2e:null 2f:["$","title",null,{"children":"호기심 해결 in 주식시장: 예측모델로 알아보는 KRX 주요특성 변천사 | Humble Data Miner | Observable"}] 34:T6fa,--- ## (2018년~2022년) 정기보고서 기간별 주요특성의 추이

${md` * 정기보고서 제출기한 결산월 | 상장회사수
(유가증권) | 상장회사수
(코스닥) | 1분기보고서 | 반기보고서 | 3분기보고서 | 사업보고서 | :-: | :-: | :-: | :-: | :-: | :-: | :-: | 12 | 800 (98.4%) | 1704 (98.6%) | 5월 16일 | 8월 14일 | 11월 14일 | (다음해)4월 1일 * 98%의 상장사가 12월 결산월로 보고서 제출 (데이터셋에서 다른 결산월 종목들 제외) * 최신 재무정보의 유효기간 = 지난보고서와 다음보고서 사이의 시간 * e.g. 2022년 1분기보고서의 유효기간 = 2022년 5월 16일 ~ 2022년 8월 14일 * 최신 재무정보의 갱신마다 개별 예측모델을 학습 * (입력)최신 보고서 유효기간내 데이터 특성 → (출력)1년후 로그수익률 * 총 20개의 예측모델: (2018,2019,2020,2021,2022년)×(1분기,반기,3분기,사업보고서) * 참고사항 * (하단)슬라이더로 적용 보고서기간과 그에 따른 모델 변경가능 * 개별모델마다 특성의 영향력 분포(좌하단)와 크기의 평균값(우하단)을 표시 * (좌하단)분포상의 원에 커서를 올리면 같은 원에 속한 특성값들 평균 표시 `}

${htl.html.fragment`

${simple_model_representation}

`}35:T148d,chart = { const dispitems = 10; const svg = d3.create("svg").attr("viewBox", [0, 0, width, height]); const x1 = axis_x1; const x2 = d3.scaleLinear(extent.impact,[width-margin.right-plotdim.width,width-margin.right]); svg.append("g").attr("transform",`translate(0,${margin.top})`).call(d3.axisTop(x1)); svg.append("g").attr("transform",`translate(0,${margin.top})`).call(d3.axisTop(x2)); const y = axis_y; const g_yl = svg.append("g").attr("transform",`translate(${margin.left+plotdim.width},0)`); const g_yr = svg.append("g").attr("transform",`translate(${width-margin.right-plotdim.width},0)`); const beeswarm = beeswarmForce() .x(d => x1(d.m[0])) .y(y(0)+y.bandwidth()/2) .r(d=>Math.max(1,Math.sqrt(d.c)/16)); const label_fname = (fi,data) => `${(fi==0?"":"평균: ")+(fi==0?groups[data.m[1]]:(data.m[1]/100000000)>1?`${Math.floor(data.m[1]/100000000)}억원`:data.m[1])}\n표본수: ${data.c}`; const dtitle = svg.append("g"); dtitle.selectAll("text") .data(series) .join("text") .attr("x",(d,i)=>i*width+width/2).attr("y",0).attr("dy",3) .attr("dominant-baseline","hanging").attr("text-anchor","middle") .attr("font-weight","bold").attr("font-size","x-large") .text(seq=>`${label_finseq(seq)}`) svg.append("text").attr("text-anchor","middle").attr("dominant-baseline","hanging") .attr("x",margin.left+plotdim.width/2).attr("y",40).text("SHAP value (impact on model output)"); svg.append("text").attr("text-anchor","middle").attr("dominant-baseline","hanging") .attr("x",width-margin.right-plotdim.width/2).attr("y",40).text("mean(|SHAP value|) (average impact on model output magnitude)"); const g_bar = svg.append("g"); const g_swarms = d3.range(series.length) .map(si=>d3.range(fields.length) .map(fi=> svg.append("g").attr("visibility","hidden").attr("transform",`translate(0,${y(series[si].stat[fi].fi)})`) )); g_swarms.forEach((g_swarms_per_seq,si)=>{ g_swarms_per_seq.forEach((g,fi)=>{ const stat = series[si].stat[fi]; g.selectAll("circle") .data(stat.swarm) .join("circle") .attr("cx",d=>d.x).attr("cy",d=>d.y).attr("r",d=>Math.max(2,d.r)) .attr("fill",d=>d3.interpolateRdYlBu(1-d.data.fnorm)) .append("title").text((d,i)=>`${label_fname(stat.fi,d.data)}`); }); }); function