if order.buy_sell == 'B': lpriceOB.append(order.price) ltimeOB.append(order.otime) lsizeOB.append(order.size) else: lpriceOS.append(order.price) ltimeOS.append(order.otime) lsizeOS.append(order.size) # if order == 'U': # nORN = data[4].strip() # sorders.process_order(stock, order, nORN, timestamp.itime, updid=ORN, price=float(data[6].strip()), size=int(data[5].strip())) # Computes the time between placing and order and canceling it if order.type == 'D': trans = sorders.query_id(order.id) if trans is not None: ldelete.append(order.otime - trans.otime) else: print('MISSING DELETED' + order.id) # sorders.process_order(stock, order, ORN) # Computes the time between placing and order and its execution if order.type in ['E', 'C']: trans = sorders.query_id(order.id) if trans.buy_sell == 'S': lexecutionsS.append(order.otime - trans.otime) ltimeES.append(order.otime) lpriceES.append(trans.price) lsizeES.append(trans.size) else:
data = mess.split(',') timestamp = ITCHtime(int(data[1].strip())) order = data[2].strip() ORN = data[3].strip() if order in ['F', 'A']: if order == 'A': price = float(data[7].strip()) else: price = float(data[8].strip()) sorders.process_order(stock, order, ORN, otime=timestamp, bos=data[5].strip(), price=price, size=int(data[6].strip())) if order == 'U': nORN = data[4].strip() sorders.process_order(stock, order, nORN, timestamp, updid=ORN, price=float(data[6].strip()), size=int(data[5].strip())) # Computes the time between placing and order and canceling it if order == 'D': trans = sorders.query_id(ORN) ldelete.append(timestamp.itime - trans[1]) sorders.process_order(stock, order, ORN) if order in ['E', 'C']: trans = sorders.query_id(ORN) if trans[2] == 'S': lexecutionsS.append(timestamp.itime - trans[1]) else: lexecutionsB.append(timestamp.itime - trans[1]) # Working with the logs # Computing a histogram using powers of 10 as discretization of time # Deletions deletions = np.log10(ldelete)
rfile = ITCHMessages(year, day, stock) sorders = OrdersProcessor() rfile.open() cpny = Company() wfile = open(datapath + '/Results/' + day + '-' + stock + '-EXEC.csv', 'w') i = 0 norders = 0 for order in rfile.get_order(): print(order.to_string()) sorders.insert_order(order) if order.type in ['E', 'C']: trans = sorders.query_id(order.ORN) wfile.write(stock + ',' + str(order.otimes) + ',' + str(trans.otime) + ',' + trans.buy_sell + ',' + str(order.size) + ',' + str(trans.price)) if order == 'C': wfile.write(',' + str(order.price) + '\n') else: wfile.write('\n') i += 1 if i % 10000 == 0: print('.', end='', flush=True) wfile.flush() wfile.close() rfile.close()
def order_exec_analysis(year, day, stock, logging=False, market=False): """ :param year: :param day: :param stock: :return: """ ## Structure for collecting statistics statistics = {v: {} for v in timelines} for v in statistics: statistics[v]['buy'] = {s: [] for s in stat} statistics[v]['sell'] = {s: [] for s in stat} rfile = ITCHMessages(year, day, stock) rfile.open() sorders = OrdersProcessor() for order in rfile.get_order(): sorders.insert_order(order) lopen = sorders.sorted_orders(otype='open') lexecuted = sorders.sorted_orders(otype='executed') lcancelled = sorders.sorted_orders(otype='cancelled') # list for storing all the orders in chonological order lorders = [] # Add to the list an open order with its time and all the partial executions for o in lopen: if not market or (time_to_nanoseconds(9, 30) < o.otime < time_to_nanoseconds(16)): lorders.append((o.otime, 'O', o.id)) # Orders still open buy maybe partially executed for xo in range(1, len(o.history)): if o.history[xo].type in ['C', 'E']: if not market or (time_to_nanoseconds( 9, 30) < o.history[xo].otime < time_to_nanoseconds(16)): lorders.append((o.history[xo].otime, f'OF{xo}', o.id)) # Add to the list an executed order with the time of all the partial # executions and the last execution for o in lexecuted: if not market or (time_to_nanoseconds(9, 30) < o.otime < time_to_nanoseconds(16)): lorders.append((o.otime, 'XI', o.id)) # Partial executions for xo in range(1, len(o.history) - 1): if o.history[xo].type in ['C', 'E']: if not market or (time_to_nanoseconds( 9, 30) < o.history[xo].otime < time_to_nanoseconds(16)): lorders.append((o.history[xo].otime, f'XF{xo}', o.id)) # Final execution if not market or (time_to_nanoseconds(9, 30) < o.history[-1].otime < time_to_nanoseconds(16)): lorders.append((o.history[-1].otime, f'XF', o.id)) # Add to the list a cancelled order with the time of the initial order # all the possible partial executions # and the time of the final cancellation for o in lcancelled: if not market or (time_to_nanoseconds(9, 30) < o.otime < time_to_nanoseconds(16)): lorders.append((o.otime, 'CI', o.id)) for xo in range(1, len(o.history) - 1): if o.history[xo].type in ['C', 'E']: if not market or (time_to_nanoseconds( 9, 30) < o.history[xo].otime < time_to_nanoseconds(16)): lorders.append((o.history[xo].otime, f'UF{xo}', o.id)) # Last item should be a cancelation (X) or a cancel/replace (U) if not market or (time_to_nanoseconds(9, 30) < o.history[-1].otime < time_to_nanoseconds(16)): lorders.append((o.history[-1].otime, 'CF', o.id)) lorders = sorted(lorders) # Processes all the itervals for the orders and registers for all the executions a # some statistics cbuy = Counter() csell = Counter() weird = 0 texec = 0 for _, op, orderid in lorders: if op == 'O': if sorders.orders[orderid].buy_sell == 'B': cbuy[sorders.orders[orderid].price] += 1 else: csell[sorders.orders[orderid].price] += 1 # sopen.append(orderid) elif op == 'CI': if sorders.cancelled[orderid].buy_sell == 'B': cbuy[sorders.cancelled[orderid].price] += 1 else: csell[sorders.cancelled[orderid].price] += 1 # scancel.append(orderid) elif op == 'CF': if sorders.cancelled[orderid].buy_sell == 'B': cbuy[sorders.cancelled[orderid].price] -= 1 else: csell[sorders.cancelled[orderid].price] -= 1 # scancel.remove(orderid) elif op == 'XI': if sorders.executed[orderid].buy_sell == 'B': cbuy[sorders.executed[orderid].price] += 1 else: csell[sorders.executed[orderid].price] += 1 # sexec.append(orderid) else: # it is an execution # If is a final execution the code is 'XF', else it has a number attached # The final execution eliminates the price from the order book so the first now is the second best price exorder = sorders.query_id(orderid) if len(op) == 2: if exorder.buy_sell == 'B': cbuy[exorder.price] -= 1 else: csell[exorder.price] -= 1 pendingbuy = sorted([v for v in cbuy.items() if v[1] > 0], reverse=True) pendingsell = sorted([v for v in csell.items() if v[1] > 0]) # Checks if the queues are empty bestbuy = pendingbuy[0][0] if len(pendingbuy) > 0 else -1 bestsell = pendingsell[0][0] if len(pendingsell) > 0 else -1 if len(op) == 2: timeline = in_timeline(exorder.history_time_length()) else: timeline = in_timeline( exorder.history_time_length(dist=int(op[2:]))) # print(f'{sorders.executed[orderid].price} {pendingbuy[0]} {pendingsell[0]}') # If any of the queues is empty the statistics make no sense so they are not computed if (bestsell != -1) and (bestbuy != -1): texec += 1 if logging: print( f'******************************************** {op[2:]}' ) print(f'ID: {orderid}') print(exorder.to_string(history=True)) # Get the execution order number if len(op) == 2: hist_exorder = exorder.history[-1] else: hist_exorder = exorder.history[int(op[2:])] # Get the price of the executed order (if the type is C then it is an execution with price) exprice = hist_exorder.price if exorder.type == 'C' else exorder.price # gap - the difference between the price of the execution and the best price of the other side # diff - the difference between the price of the execution and the second best price if exorder.buy_sell == 'B': buy_sell = 'buy' gap = bestsell - exprice diff = exprice - bestbuy if logging: print( f'BUY: {exprice} / GAP: {gap:3.2f} / DIFF: {diff:3.2f}' ) else: buy_sell = 'sell' gap = exprice - bestbuy diff = bestsell - exprice if logging: print( f'SELL: {exprice} / GAP: {gap:3.2f} / DIFF: {diff:3.2f}' ) if logging: print(f'QSELL5={pendingsell[:5]}') print(f'QBUY5={pendingbuy[:5]}') print(f'BBUY={bestbuy} BSELL={bestsell}') print( f'LQBUY={sum_count(pendingbuy)} LQSELL={sum_count(pendingbuy)}' ) if gap < 0 or diff < 0: weird += 1 pass # print(f'?????????????????????????????????????????????????????????????') # print(f'OP: {buy_sell} OTHER= {gap} SECOND= {diff}') # print(f'ID: {orderid}') # print(exorder.to_string(history=True)) # print(f'P:{exprice} BS: {bestsell} BB: {bestbuy}') # print(f'QSELL5={pendingsell[:5]}') # print(f'QBUY5={pendingbuy[:5]}') # print(f'BBUY={bestbuy} BSELL={bestsell}') # print(f'LQBUY={sum_count(pendingbuy)} LQSELL={sum_count(pendingbuy)}') else: statistics[timelines[timeline]][buy_sell]['price'].append( exprice) statistics[timelines[timeline]][buy_sell]['lenbuy'].append( sum_count(pendingbuy)) statistics[ timelines[timeline]][buy_sell]['lensell'].append( sum_count(pendingsell)) statistics[ timelines[timeline]][buy_sell]['lenbuy5'].append( sum_count(pendingbuy, lim=5)) statistics[ timelines[timeline]][buy_sell]['lensell5'].append( sum_count(pendingsell, lim=5)) statistics[ timelines[timeline]][buy_sell]['lenbuy10'].append( sum_count(pendingbuy, lim=10)) statistics[ timelines[timeline]][buy_sell]['lensell10'].append( sum_count(pendingsell, lim=10)) statistics[timelines[timeline]][buy_sell][ 'otherprice'].append(diff) statistics[timelines[timeline]][buy_sell]['gap'].append( gap) statistics[timelines[timeline]][buy_sell]['size'].append( hist_exorder.size) print(f"W={weird} TEX={texec}") for st in stat: for v in timelines: statistics[v]['buy'][st] = np.array(statistics[v]['buy'][st]) statistics[v]['sell'][st] = np.array(statistics[v]['sell'][st]) return statistics
def order_statistics(year, day, stock): """ Computes statistics for the orders of a stock :param year: :param day: :param stock: :return: """ statistics = { 'buy': { 'ordersize': [], 'ordertime': [], 'orderprice': [], 'executionsize': [], 'executionprice': [], 'executiondeltatime': [], 'executiontime': [], 'deletedeltatime': [] }, 'sell': { 'ordersize': [], 'ordertime': [], 'orderprice': [], 'executionsize': [], 'executionprice': [], 'executiontime': [], 'executiondeltatime': [], 'deletedeltatime': [] }, } i = 0 norders = 0 rfile = ITCHMessages(year, day, stock) sorders = OrdersProcessor() rfile.open() for order in rfile.get_order(): sorders.insert_order(order) if not args.market or (time_to_nanoseconds(9, 30) < order.otime < time_to_nanoseconds(16)): if order.type in ['F', 'A', 'U']: norders += 1 # if 0 < order.price < 5000: if order.buy_sell == 'S': statistics['sell']['ordersize'].append(order.size) statistics['sell']['ordertime'].append(order.otime) statistics['sell']['orderprice'].append(order.price) else: statistics['buy']['ordersize'].append(order.size) statistics['buy']['ordertime'].append(order.otime) statistics['buy']['orderprice'].append(order.price) # If is a cancel/replace order consider also a deletion if order.type in ['U']: trans = sorders.query_id(order.oid) if order.buy_sell == 'S': statistics['sell']['deletedeltatime'].append(order.otime - trans.otime) else: statistics['buy']['deletedeltatime'].append(order.otime - trans.otime) # Computes the time between placing and order and canceling it if order.type == 'D': trans = sorders.query_id(order.id) if trans is not None: if trans.buy_sell == 'S': statistics['sell']['deletedeltatime'].append( order.otime - trans.otime) else: statistics['buy']['deletedeltatime'].append( order.otime - trans.otime) else: print('MISSING DELETED' + order.id) # Computes the time between placing and order and its execution if order.type in ['E', 'C']: trans = sorders.query_id(order.id) if trans.buy_sell == 'S': statistics['sell']['executiondeltatime'].append( order.otime - trans.otime) statistics['sell']['executiontime'].append(order.otime) if order.type == 'E': statistics['sell']['executionprice'].append( trans.price) else: # Execution with price statistics['sell']['executionprice'].append( order.price) statistics['sell']['executionsize'].append(order.size) else: statistics['buy']['executiondeltatime'].append( order.otime - trans.otime) statistics['buy']['executiontime'].append(order.otime) if order.type == 'E': statistics['buy']['executionprice'].append(trans.price) else: # Execution with price statistics['buy']['executionprice'].append(order.price) statistics['buy']['executionsize'].append(order.size) # Convert everything to numpy arrays for v in statistics: for att in statistics[v]: statistics[v][att] = np.array(statistics[v][att]) return statistics