Esempio n. 1
0
def serialize_and_pack_portfolio_details(game_id: int, user_id: int):
    out_dict = dict(data=[], headers=list(PORTFOLIO_DETAIL_MAPPINGS.values()))
    balances = get_active_balances(game_id, user_id)
    if balances.empty:
        s3_cache.set(f"{game_id}/{user_id}/{CURRENT_BALANCES_PREFIX}", json.dumps(out_dict))
        return
    cash_balance = get_current_game_cash_balance(user_id, game_id)
    symbols = balances["symbol"].unique()
    prices = get_most_recent_prices(symbols)
    df = balances.groupby("symbol", as_index=False).aggregate({"balance": "last", "clear_price": "last"})
    df = df.merge(prices, on="symbol", how="left")
    df["Value"] = df["balance"] * df["price"]
    total_portfolio_value = df["Value"].sum() + cash_balance
    df["Portfolio %"] = (df["Value"] / total_portfolio_value)
    close_prices = get_last_close_prices(symbols)
    df = df.merge(close_prices, how="left")
    df["Change since last close"] = ((df["price"] - df["close_price"]) / df["close_price"])
    del df["close_price"]
    symbols_colors = assign_colors(symbols)
    df["color"] = df["symbol"].apply(lambda x: symbols_colors[x])
    df.rename(columns=PORTFOLIO_DETAIL_MAPPINGS, inplace=True)
    df.fillna(NA_TEXT_SYMBOL, inplace=True)
    records = df.to_dict(orient="records")
    out_dict["data"] = records
    s3_cache.set(f"{game_id}/{user_id}/{CURRENT_BALANCES_PREFIX}", json.dumps(out_dict))
Esempio n. 2
0
def process_order(order_id: int):
    timestamp = time.time()
    if get_order_expiration_status(order_id):
        add_row("order_status",
                order_id=order_id,
                timestamp=timestamp,
                status="expired",
                clear_price=None)
        return

    order_ticket = query_to_dict("SELECT * FROM orders WHERE id = %s",
                                 order_id)[0]
    symbol = order_ticket["symbol"]
    game_id = order_ticket["game_id"]
    user_id = order_ticket["user_id"]
    buy_or_sell = order_ticket["buy_or_sell"]
    quantity = order_ticket["quantity"]
    order_type = order_ticket["order_type"]

    market_price, _ = fetch_price(symbol)

    # Only process active outstanding orders during trading day
    cash_balance = get_current_game_cash_balance(user_id, game_id)
    current_holding = get_current_stock_holding(user_id, game_id, symbol)
    if during_trading_day():
        if execute_order(buy_or_sell, order_type, market_price,
                         order_ticket["price"], cash_balance, current_holding,
                         quantity):
            order_status_id = add_row("order_status",
                                      order_id=order_id,
                                      timestamp=timestamp,
                                      status="fulfilled",
                                      clear_price=market_price)
            update_balances(user_id, game_id, order_status_id, timestamp,
                            buy_or_sell, cash_balance, current_holding,
                            market_price, quantity, symbol)
            serialize_and_pack_pending_orders(
                game_id, user_id)  # refresh the pending orders table
            add_fulfilled_order_entry(
                game_id, user_id,
                order_id)  # add the new fulfilled orders entry to the table
            serialize_and_pack_portfolio_details(game_id, user_id)
        else:
            # if a market order was placed after hours, there may not be enough cash on hand to clear it at the new
            # market price. If this happens, cancel the order and recalculate the purchase quantity with the new price
            if order_type == "market":
                cancel_order(order_id)
                updated_quantity = cash_balance // market_price
                if updated_quantity <= 0:
                    return

                place_order(user_id, game_id, symbol, buy_or_sell,
                            cash_balance, current_holding, order_type,
                            "Shares", market_price, updated_quantity,
                            order_ticket["time_in_force"])
                serialize_and_pack_portfolio_details(game_id, user_id)
Esempio n. 3
0
def add_virtual_cash(game_id: int, user_id: int, dividend_id: int,
                     amount: float):
    current_cash = get_current_game_cash_balance(user_id, game_id) + amount
    now = datetime_to_posix(dt.now())
    add_row("game_balances",
            user_id=user_id,
            game_id=game_id,
            timestamp=now,
            balance_type='virtual_cash',
            balance=current_cash,
            dividend_id=dividend_id,
            transaction_type="stock_dividend")
Esempio n. 4
0
def get_user_portfolio_value(game_id: int, user_id: int, cutoff_time: float = None) -> float:
    """This works slightly differently than the method currently used to generates the leaderboard and charts, which
    depends on the dataframe that make_historical_balances_and_prices_table produces, using pandas.merge_asof. this
    minor discrepancy could result in leaderboard orders in the game panel not precisely matching how they are
    calculated internally here."""
    cash_balance = get_current_game_cash_balance(user_id, game_id)
    balances = get_active_balances(game_id, user_id)
    symbols = balances["symbol"].unique()
    if len(symbols) == 0:
        return cash_balance
    prices = get_most_recent_prices(symbols, cutoff_time)
    df = balances[["symbol", "balance"]].merge(prices, how="left", on="symbol")
    df["value"] = df["balance"] * df["price"]
    return df["value"].sum() + cash_balance
Esempio n. 5
0
def apply_stock_splits(start_time: float = None, end_time: float = None):
    splits = get_splits(start_time, end_time)
    if splits.empty:
        return

