Signal research

1.A researcher standardizes each name's earnings yield by the mean and spread of that yield on the whole tape, then studies the 2014 scores. The 2014 score for a name uses yields printed in 2016. What entered the score?

2.The large-cap backtest keeps today's index members and replays them from 2004. Names that were removed after going bankrupt never appear. The long-only result looks smooth. What is missing from that tape?

3.The earnings file stores one number per quarter, and that cell is overwritten when the company restates. The backtest reads the cell on the original announcement date. What is the announcement-day earnings figure?

4.Forty entry thresholds were tried on the same equity tape. The shipped rule is the one whose t-stat cleared 2, and the note compares that t-stat with the usual single-test bar. What did the search do to that comparison?

5.The signal's edge is measured on overlapping 20-day returns, and the t-stat treats each window as its own draw. About nineteen of every twenty windows share most of their days. What happens to the standard error?

6.The published Sharpe marks every fill at the midpoint and subtracts nothing for the spread, impact, or borrow. The book turns over several times a month in small names. What is that Sharpe?

7.The signal is the close-to-close return, and the backtest buys at that same close. The position then earns the next day's move. What is wrong with the fill?

8.A name closes at 80 and opens at 40 the next day. The tape shows a two-for-one split and no news. The backtest treats the drop as a crash and shorts the open. What is the open?

9.When a name leaves the tape, the backtest writes a zero return for its last day and drops it. Several of those names were acquired for cash well below the last print, and a few went to zero. What does the zero do to a long book?

10.A name may be traded on a day only if its average daily volume over the whole taped history is large. The 2012 decision uses volume from 2018. What is that filter using?

11.Before ranking, the researcher clips every score at the 1st and 99th percentiles of the scores on the entire tape. A 2011 score is clipped with cutoffs that move when 2020 is added. What do those cutoffs contain?

12.Daily strategy returns are randomly split into five folds. The model is trained on four folds and scored on the fifth, and the note calls the fifth fold out of sample. What did the random split break?

13.The label is the return from the close to ten days later. Fold two starts the day after fold one ends, and both folds are used in the same fit. Training labels in fold one reach across the boundary. What do those labels contain?

14.The desk keeps the same signal and lowers the assumed spread until the backtest Sharpe clears 1. The shipped note uses that spread. What is the spread in that note?

15.After the equity curve is plotted, the worst month is deleted and the Sharpe is recomputed. The note quotes the recomputed number as the strategy's Sharpe. What was deleted?

16.The book is long high-beta names and flat the index through a long rally. The note calls the gain alpha because the portfolio was not the index itself. What is the gain?

17.The index hedge for day t is a beta fit on a window that includes day t's own return. The hedge is applied to day t. What did the hedge see?

18.Exchange short interest is published about two weeks after the as-of date. The backtest trades on the as-of date using that file. What is knowable on the as-of date?

19.A holdings signal is taken from the quarterly ownership filings. The backtest changes positions on the quarter-end date in the filing. Filings are due weeks after quarter-end. What is in the quarter-end book?

20.The short book targets names with high borrow fees. The backtest never subtracts a borrow fee. The short side looks strong. What did the path omit?

21.A futures signal buys when the back-adjusted continuous price is below 100. Each later roll shifts the older prices used to build that series. What is the old level of 100?

22.The backtest buys at the bid and sells at the ask whenever the quote flickers, and it marks the inventory at the midpoint. The round trip shows a profit on almost every flicker. Where does the profit come from?

23.A signal uses the last local close in Asia to trade that market, and the backtest fills at that close. The local session had already ended, and the price on the screen is hours old. What fill did the backtest take?

24.The earnings signal is computed after the announcement-day move, and the backtest buys at the prior close. The position earns the announcement jump. What did the prior-close fill include?

25.The vendor's addition file is stamped a week before the announcement, and the backtest buys then. The real announcement came later, and the price ran up after it. What does the early stamp do?