ramp(color, n) { const canvas = DOM.canvas(n, 1); const context = canvas.getContext("2d"); canvas.style.margin = "0 -14px"; canvas.style.width = "calc(100% + 28px)"; canvas.style.height = "40px"; canvas.style.imageRendering = "-moz-crisp-edges"; canvas.style.imageRendering = "pixelated"; for (let i = 0; i < n; ++i) { context.fillStyle = color(i / (n - 1)); context.fillRect(i, 0, 1, 1); } return canvas; } svg.append("image") .attr("transform",`translate(${margin.left+20},${height/2}) rotate(90)`) .attr("width", height/2-20) .attr("height", 20) .attr("preserveAspectRatio", "none") .attr("xlink:href", ramp(d3.interpolateRdYlBu,300).toDataURL()); svg.append("text") .attr("text-anchor","start").attr("dominant-baseline","hanging").attr("font-size",14).attr("font-weight","bold") .attr("fill",d3.interpolateRdYlBu(0)) .attr("x",margin.left+20).attr("y",height/2).text("높은특성값"); svg.append("text") .attr("text-anchor","start").attr("dominant-baseline","auto").attr("font-size",14).attr("font-weight","bold") .attr("fill",d3.interpolateRdYlBu(1)) .attr("x",margin.left+20).attr("y",height-20).text("낮은특성값"); const name_field = i=>fields[i][1]; function _update(ns) { const tr1 = svg.transition().duration(anim_dur).ease(d3.easeLinear); const tr2 = svg.transition().duration(anim_dur).ease(d3.easeCubic); y.domain(ns.stat.map(d=>d.fi)); g_yl.selectAll("text").attr("text-anchor","middle").attr("dx",31); g_yl.transition(tr2).call(d3.axisRight(y).tickFormat(name_field)); g_yr.transition(tr2).call(d3.axisLeft(y).tickFormat(x=>"")); g_bar.selectAll("rect") .data(ns.stat,d=>d.fi) .join( enter => enter.append("rect") .attr("fill",d=>fields[d.fi][2]).attr("opacity",d=>d.impact/extent.impact[1]) .attr("x",x2(0)).attr("y",d=>y(d.fi)) .attr("width",d=>x2(d.impact)-x2(0)).attr("height",y.bandwidth()), update => update.transition(tr2) .attr("opacity",d=>d.impact/extent.impact[1]) .attr("y",d=>y(d.fi)) .attr("width",d=>x2(d.impact)-x2(0)).attr("height",y.bandwidth()), ); const next_si = series.findIndex(s=>s==ns); g_swarms.forEach((g_swarms_per_seq,si)=>{ g_swarms_per_seq.forEach((g,fi)=>{ const stat = series[si].stat[fi]; const show = next_si==si; if(show) g.attr("visibility","visible"); else g.transition().delay(anim_dur).attr("visibility","hidden"); g.transition(tr2).attr("transform",`translate(0,${y(stat.fi)})`).attr("opacity",show?1:0); }); }); dtitle.transition(tr2).attr("transform",`translate(${-next_si*width},0)`); } _update(series[0]); return Object.assign(svg.node(), { update(nextseq) { _update(nextseq); } }); }36:Tc12,## 결과 해석 * (전체기간동안) 일반적으로 영향력이 큰 특성 1. **종목그룹**: "자체적으로 많은 정보를 내포" * (납득가능한 사례) 코로나 시기의 **해운**, **게임사**, **식품** 등의 영향력 증대 * (대세라기엔 너무한) 2019년 **생명공학**,**제약** 2022년 **화학(2차전지)** * (도대체 왜?) 뜬금 튀어나오는 **종이와 목재**, **가스유틸리티** * (대암흑기) **종목그룹**만이 거의 유일한 변별특성이던 시기: 진정한 도박장? 2. **시가총액** * (변별력이 없음)수백억대 종목들이 절반(43%)이고 그중에서도 극소수에 편중 * (큰 등락폭)**시총**이 낮을수록 가격의 변동성도 크다 * 기간별 특이사항 * (2018,2019년 고요한 암흑기) **종목그룹**,**시가총액**만이 크게 중요하던 시기 * (2020년 코로나 펜데믹 르네상스) 한해뿐이지만 기업의 성적(**자본/시총**)이 중요했던 시기 * (선명한 선형성)**자본/시총** 크기에 비례한 수익률 (빨간색이 오른쪽, 파란색이 왼쪽) * (2021년 하락장엔 안전이 우선) 반기 조금넘게 지속된 현금부자(**잉여금/시총**) 기업들의 인기 * (약한 선형성)**잉여금/시총** 크기에 비례한 수익률 (붉은계열이 오른쪽, 파란계열이 왼쪽) * (2022년 돌아온 암흑기) 또다시 종목그룹에만 지배당하는 시장 * (2023년 상반기?) 