    last_prices = get_most_recent_prices(splits["symbol"].to_list())
    for i, row in splits.iterrows():
        symbol = row["symbol"]
        numerator = row["numerator"]
        denominator = row["denominator"]
        exec_date = row["exec_date"]
        df = get_active_holdings_for_games(symbol)
        if df.empty:
            continue

        # make order status updates here. implement dask or spark here when the job getting to heavy to run locally
        # (e.g. see https://docs.dask.org/en/latest/dataframe-api.html#dask.dataframe.read_sql_table)
        df["transaction_type"] = "stock_split"
        df["order_status_id"] = None
        df["timestamp"] = exec_date
        df["fractional_balance"] = df["balance"] * numerator / denominator
        df["balance"] = df["balance"] * numerator // denominator
        df["fractional_balance"] -= df["balance"]

        # identify any fractional shares that need to be converted to cash
        mask = df["fractional_balance"] > 0
        fractional_df = df[mask]
        last_price = float(last_prices.loc[last_prices["symbol"] == symbol,
                                           "price"].iloc[0])
        for _, fractional_row in fractional_df.iterrows():
            game_id = fractional_row["game_id"]
            user_id = fractional_row["user_id"]
            cash_balance = get_current_game_cash_balance(user_id, game_id)
            add_row("game_balances",
                    user_id=user_id,
                    game_id=game_id,
                    timestamp=exec_date,
                    balance_type="virtual_cash",
                    balance=cash_balance +
                    fractional_row["fractional_balance"] * last_price,
                    transaction_type="stock_split")

        # write updated balances
        del df["fractional_balance"]
        with engine.connect() as conn:
            df.to_sql("game_balances", conn, index=False, if_exists="append")
Esempio n. 6
0
def compile_and_pack_player_leaderboard(game_id: int, start_time: float = None, end_time: float = None):
    user_ids = get_active_game_user_ids(game_id)
    usernames = get_game_users(game_id)
    user_colors = assign_colors(usernames)
    records = []
    for user_id in user_ids:
        user_info = get_user_information(user_id)  # this is where username and profile pic get added in
        cash_balance = get_current_game_cash_balance(user_id, game_id)
        balances = get_active_balances(game_id, user_id)
        stocks_held = list(balances["symbol"].unique())
        portfolio_value = get_user_portfolio_value(game_id, user_id)
        stat_info = make_stat_entry(color=user_colors[user_info["username"]],
                                    cash_balance=cash_balance,
                                    portfolio_value=portfolio_value,
                                    stocks_held=stocks_held,
                                    return_ratio=rds.get(f"{RETURN_RATIO_PREFIX}_{game_id}_{user_id}"),
                                    sharpe_ratio=rds.get(f"{SHARPE_RATIO_PREFIX}_{game_id}_{user_id}"))
        records.append({**user_info, **stat_info})

    if check_single_player_mode(game_id):
        for index in TRACKED_INDEXES:
            index_info = query_to_dict("""
                SELECT name as username, avatar AS profile_pic 
                FROM index_metadata WHERE symbol = %s""", index)[0]
            portfolio_value = get_index_portfolio_value(game_id, index, start_time, end_time)
            stat_info = make_stat_entry(color=user_colors[index_info["username"]],
                                        cash_balance=None,
                                        portfolio_value=portfolio_value,
                                        stocks_held=[],
                                        return_ratio=rds.get(f"{RETURN_RATIO_PREFIX}_{game_id}_{index}"),
                                        sharpe_ratio=rds.get(f"{SHARPE_RATIO_PREFIX}_{game_id}_{index}"))
            records.append({**index_info, **stat_info})