26.A small-cap book sends orders equal to half of each name's average daily volume and fills them all at the midpoint. Size is then scaled up tenfold and the same fills are kept. What did scaling up skip?

27.Twelve features are combined with weights fit on the full taped history. The Sharpe of that combination, on that same history, is the number in the abstract. What is the abstract reporting?

28.A walk-forward refits the model every year. After all the years are done, the researcher keeps the lookback whose average across those years was best, and quotes that average. What is the quoted lookback?

29.The information coefficient is the correlation of today's signal with today's return, and both are known at the close. The desk wants to trade the close. What is that coefficient?

30.Winners are 55 percent of trades and average 40 cents. Losers average one dollar. The note ships the rule because the hit rate is above one half. What is the average trade?

31.The researcher computes a Sharpe on overlapping 5-day returns and multiplies by the square root of 252. The note calls the product an annual Sharpe. What did that scaling treat each window as?

32.Strategy profits arrive in clusters around the same macro prints. A bootstrap draws single days at random and rebuilds the path. The interval around the mean is very tight. What did the draw assume?

33.Two futures price levels are regressed on each other over ten years. The slope's t-stat is large. The desk trades the residual back toward the fitted line, using that same fit. What is the large t-stat evidence of?

34.A signal z-scores the price level of a futures market that has trended for years, using the sample mean of the level. Crosses of the mean are traded as mean reversion. What is the mean of a trending level?

35.Position size each morning is target volatility divided by a realized-volatility number. That number is the full-sample volatility of the strategy, one figure for every date. What does this morning's size depend on?

36.Sector labels are taken from the vendor's current map and applied to every past year. A name that moved from banks to software last year is labeled software in 2012. What is the 2012 label?

37.The book earns local-currency returns and the note compares them with a dollar benchmark. The currency moved a long way in favor of the local currency. The difference is called alpha. What is mixed into that difference?

38.The yield signal on Monday uses a dividend the company declared on Thursday. The backtest buys on Monday and collects it. What was knowable on Monday?

39.A name is halted for three days. The backtest marks it at the last print before the halt and shows no profit or loss until it reopens down sharply. What did those flat days do?

40.The signal is the US listing's close. The backtest sells the local line at that local line's close, which printed hours earlier. What price did the sale take?

41.The FX signal is the published fix. The backtest trades after the fixing window and fills at the fix that was just published. Who can still deal at that fix?

42.Every short is assumed located, including names the street could not borrow. Those names are exactly the ones the signal prefers. What did the backtest assume?

43.Borrow is charged at a flat 50 basis points a year. The signal's shorts are the names that were special, with fees many times that. What does the flat fee do?

44.History is joined on ticker. The ticker now belongs to a software firm. Ten years ago the same ticker was a mining firm that later changed symbols. The backtest treats the rows as one company. What is in the early rows?

45.A parent drops sharply on the day it spins off a division. The backtest records the drop and drops the new shares, which the holder received. What happened to the holder's wealth?

46.The short book prefers high-dividend names. The backtest credits price returns only and never charges the dividend. What does a short owe when a dividend is paid?

47.Each feature is replaced by the residual of a regression on the forward return, fit in sample. The residual is then used to predict that same forward return. What did the residual use?

48.Features are scaled with a mean and spread fit on every row, and only then are the rows split into train and test. The test scores use those scaled values. What did the scale see?

49.Five models are scored on the test window. The note reports the test score of the model that won that comparison. What was the test window used for?

50.Training stops when the error on the test window stops falling. The weights from that stopping point are the ones reported on that same window. What did the stopping rule see?

51.The holdout chart is on the research screen while thresholds are changed. The final note calls the holdout untouched. What is the holdout?

52.About two hundred configurations were tried. The best Sharpe is 1.2, and the note says 1.2 clears a bar that was written for a single prespecified rule. What is wrong with that comparison?

