i3 dZddlmZddlmZmZddlmZddlZ ddl Z ddl m ZddlmZdZd1dZddd2d Zd d d  d3d Zdd d4dZdddd d5dZdd d6dZdddd d7dZeGddZeGddZedk(re j<j?dZ e jBd d!d"#Z"d$e jFe jIdd%e%e"&zZ&d'e&zd(ze jFe jIdd)e%e"&zZ'e jPde'ie"*Z)e jPde&ie"*Z*ed+d,d d+-Z+e+jYe)e*Z-e.d.e+j^e.e-gd/jajcd0yy#e$rZdZdZeZYdZ[dZ[wwxYw)8u Mean-reversion engines: * **Single-asset** — rolling z-score on one ``close`` series (:class:`StatArbEngine`). * **Pairs / stat-arb** — Engle–Granger-style hedge ratio (OLS), ADF on the residual spread, then rolling z-score on that spread (:class:`PairStatArbEngine`). Dependencies: pandas, numpy; pairs mode also requires **statsmodels**. :class:`PairStatArbEngine` fits the hedge **only on past data** (expanding or rolling window), forms a **causal** spread at each bar, runs ADF on that spread, and sets **basket PnL** to ``log_ret_y - β_{t-1}·log_ret_x`` (or simple returns when not using logs). A full-sample OLS is still stored on the instance for diagnostics (``ols_result``) but does not drive signals. ) annotations) dataclassfield)AnyN)adfullerc@tttdty)Nznstatsmodels is required for pair / ADF logic. Install with: python -m pip install statsmodels Original error: )smr ImportError _SM_ERRORA/opt/rentech/trading_bot/RenTech/strategy_stack/statarb_engine.py_require_statsmodelsr$s- zX% (k +  &r ) min_periodsc>||ntd|dz}|jtj}|j ||j }|j ||j d}||z |jdtjz }|S)z| Z-score of the current value vs trailing window mean and std dev. Uses sample std (ddof=1) inside the window. )windowr)ddof) maxastypenpfloat64rollingmeanstdreplacenan)seriesrrmpxmusigzs rrolling_zscorer&-s $/SFaK5HB bjj!A &b 1 6 6 8B ))6r) 2 6 6A 6 >C R3;;sBFF++A Hr @rentry_zexit_zcttjt|tj}tjd}|j}|j }t t|D]}||}tj|s|||<#|dk(r7|| krtjd}nU||k\rPtjd}n:|dk(r||k\r0tjd}n||krtjd}|||<tj||tjS)z Bar-by-bar position state machine on z-scores. Flat -> long (+1) when z <= -entry_z; flat -> short (-1) when z >= entry_z. Long exits when z >= exit_z; short exits when z <= exit_z. )dtyperr)indexr,) rzeroslenint8r.to_numpyrangeisfinitepdSeries) r%r)r*poscurrentidxvalsizis rrun_mean_reversion_positionsr=;s ((3q6 )CggajG ''C :: ? F F HB 2wQKCPRG9UVV%,BFF2i= !"Y-2F2Frzz2RE%,BFF2i= !"Y-2F2Frzz2RE  B  B BA WWQ F WWQ Fgga G WWQ F 1X#  #1s<001EE ; " U3-U3-//2773;#7"AB IIOOGWDOIMD!Q $q'N $q'Nr!uuRU|B q BJq Nq +#. ((C &$ &$ '% &$     s#'L  LLrRcPttj||dj}t |dkrt dt ||rt j|dn!|djt j}|rt j|dn!|djt j}tj|}tj||j}tj|j|j }tj|j"|j } t%|j&j(d} t%|j&j(d} | || | |fS) u First-stage Engle–Granger regression: regress Y on X (with constant). With logs: log(Y) = alpha + beta log(X) + epsilon; spread = epsilon (OLS residual). Returns ------- spread Cointegrating residual series (aligned to common index). fitted In-sample fitted values (same index). alpha, beta Intercept and hedge coefficient on X. ols_result statsmodels ``OLS`` results object. rVrPz/Need more overlapping bars for regression; got rBrDrXrr)rr5r[r\r0rJrr]rrr add_constantOLSfitr6 fittedvaluesr.residrdparamsiloc) rBrDrRrerMr"Xolsrlrkalphabetas rengle_granger_spreadrs1, 'g> ? F F HB 2w|J3r7)TUU!(r)}bm.B.B2::.NA!(r)}bm.B.B2::.NA A &&A,   C YYs''rxx 8F YYsyy 1F #**//!