조금씩 재무정보(**순익/시총**,**잉여/시총**)의 영향력이 커지는것 같은데? ## 투자자 여러분들께 남기고 싶은 말 * 시대별로 대세인 산업군에 투자금이 쏠릴수 있지만 그래도 좀 따져봐야 * 그 산업의 성장이 기존의 다른산업과 엮여서 경제성장을 이끌수 있는지? * 그 산업에 속한 기업들이 투여한 자본금에 따라 시총대비 성장하는지? * 국장을 박스피라고 조롱하는 여러가지 이유가 있겠으나... * 자본시장의 투자주체가 성적표에 근거해 정말 자본이 필요한 기업에 투자한다면 * 방만한 금융감독기관 없이도 시장이 불량한 지배구조의 기업을 걸러낼수 있지 않을지? * 그래야 비로소 시장이 합리적으로 동작하고 경제성장에 비례해 시총도 늘지 않을지?37:Tabc,arrow = { const svg = d3.create('svg').attr('viewBox', [0, 0, width, 10]); const markerBoxWidth=6; const markerBoxHeight=6; const refX = markerBoxWidth; const refY = markerBoxHeight/2; // See https://developer.mozilla.org/en-US/docs/Web/SVG/Element/marker const defs = svg.append('defs'); defs.append('marker') .attr('id', `arrowinc`) .attr('viewBox', [0, 0, markerBoxWidth, markerBoxHeight]) .attr('refX', refX/2) .attr('refY', refY) .attr('markerWidth', markerBoxWidth) .attr('markerHeight', markerBoxHeight) .attr('orient', 'auto-start-reverse') .append('path') .attr('d', d3.line()([[0,0],[0,markerBoxHeight],[markerBoxWidth,markerBoxHeight/2]])) .attr('stroke', 'red').attr('fill', 'red'); defs.append('marker') .attr('id', `arrowdec`) .attr('viewBox', [0, 0, markerBoxWidth, markerBoxHeight]) .attr('refX', refX/2) .attr('refY', refY) .attr('markerWidth', markerBoxWidth) .attr('markerHeight', markerBoxHeight) .attr('orient', 'auto-start-reverse') .append('path') .attr('d', d3.line()([[0,0],[0,markerBoxHeight],[markerBoxWidth,markerBoxHeight/2]])) .attr('stroke', 'blue').attr('fill', 'blue'); defs.append('marker') .attr('id', `circle`) .attr('viewBox', [0, 0, markerBoxWidth, markerBoxHeight]) .attr('refX', refX) .attr('refY', refY) .attr('markerWidth', markerBoxWidth) .attr('markerHeight', markerBoxHeight) .attr('orient', 'auto-start-reverse') .append('path') .attr('d', d3.line()([[0,0],[0,markerBoxHeight],[markerBoxWidth,markerBoxHeight/2]])) .attr('stroke', 'blue').attr('fill', 'blue'); defs.append('marker') .attr('id', `arrow`) .attr('viewBox', [0, 0, markerBoxWidth, markerBoxHeight]) .attr('refX', refX) .attr('refY', refY) .attr('markerWidth', markerBoxWidth) .attr('markerHeight', markerBoxHeight) .attr('orient', 'auto-start-reverse') .append('path') .attr('d', d3.line()([[0,0],[0,markerBoxHeight],[markerBoxWidth,markerBoxHeight/2]])) .attr('stroke', 'black').attr('fill', 'black'); ccodes.forEach((color,ci)=>{ defs.append('marker') .attr('id', `arrow${ci}`) .attr('viewBox', [0, 0, markerBoxWidth, markerBoxHeight]) .attr('refX', refX) .attr('refY', refY) .attr('markerWidth', markerBoxWidth) .attr('markerHeight', markerBoxHeight) .attr('orient', 'auto-start-reverse') .append('path') .attr('d', d3.line()([[0,0],[0,markerBoxHeight],[markerBoxWidth,markerBoxHeight/2]])) .attr('stroke', color).attr('fill', color); }); //svg.append("path").attr("marker-end","url(#arrow)").attr("fill","none").attr("stroke","black").attr("d",`M 