        return False

    # The strategy needs to be active.
    if strategy['action'] == 'hold':
        return False

    # We need to know the stock price.
    if not strategy['price_at'] or not strategy['price_eod']:
        return False

    return True


if __name__ == '__main__':
    analysis = Analysis(logs_to_cloud=False)
    trading = Trading(logs_to_cloud=False)
    twitter = Twitter(logs_to_cloud=False)

    # Look up the metadata for the tweets.
    tweets = twitter.get_all_tweets()

    events = []
    for tweet in tqdm(tweets):
        event = {}

        timestamp_str = tweet['created_at']
        timestamp = trading.utc_to_market_time(datetime.strptime(
            timestamp_str, '%a %b %d %H:%M:%S +0000 %Y'))
        text = twitter.get_tweet_text(tweet)
        event['timestamp'] = timestamp
        event['text'] = text
        event['link'] = twitter.get_tweet_link(tweet)

        # Extract the companies.
        companies = analysis.find_companies(tweet)

        strategies = []
        for company in companies:

            # What would have been the strategy?
            market_status = get_market_status(timestamp)
            strategy = trading.get_strategy(company, market_status)

            # What was the price at tweet and at EOD?
            price = trading.get_historical_prices(
                company['ticker'], timestamp)
            if price:
                strategy['price_at'] = price['at']
                strategy['price_eod'] = price['eod']
            else:
                strategy['price_at'] = None
                strategy['price_eod'] = None

            strategies.append(strategy)

        event['strategies'] = strategies

        events.append(event)

    # Make sure the events are ordered by ascending timestatmp.
    events = sorted(events, key=lambda event: event['timestamp'])

    # Print out the formatted benchmark results as markdown.
    print('## Benchmark Report')
    print()
    print('This breakdown of the analysis results and market performance vali'
          'dates the current implementation against historical data.')
    print()
    print('Use this command to regenerate the benchmark report after changes '
          'to the algorithm or data:')
    print('```shell')
    print('python benchmark.py > benchmark.md')
    print('```')

    print()
    print('### Events overview')
    print()
    print('Here\'s each tweet with the results of its analysis and individual '
          'market performance.')

    for event in events:
        strategies = event['strategies']

        if strategies:
            timestamp = format_timestamp(event['timestamp'], weekday=True)
            print()
            print('##### [%s](%s)' % (timestamp, event['link']))
            print()
            lines = ['> %s' % line for line in event['text'].split('\n')]
            print('\n\n'.join(lines))
            print()
            print('*Strategy*')
            print()
            print('Company | Root | Sentiment | Strategy | Reason')
            print('--------|------|-----------|----------|-------')

            for strategy in strategies:
                root = '-' if 'root' not in strategy else strategy['root']
                sentiment = strategy['sentiment']
                sentiment_emoji = get_sentiment_emoji(sentiment)
                print('%s | %s | %.1f %s | %s | %s' % (
                    strategy['name'],
                    root,
                    sentiment,
                    sentiment_emoji,
                    strategy['action'],
                    strategy['reason']))

            print()
            print('*Performance*')
            print()
            print('Ticker | Exchange | Price @ tweet | Price @ close | Gain')
            print('-------|----------|---------------|---------------|-----')

            for strategy in strategies:
                price_at = strategy['price_at']
                price_eod = strategy['price_eod']
                if price_at and price_eod:
                    price_at_str = format_dollar(price_at)
                    price_eod_str = format_dollar(price_eod)
                else:
                    price_at_str = '-'
                    price_eod_str = '-'
                ratio = get_ratio(strategy)
                gain = format_ratio(ratio)
                print('%s | %s | %s | %s | %s' % (
                    strategy['ticker'],
                    strategy['exchange'],
                    price_at_str,
                    price_eod_str,
                    gain))

    print()
    print('### Fund simulation')
    print()
    print(('This is how an initial investment of %s would have grown, includin'
           'g fees of 2 \u00d7 %s per pair of orders. Bold means that the data'
           ' was used to trade.') % (
               format_dollar(FUND_DOLLARS), format_dollar(TRADE_FEE)))
    print()
    print('Time | Trade | Gain | Value | Return | Annualized')
    print('-----|-------|------|-------|--------|-----------')
    start_date = events[0]['timestamp']
    value = FUND_DOLLARS
    print('*Initial* | - | - | *%s* | - | -' % format_dollar(value))

    previous_trade_date = None
    for event in events:
        date = event['timestamp']
        strategies = event['strategies']

        # Figure out what to spend on each trade.
        num_actionable_strategies = sum(
            [1 for strategy in strategies if should_trade(
                strategy, date, previous_trade_date)])
        budget = trading.get_budget(value, num_actionable_strategies)

        trade = False
        for strategy in strategies:
            trade = should_trade(strategy, date, previous_trade_date)

            price_at = strategy['price_at']
            price_eod = strategy['price_eod']

            if trade:
                # Use the price at tweet to determine stock quantity.
                quantity = int(budget // price_at)

                if quantity:
                    # Pay the fees for both trades.
                    value -= 2 * TRADE_FEE

                    # Calculate the returns depending on the strategy.
                    if strategy['action'] == 'bull':
                        value -= quantity * price_at  # Buy
                        value += quantity * price_eod  # Sell
                    elif strategy['action'] == 'bear':
                        value += quantity * price_at  # Short
                        value -= quantity * price_eod  # Cover
                else:
                    # Not enough budget to buy even one share.
                    trade = False
            else:
                quantity = 0

            total_ratio = value / FUND_DOLLARS
            total_return = format_ratio(total_ratio)

            if date != start_date:
                days = (date - start_date).days
                if days > 0:
                    annualized_ratio = pow(total_ratio, 365.0 / days)
                else:
                    annualized_ratio = 1
                annualized_return = format_ratio(annualized_ratio)
            else:
                annualized_return = '-'

            date_str = format_timestamp(date)
            trade_str = '%s %s' % (
                strategy['ticker'],
                get_sentiment_emoji(strategy['sentiment']))
            ratio = get_ratio(strategy)
            gain = format_ratio(ratio)

            if trade:
                date_str = '**%s**' % date_str
                trade_str = '**%s**' % trade_str

            print('%s | %s | %s | %s | %s | %s' % (
                date_str,
                trade_str,
                gain,
                format_dollar(value),
                total_return,
                annualized_return))

        if trade:
            previous_trade_date = date