53.The two hundred trials share most of their trades, and the note says multiple testing therefore does not apply. The shipped rule is still the trial that won. What does the overlap change?

54.Voting shares and non-voting shares of the same firm are both in the universe. The t-stat treats them as two names. The positions move together almost one for one. What did the t-stat count?

55.An ETF and the basket of names inside it are both in the book. The signal buys the ETF when its last print lags the basket, and the backtest fills that stale ETF print. What is the fill?

56.A pairs hedge ratio is an ordinary least-squares fit on the whole ten-year tape. Trades over those ten years use that ratio, and the note calls them out of sample because the residual looked stationary. What did the ratio see?

57.A stop sits 2 percent under the prior close. The next open is 6 percent under that close. The backtest exits at the stop price. What fill was available at the open?

58.On each daily bar the backtest buys the low and sells the high. The path is smooth. What does that path require the desk to have known?

59.The rule trades at the open only if that day's volume will finish above its 20-day average. The decision is made at the open. What is the volume test using?

60.The confirmation years were on the same charts the researcher used while editing the rule. The code is then frozen and those years are quoted as confirmation. What did those years do during the edits?

61.The code is frozen, but the vendor restates history and the researcher reruns the same code. The Sharpe moves. Why did a frozen program change its answer?

62.The database is advertised as point-in-time, but the link from a date to a company uses the ticker the firm has today. A reused ticker pulls in the wrong firm's old rows. What failed?

63.Fundamentals are stored on the quarter's period-end date. The filing that revealed them is weeks later. The backtest trades at the period end. What is known at period end?

64.The researcher computes results at 1, 5, 10, and 20 day holds, then writes the abstract about the hold that looked best. The methods section calls that hold prespecified. What was prespecified?

65.The same rule is run in twenty countries. The note reports the two countries where it worked and does not mention the other eighteen. What is the reported pair?

66.The headline Sharpe drops the crisis months and quotes the rest as the strategy's performance. An investor in the book would have lived through those months. What is the headline number?

67.The tape is cut so the backtest begins the month after a large drawdown. Earlier years exist and are not shown. Why does the curve start there?

68.The entry threshold is the value that maximized Sharpe on this tape. The note then quotes that Sharpe as the threshold's out-of-sample performance. What was maximized?

69.A cost of 20 basis points per year is subtracted no matter how often the book trades. The researcher then shortens the hold from a month to a day and keeps the same 20 basis points. What did the shorter hold do to costs?

70.The signal moves slowly, but the book rebalances every day and pays the spread each time. Most daily changes are smaller than the spread. What happens to the edge?

71.The book is scaled from a small research size to a large one. Fills stay at the old prices. What was held constant?

72.The cost line is a commission of a few cents and nothing for impact. Orders are a large fraction of the day's volume in the names being traded. What is missing from the cost line?

73.The book is dollar neutral, long high-beta names and short low-beta names, through a rally. There is no index hedge. The gain is called market neutral. What exposure remains?

74.Single-name shorts already offset part of the long book. The index hedge is then set equal to the full long market value, and the shorts are left in place. What is the net market exposure?

75.The researcher subtracts each sector's mean from the signal, then buys the global top decile and sells the global bottom decile. One sector dominates both deciles because its scores are more spread out. The note calls the book sector neutral. What is true of the weights?

76.The signal is built from the opening print. The backtest still earns the close-to-open gap that ended at that print. What is the gap relative to the signal?

77.The closing-auction imbalance is published a few minutes before the auction. The backtest trades at 3:00 using the imbalance file from that auction. What is the 3:00 book using?

78.Day-of-week effects are estimated on this tape, and the same tape is used to quote the winner. No later period is shown. What is the quoted effect?

79.The drift study buys at the close before the earnings announcement and holds through the announcement. The note calls the result post-announcement drift. What does the position earn that drift does not include?

80.The surprise is the print minus a consensus that was revised during the call, after the print had started. The backtest uses that revised consensus at the announcement. What is the consensus?