    benchmark = get_game_info(game_id)["benchmark"]  # get game benchmark and use it to sort leaderboard
    records = sorted(records, key=lambda x: -x[benchmark])
    output = dict(days_left=_days_left(game_id), records=records)
    s3_cache.set(f"{game_id}/{LEADERBOARD_PREFIX}", json.dumps(output))
    def test_apply_dividends_to_games(self):
        user_id = 1
        date = datetime_to_posix(dt.now().replace(hour=0,
                                                  minute=0,
                                                  second=0,
                                                  microsecond=0))
        amzn_dividend = 10
        tsla_dividend = 20
        envidia_dividend = 15
        fake_dividends = ([['AMZN', 'Amazon INC', amzn_dividend, date],
                           ['TSLA', 'Tesla Motors', tsla_dividend, date],
                           ['NVDA', 'Envidia SA', envidia_dividend, date]])
        fake_dividends = pd.DataFrame(
            fake_dividends,
            columns=['symbol', 'company', 'amount', 'exec_date'])

        insert_dividends_to_db(fake_dividends)
        user_1_game_8_balance = get_current_game_cash_balance(user_id, 8)
        user_1_game_3_balance = get_current_game_cash_balance(user_id, 3)
        should_remain_1_000_000 = get_current_game_cash_balance(user_id, 6)
        apply_dividends_to_stocks()

        # user 1, game 8 is holding NVDA. we should see their holding * the dividend in their updated cash balance
        envidia_holding = get_current_stock_holding(user_id, 8, "NVDA")
        self.assertEqual(
            get_current_game_cash_balance(user_id, 8) - user_1_game_8_balance,
            envidia_holding * envidia_dividend)

        # user 1, game 3 is holding AMZN, TSLA, and NVDA. Their change in cash balance should be equal to the sum of
        # each position * its corresponding dividend
        active_balances = get_active_balances(3, user_id)
        merged_table = fake_dividends.merge(active_balances,
                                            how="left",
                                            on="symbol")
        merged_table[
            "payout"] = merged_table["amount"] * merged_table["balance"]
        total_dividend = merged_table["payout"].sum()
        self.assertEqual(
            get_current_game_cash_balance(user_id, 3) - user_1_game_3_balance,
            total_dividend)

        # user 1 isn't holding any dividend-paying shares in game 6. they should have no cash balance change
        self.assertEqual(get_current_game_cash_balance(user_id, 6),
                         should_remain_1_000_000)
    def test_internal_splits_logic(self):
        """test_harvest_stock_splits_external_integration ensures that our external integration is working. this test
        verifies that once we have splits they are appropriately applied to users balances. we'll cover a straight-
        forward split, a split that leaves fractional shares, and a reverse split that leaves fractional shares"""
        game_id = 3
        user_id = 1
        exec_date = simulation_end_time + 1
        splits = pd.DataFrame([
            {
                "symbol": "AMZN",
                "numerator": 5,
                "denominator": 1,
                "exec_date": exec_date
            },
            {
                "symbol": "TSLA",
                "numerator": 5,
                "denominator": 2,
                "exec_date": exec_date
            },
            {
                "symbol": "SPXU",
                "numerator": 2.22,
                "denominator": 3.45,
                "exec_date": exec_date
            },
        ])
        cash_balance_pre = get_current_game_cash_balance(user_id, game_id)
        balances_pre = get_active_balances(game_id, user_id)
        with patch("backend.logic.stock_data.get_splits") as db_splits_mock:
            db_splits_mock.return_value = splits
            apply_stock_splits()

        cash_balance_post = get_current_game_cash_balance(user_id, game_id)
        balances_post = get_active_balances(game_id, user_id)

        pre_amzn = balances_pre[balances_pre["symbol"] ==
                                "AMZN"].iloc[0]["balance"]
        post_amzn = balances_post[balances_post["symbol"] ==
                                  "AMZN"].iloc[0]["balance"]
        self.assertEqual(pre_amzn * 5 / 1, post_amzn)

        pre_tsla = balances_pre[balances_pre["symbol"] ==
                                "TSLA"].iloc[0]["balance"]
        post_tsla = balances_post[balances_post["symbol"] ==
                                  "TSLA"].iloc[0]["balance"]
        self.assertEqual(pre_tsla * 5 // 2, post_tsla)

        pre_spxu = balances_pre[balances_pre["symbol"] ==
                                "SPXU"].iloc[0]["balance"]
        post_spxu = balances_post[balances_post["symbol"] ==
                                  "SPXU"].iloc[0]["balance"]
        self.assertEqual(pre_spxu * 2.22 // 3.45, post_spxu)

        last_prices = get_most_recent_prices(["AMZN", "TSLA", "SPXU"])
        last_tsla_price = last_prices[last_prices["symbol"] ==
                                      "TSLA"].iloc[0]["price"]
        last_spxu_price = last_prices[last_prices["symbol"] ==
                                      "SPXU"].iloc[0]["price"]
        self.assertAlmostEqual(
            cash_balance_pre +
            (pre_tsla * 5 / 2 - post_tsla) * last_tsla_price +
            (pre_spxu * 2.22 / 3.45 - post_spxu) * last_spxu_price,
            cash_balance_post, 3)