GLOBAL SEARCH
搜索在服务端完成，题目解析与答案不会进入搜索结果。登录后可搜索自己的收藏题单。
找到 30 个结果
中文题目A desk studies how aggressiveness $X$ affects trade profitability $Y$, but $Y$ is observed only for orders that actually fill. Fill probability is higher when latent market demand $D$ is strong, and stronger demand also tends to improve profitability. Why can regressing observed
打开 →A desk only records post-launch performance for strategies that first clear an internal backtest hurdle. Why does regressing realized performance on backtest score inside the launched set generally fail to recover the unconditional relationship?
打开 →In a market the per-trade adverse-selection cost to a dealer is alpha*delta, where alpha is the informed fraction and delta is the value gap. Noise traders are willing to pay at most a half-spread of 0.30 before they stop trading entirely. The value gap is delta = 1.0. Above what
打开 →When a resting quote fills, with probability 0.30 it is an informed pick-off that moves 0.05 against the maker; with probability 0.70 it is noise flow with zero adverse move. Ignoring rebates, what minimum half-spread makes expected per-fill PnL exactly zero?
打开 →You capture a half-spread of 0.04 per share when filled. A fraction 0.35 of your fills come from informed traders, and on those fills the fair value moves 0.10 per share against you (uninformed fills have zero subsequent move). What is your expected net edge per fill, per share?
打开 →Why is 'pick the model with the lowest train error' a bad rule for model selection?
打开 →Three items arrive in uniformly random order. Their qualities are NOT all distinct: two of them have quality 2 (tied for best) and one has quality 1. After each item you observe its quality relative to those seen so far, reported as 'higher', 'tied', or 'lower' (so a tie is visib
打开 →Why can studying only executed trades bias the estimated effect of an execution rule, even if the rule assignment itself was randomized upstream?
打开 →A buy trade occurs at 100.02 when the midpoint is 100. Five seconds later the midpoint is 100.03. What are the effective spread, realized spread, and adverse-selection component?
打开 →A sell trade occurs at 49.96 when the midpoint is 50. Five seconds later the midpoint is 49.94. What are the effective spread, realized spread, and adverse-selection component?
打开 →A market maker wants net edge 0.018 per fill after adverse-selection loss. If expected loss is 0.007, rebate is 0.001, and current half-spread is 0.012, what half-spread is required and how much extra widening is needed?
打开 →sql · select · join · group-by · window-functions · cte · null · timezone
打开 →Why is a bad inventory state especially dangerous when adverse selection is also high?
打开 →国内某 SSE 接入团队接到任务: 把老 C++ + Boost.Asio 写的行情接入网关重构成 Rust, 单台机器要同时维持 8000 条 TCP 长连接, 把 510300.SH (沪深300 ETF) 等几百只标的的 tick 流落到内部撮合面板。架构师扫一眼说: 上 tokio——别想着每条连接派一个 OS 线程, 8000 个 OS 线程在调度...
打开 →Two venues route flow to your quotes. Venue A flow is 5% informed; Venue B flow is 40% informed. Both have the same value gap delta = 0.5, and informed flow costs you delta per fill while you earn half-spread s = 0.10 on all fills. You can only post on one venue. Compute expected
打开 →周二上午 11:14，某沪上 私募 量化 团队的研究员刚跑完 沪深300 ETF（510300）上 5 秒级订单流信号的回测：样本内 Sharpe 5.2，样本外 Sharpe 4.8。基金经理盯着权益曲线只问一句：「上规模交易会发生什么？」研究员不知道——回测假设每一笔成交都按中价（mid）拿到、零市场冲击。在 200,000 份的元订单（metaorde...
打开 →某 沪深 300 私募 的 交易员 提单：『上 一 季 在 ticks 表 上 15 秒 出结果 的 1 分钟 VWAP per symbol 查询，今天 跑 了 11 分钟。Postgres 仓库 正在 维持 每秒 9 万 写入 来自 沪深 行情 网关，EXPLAIN 在 一条 仅 触 三日 数据 的 查询 上 报 1.8 亿 缓冲 读取』。L2 的 卫生...
打开 →下午 15:30 CST，某 A 股 量化 私募。沪深 收盘 加 15:00 行情 结算 落定 之后 半 小时，行情 vendor 的 tick 510300 20260523.csv.gz 落 在 共享 挂载 /data/market data/ 上。cron 调起 ingest ticks.sh 。接下来 九十 秒 内，文件 必须 被 加载 到 暂存 表...
打开 →某个周二早晨，沪深 300 量化私募的基金经理走过来：『把过去两周 510050 、 510500 、 510300 的日 VWAP 拉给我，按当日收益做横截面排名，只要 close 非空的行』。数据存在研究数据仓库里——一台部署在内网的 Postgres / PolarDB O 上， bars 1m 1 分钟 K 线表和 instrument 维度表通过外...
打开 →A dealer faces order flow where a fraction 0.4 of traders are informed (always trade in the correct direction) and 0.6 are noise traders who buy or sell with equal probability. The true value is equally likely high or low, so informed traders buy half the time and sell half the t
打开 →A dealer has just observed a buy order and updated the posterior probability that value is high to 0.70 (value high = 102, low = 98). A second buy order then arrives. In this model informed traders buy when value is high with probability 0.9 and when value is low with probability
打开 →A dealer quotes a single ask price A for a stock whose value V is uniformly distributed on [40, 60]. An informed counterparty buys only when V > A (the quote is too cheap). Conditional on getting filled at ask A, what is the expected true value of the stock, and what does this im
打开 →A dealer quotes a two-sided market and earns the half-spread s on every fill. A fraction alpha = 0.15 of fills are informed and cost the dealer the full value gap delta = 0.8 (the dealer is on the wrong side for the entire move). The remaining fills are noise and the dealer keeps
打开 →A dealer's clientele is 20% informed and 80% noise on the current trade. Informed traders, when present, always pick the profitable side; noise traders are random. The dealer has just been adversely picked off (filled on the wrong side). Going into the NEXT incoming order, the de
打开 →The research desk wants not only labels but also synthetic feature draws conditional on each class for stress testing. Which side is the more natural starting point?
打开 →周一早盘，某私募的时间序列研究员把过去 200 个交易日的对冲组合超额收益丢进 statsmodels。她想确认这条曲线是不是一个干净的 ARMA 过程——若是，残差就是一组白噪声，可以挂上下一阶段的 GARCH；若不是，她得回去重做特征工程。问题是：用 AR(1)、MA(1)、ARMA(1, 1) 还是 ARMA(2, 1)？拟合完之后怎么知道这一支模型确...
打开 →Three candidate thresholds on the same classifier yield t=0.3 -> FP=18, FN=4; t=0.5 -> FP=9, FN=7; t=0.7 -> FP=4, FN=14. If one false negative costs 5 units and one false positive costs 1 unit(s), which threshold minimizes expected classification cost over this sample?
打开 →A PM is backtesting a signal-driven strategy and also wants to value a stop-loss overlay that behaves like an option. Which measure should dominate each part of the workflow?
打开 →Two hyperparameter settings differ in mean CV score by only 0.001, while the estimated standard error is 0.010. What is the first sensible interpretation?
打开 →The current best setting sits at extreme values on both the learning-rate and regularization grids. What should your next search action be?
打开 →