$ %E # $D 65$ ++r AICc)autolag regressionmaxlagct|jjtj}t |dkrdt t |dSd|i}|||d<|||d<t|fi|\}}}} } } | t| ni} t|t|t |t | | t|| jdtdk| d S) u Augmented Dickey–Fuller test on the spread (stationarity of residual). ``regression`` is passed to ``adfuller`` (default ``'c'`` = constant only). rP too_few_obs)errorrjrrrz5%r) adf_statisticpvalueused_lagn_obscritical_valuesis_stationary_5pctic_best) rr\rrrr0rZrdictrdboolget) rkrrrskwstatrusedlagnobscriticbests r adf_on_spreadrs rzz*A 1v{&SQ[99& 3B9  8 080Ab0A-D&'4v)4:rDt-LT"4$((4u*F#FG r cReZdZUdZdZded<dZded<dZded <d Zd ed <dd Z y) StatArbEnginezC Single-asset rolling z-score mean reversion on ``close``. rZrr'rdr)rr*r>strr@cJ|j|jvr-t|jdt|j|j }t ||j|j }||d<t||j|j|d<|S)Nz not in columns: zscorer(micro_position) r@rCKeyErrorlistcopyr&rr=r)r*)selfohlcvoutr%s r transformzStatArbEngine.transforms >> .dnn-->tEMM?R>STU Ujjl 3t~~. <H < t||DKK!   r N)r pd.DataFramereturnr) __name__ __module__ __qualname____doc__r__annotations__r)r*r@rr r rrrs6FCGUFEIs r rceZdZUdZdZded<dZded<dZded <d Zd ed <d Z ded<dZ ded<dZ ded<dZ ded<dZ ded<eeddZded<ed ddZded <d d! d#d"Zy )$PairStatArbEngineu Two-leg stat-arb: **causal** OLS hedge (expanding or rolling), ADF on that spread, rolling z-score, and **basket** bar returns for PnL. The **dependent** asset (Y) should match the name you use for downstream PnL (e.g. ``trade_ticker``); **X** is the hedge leg (e.g. sector ETF or basket). Columns added to the returned frame (aligned to Y after inner join): ``close_x``, ``pair_spread``, ``spread_fitted``, ``hedge_alpha``, ``hedge_beta`` (time-varying), ``zscore``, ``micro_position``, ``spread_ret`` (``diff`` of causal spread), component returns, and ``basket_ret`` — the hedgeable return ``r_y - β_{t-1} r_x`` (log or simple, matching ``use_log_prices``). rrZrr'rdr)rr*Truse_log_pricesN int | NonerSrP min_hedge_obsr str | None adf_autolagrradf_regressionr>r@F)default_factoryinitreprdict[str, Any] adf_report)defaultrrr ols_result)attach_y_ohlcvc||j}||jvs||jvrt|dt|||}t |d|d|j ^}}||_|j}||dkrd}t|d|d|j ||j\} } } } t| |j|j |_ t| |j} t!| |j"|j$ }|r*|j'| j(j+}n t-j.| j( }|dj'|j(}|dj'|j(}| j'|j(}||d<| j'|j(|d <| j'|j(|d <| j'|j(|d<||d<| j'|j(|d<|j'|j(j1dj3t4j6|d<|d j9|d<|j rUt5j:||j=dz |d<t5j:||j=dz |d<n&|j?|d<|j?|d<|d|j=d|dzz |d<|S)u Causal spread from past-only OLS; ADF on that spread; z-score; MR positions. ``ols_result`` is set to **full-sample** Engle–Granger OLS for diagnostics only. Signals and ``adf_report`` use the **causal** spread series. If ``attach_y_ohlcv``, start from ``ohlcv_y`` rows that align with the merge; otherwise return only merge columns + signals. z required on both framesr?rBrDr{NrrQ)rrr(rX pair_spread spread_fitted hedge_alpha hedge_betarr spread_retrret_yret_x basket_ret) r@rCrrOrrrrSrzrrrrrr&rr=r)r*reindexr.rr5r[fillnarrr1diffr]shift pct_change)rrKrLrpyrNruols_fshwrkrlalpha_sbeta_sr%r7rcycxhbs rrzPairStatArbEngine.transform7s ^^ W__ $'//(AbT!9:; ;$WgD) 9  9 '' F !    >bAgB*; 9  9 ''(( + '(8H8HUYUhUhi 64;; /*1dll4;;W //&,,/446C,,V\\2C I  & &syy 1 I  & &syy 1 ^^CII &I#^^CII6M%~~cii8O$__SYY7ML #)),H # CII 6 = =a @ G G P  .335L   66"rxx{"23CL66"rxx{"23CL==?CL==?CLL288A;W+EEL r )rKrrLrrrrr)rrrrrrr)r*rrSrrrr@rrrrrr r rrrs FCGUFEND#L*#M3#K#NCIs!&t%e!TJTDu5AJA $ EEE  E  Er r__main__z 2024-01-01ih)periodsfreqdg?)sizeg? g?rX(g?)rr)r*rzADF (causal spread):)r>rDrrrr)rNone)r pd.SeriesrrZrrrr)r%rr)rdr*rdrr)rKrrLrr@rrr) rBrrDrrRrrSrrTrZrz1tuple[pd.Series, pd.Series, pd.Series, pd.Series])rBrrDrrRrrz.tuple[pd.Series, pd.Series, float, float, Any]) rkrrrrrrrrr)2r __future__r dataclassesrrtypingrnumpyrpandasr5statsmodels.apiapir statsmodels.tsa.stattoolsrr r err&r=rOrzrrrrrrandom default_rngrng date_ranger9cumsumnormalr0r"rMr[dfydfxengrrprintrr\headr r rrs #( 2 I QU  " %4%4%4  %4  %4X        &# D D D D  D  D7DV #, #, #, #, 4 #,R  ! !! !  !  !H  , aa aH z ))   "C "-- c