81.The payrolls feature on release morning is the revised figure published in a later month, not the first print. The backtest trades the release. What could the desk have traded on?

82.A feature is built from the policy-statement sentence. It is aligned to the last trade before the statement's timestamp, and the backtest sells at that trade. What is the sentence relative to that trade?

83.Slippage is a regression fit on the same days the backtest reports. The net Sharpe on those days uses that fitted slippage. What did the cost model see?

84.A special dividend is cancelled after the backtest's tape is built. The tape still credits the cash. What did the long receive?

85.A buy limit for the full order is marked filled whenever the low touches the limit, even when the volume printed at that price is a small fraction of the order. What did the fill assume?

86.A name is kept only if next year's earnings are already in the file. The rule is applied to every past date, including dates when next year had not been reported. What does the screen require?

87.Portfolio weights each day come from a covariance matrix estimated once, on the whole tape. The backtest of those weights is called out of sample. What is inside the matrix?

88.The same crash is used to design the hedge and then cited as proof the hedge works. What is the crash in the proof?

89.A factor published in a paper is backtested from the early tape through the paper's publication date. The note calls that window out of sample because the researcher did not personally fit it. What is that window relative to the paper?

90.The Sharpe is computed only on days with a position, and it is annualized as if those days were the whole year. Many days are flat. What does dropping the flat days do?

91.Weights are computed from the closing prices and then applied as if they had been held from the open, earning the open-to-close move. What did the open-to-close piece require?

92.At 10:00 a missing morning return is filled with that name's close-to-close return for the same day. The book trades at 10:00. What is the filler?

93.Entry, exit, hold, and universe were all searched. The note applies a multiple-testing haircut as if only the four holds had been tried. What did the haircut miss?

94.The vendor keeps only the latest analyst consensus and serves that same number for every past date. The backtest reads it on each past date. What is on those past dates?

95.The opening-auction order is several times the typical auction size, and the backtest fills it all at the historical auction print. What did that fill ignore?

96.The book is market neutral across names. Borrow is charged on the net short, which is near zero, not on each shorted share. What was left out?

97.A special dividend is counted once as a cash payment and again as price appreciation that is reinvested. Both amounts enter the total return. What happened to that cash?

98.Local stock returns from years ago are converted to dollars with today's exchange rate, one rate for the whole tape. What is wrong with the old dollar returns?

99.A London cash close is paired, on the same calendar date, with a New York print from late afternoon, after London had shut. The backtest treats the London close as a reaction to that print. What is the London close relative to the print?

100.The signal is the futures settlement, which is published after the close. The backtest sells at that settlement during the session. What is the settlement during the session?

101.American option prices are compared with a model that cannot exercise early. The backtest sells the market and buys the model whenever the market is richer, and it marks the gap as profit. What is in that gap?

102.The rule loses money on the first pass. The researcher flips its sign and reports the flipped path as the result. No later sample is shown. What is the flipped path?

103.Rules are kept only when the training window and the test window both clear a t-stat of 2. The researcher keeps trying rules until one does. The test hurdle is described as untouched. What was the test hurdle part of?

104.Some names are marked at the closing auction and others at a quoted midpoint from earlier. A cross-sectional gap between those marks is traded. What is the gap?

105.The code lags the signal by one day and also shifts the return column by one day in the same direction. The two shifts cancel, and the signal lines up with the same day's return again. What did the second shift do?

106.Names that fail a tradable-price check stay in the cross-sectional rank and can be selected. The backtest fills a selected halt at the last print from before the halt. What is that fill?

107.The researcher standardizes each name by its own mean and volatility over the whole tape, then ranks the standardized scores each day. A 2015 rank uses a mean that includes 2021. What moves the 2015 rank when 2021 is added?

108.The note reports the median Sharpe across fifty random seeds. The book that is shipped is the single seed with the highest Sharpe, and the median is quoted beside it as if it were that book. What does the median describe?