0 5 h ${width}`); return svg.node(); }38:Tee9,single_sample_shap_values = { const ps = {w:400,h:300+90}; const svg = d3.create("svg").attr("width",ps.w).attr("height",ps.h); const samples = raw_data.samples; const rng = d3.randomInt(samples.length); const nr = ps.w/3; const g = svg.append("g"); function resample(){ g.selectAll("*").remove(); const sample = samples[rng()]; const bias = sample.slice(-1)[0]; const ordered = sample.slice(0,-1).map((v,fi)=>[v,fi]).toSorted((a,b)=>b[0]-a[0]).filter(v=>v[0]!=0); //console.log(ordered) const acc = ordered.reduce((acc,cur)=>{acc.push(acc[acc.length-1]+cur[0]); return acc;},[bias]); const xlines = d3.range(acc.length-1).map((_,i)=>({x1:acc[i],x2:acc[i+1]})); const x = d3.scaleLinear(d3.extent(acc),[10,ps.w-20]); const y = d3.scaleBand(d3.range(xlines.length),[40,ps.h-60]) g.selectAll("path") .data(xlines) .join("path") .attr("marker-end",d=>d.x2>d.x1?"url(#arrowinc)":d.x2d.x2>d.x1?"red":d.x2`M ${x(d.x1)} ${y(i)+y.bandwidth()/2} H ${x(d.x2)}`); g.selectAll("circle") .data(xlines) .join("circle") .attr("visibility",d=>d.x1==d.x2?"visible":"hidden") .attr("cx",d=>x(d.x2)).attr("cy",(d,i)=>y(i)+y.bandwidth()/2).attr("r",3).attr("fill","grey"); g.selectAll("text.textvalue") .data(xlines) .join("text") .attr("font-size",10).attr("class","textvalue") .attr("text-anchor","middle").attr("dominant-baseline","hanging") .attr("x",d=>x((d.x1+d.x2)/2)).attr("y",(d,i)=>y(i)+y.bandwidth()/2) .text((d,i)=>`${ordered[i][0]>0?"+":""}${ordered[i][0].toFixed(3)}`); g.selectAll("text.textfield") .data(xlines) .join("text") .attr("font-size",10).attr("class","textfield") .attr("text-anchor","middle").attr("dominant-baseline","auto") .attr("x",d=>x((d.x1+d.x2)/2)).attr("y",(d,i)=>y(i)+y.bandwidth()/2) .text((d,i)=>`${fields[ordered[i][1]][1]}`); const iv_right = x(acc[0])({si:si,y:seq.y,q:seq.q,d:seq.d})); const ygrps = seqs.reduce((acc,cur,i)=>{ if((i%4)==0) acc.push({y:cur.y,dset:new Set(),dlist:[]}); acc[acc.length-1].dset.add(cur.d[0]); acc[acc.length-1].dset.add(cur.d[1]); acc[acc.length-1].dlist.push(cur.d); return acc; },[]); ygrps.forEach((ygrp,i)=>{ ygrp.dset = [...ygrp.dset].map(d=>new Date(d)); ygrp.dlist = ygrp.dlist.map(d=>[new Date(d[0]),new Date(d[1])]); ygrp.x = d3.scaleTime(d3.extent(ygrp.dset),[margin.left,ps.w-margin.right]); }); const ty = ygrps[ygrps.length-1].y+1; ygrps.push({ y:ty, dset:ygrps[ygrps.length-1].dset.map(d=>new Date(d)), dlist:ygrps[ygrps.length-1].dlist.map(d=>[new Date(d[0]), new Date(d[1])]), }); const lastgrp = ygrps[ygrps.length-1]; lastgrp.dset.forEach(d=>{d.setFullYear(d.getFullYear()+1)}); lastgrp.dlist.forEach(d=>{d[0].setFullYear(d[0].getFullYear()+1);d[1].setFullYear(d[1].getFullYear()+1);}); lastgrp.x = d3.scaleTime(d3.extent(ygrps[ygrps.length-1].dset),[margin.left,ps.w-margin.right]); const stair_h = 14; const y_stair = d3.scaleBand(ygrps.map(d=>d.y),[margin.top+stair_h, (margin.top+stair_h)+(ps.h-margin.top-stair_h)*ygrps.length/(ygrps.length-1)]); const rects = ygrps.map( grp=>grp.dlist.map( term=>[{ x:grp.x(term[0]), y:y_stair(grp.y)-stair_h, w:grp.x(term[1])-grp.x(term[0]), h:stair_h}])).flat(2); const g_rects = svg.append("g").selectAll("rect") .data(rects) .join("rect") .attr("stroke","black").attr("fill","none") .attr("x",d=>d.x).attr("y",d=>d.y).attr("width",d=>d.w).attr("height",d=>d.h); ygrps.forEach((ygrp,yi)=>{ svg.append("g").attr("transform",`translate(0,${y_stair(ygrp.y)-stair_h})`) .call(d3.axisTop(ygrp.x).tickValues(ygrp.dset) .tickFormat(d=>`${d.getFullYear()}.${d.getMonth()+1}.${d.getDate()}`)); }); const g_inout = svg.append("g"); const g_in_box = g_inout.append("rect") .attr("fill","grey").attr("stroke","none") .attr("x",rects[0].x).attr("y",rects[0].y).attr("width",rects[0].w).attr("height",rects[0].h); const g_out_box = g_inout.append("rect") .attr("fill","grey").attr("stroke","none") .attr("x",rects[4].x).attr("y",rects[4].y).attr("width",rects[4].w).attr("height",rects[4].h); const g_in_text = g_inout.append("text") .style("text-shadow","1px 1px 1px black") .attr("x",Math.round(rects[0].x+rects[0].w/2)).attr("y",Math.round(rects[0].y+rects[0].h)) .attr("text-anchor","middle").attr("fill","white").attr("font-size",12).text("[입력]특성"); const g_out_text = g_inout.append("text") .style("text-shadow","1px 1px 1px black") .attr("x",Math.round(rects[4].x+rects[4].w/2)).attr("y",Math.round(rects[4].y+rects[4].h)) .attr("text-anchor","middle").attr("fill","white").attr("font-size",12).text("[출력]로그수익률"); const mbox_h = 20; const marrow_y1 = (rects[4].y-rects[0].y-rects[0].h)/2-mbox_h/2; const marrow_y2 = rects[4].y-rects[0].y-rects[0].h; const g_model = svg.append("g").attr("transform",`translate(${rects[0].x+rects[0].w/2},${rects[0].y+rects[0].h})`); g_model.append("path") .attr("marker-end","url(#arrow)").attr("stroke","black") .attr("d", `M 0 0 V ${marrow_y1}`); g_model.append("path") .attr("marker-end","url(#arrow)").attr("stroke","black") .attr("d", `M 0 ${(rects[4].y-rects[0].y-rects[0].h)/2+mbox_h/2} V ${rects[4].y-rects[0].y-rects[0].h}`); g_model.append("rect") .attr("stroke","none").attr("fill","black") .attr("x",-mbox_h/2+2).attr("y",(rects[4].y-rects[0].y-rects[0].h)/2-mbox_h/2+2) .attr("width",mbox_h).attr("height",mbox_h); g_model.append("rect") .attr("stroke","black").attr("fill","white") .attr("x",-mbox_h/2).attr("y",(rects[4].y-rects[0].y-rects[0].h)/2-mbox_h/2) .attr("width",mbox_h).attr("height",mbox_h); g_model.append("text") .attr("x",0).attr("y",marrow_y1).attr("dx",4).attr("dy",-2) .attr("text-anchor","start").attr("fill","black").attr("font-size",10).text("매수"); g_model.append("text") .attr("x",0).attr("y",marrow_y2).attr("dx",4).attr("dy",-2) .attr("text-anchor","start").attr("fill","black").attr("font-size",10).text("매도"); _update(series[0]); function _update(ns){ const si = series.findIndex(s=>s==ns); const tr2 = svg.transition().duration(anim_dur).ease(d3.easeCubic); g_in_box.transition(tr2) .attr("x",rects[si].x).attr("y",rects[si].y).attr("width",rects[si].w).attr("height",rects[si].h); g_out_box.transition(tr2) .attr("x",rects[si+4].x).attr("y",rects[si+4].y).attr("width",rects[si+4].w).attr("height",rects[si+4].h); g_in_text.transition(tr2) .attr("x",Math.round(rects[si].x+rects[si].w/2)).attr("y",Math.round(rects[si].y+rects[si].h)).attr("dy",-2); g_out_text.transition(tr2) .attr("x",Math.round(rects[si+4].x+rects[si+4].w/2)).attr("y",Math.round(rects[si+4].y+rects[si+4].h)).attr("dy",-2); g_model.transition(tr2) .attr("transform",`translate(${rects[si].x+rects[si].w/2},${rects[si].y+rects[si].h})`); } return Object.assign(svg.node(), { update(nextseq) { _update(nextseq); } }); }3a:[{"id":"65f982813677130ff76d63f3fc5f216ff13e803e14738afe52ff4f233e06fefa5e5e25bade3383f248c4a3e35180b1f1130ba03a6bf82f0ee5c9f73ad9a04cea","href":"https://static.observableusercontent.com/files/65f982813677130ff76d63f3fc5f216ff13e803e14738afe52ff4f233e06fefa5e5e25bade3383f248c4a3e35180b1f1130ba03a6bf82f0ee5c9f73ad9a04cea","name":"KRX_TreeSHAP_2018_2022@2.json","type":"application/json","lastModified":1718274374682.938,"size":2263067}] 32:{"id":"39dc4b3fdcf706d8","slug":"krx-feature-trend-using-shap-values-from-2018-to-2022","title":"호기심 해결 in 주식시장: 예측모델로 알아보는 KRX 주요특성 변천사","thumbnail":"df8dcb968ccf659d58e5f5ac4c26d4530dd79b784bdee5d5e7aba2bc5dcf496e","access_level":"public","create_time":"$D2024-06-11T17:13:37.275Z","update_time":"$D2024-06-17T13:29:12.067Z","view_time":null,"first_public_time":"$D2024-06-16T23:42:31.409Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"50faa4da52a9ff9b","login":"humble-data-miner","name":"Humble Data Miner","avatar_url":"https://avatars-next.observableusercontent.com/3046b1bea86c5e825baea7c0c403d90ad9690b5396b11b5d02b3a0839fbcb946","tier":"starter_2024"},"version":943,"publish_version":null,"role":null,"authored":false,"authors":[{"type":"individual","id":"4e2fc41e06361cc3","login":"humble-data-miner","name":"Humble Data Miner","avatar_url":"https://avatars-next.observableusercontent.com/3046b1bea86c5e825baea7c0c403d90ad9690b5396b11b5d02b3a0839fbcb946","tier":null}],"members":[{"type":"individual","id":"4e2fc41e06361cc3","login":"humble-data-miner","name":"Humble Data Miner","avatar_url":"https://avatars-next.observableusercontent.com/3046b1bea86c5e825baea7c0c403d90ad9690b5396b11b5d02b3a0839fbcb946","tier":null,"role":"editor"},{"type":"team","id":"50faa4da52a9ff9b","login":"humble-data-miner","name":"Humble Data Miner","avatar_url":"https://avatars-next.observableusercontent.com/3046b1bea86c5e825baea7c0c403d90ad9690b5396b11b5d02b3a0839fbcb946","tier":"starter_2024","role":"viewer"}],"comments":[],"custom_thumbnail":true,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":{"id":"ffed4a92f8068ba7","slug":"myth-buster-in-person","title":"Myth Buster in Person","owner":{"id":"50faa4da52a9ff9b","login":"humble-data-miner","name":"Humble Data Miner"}},"type":"notebook","body":{"title":"호기심 해결 in 주식시장: 예측모델로 알아보는 KRX 주요특성 변천사","cells":[{"id":0,"value":"# 호기심 해결 in 주식시장:
예측모델로 알아보는 KRX 주요특성 변천사","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":18,"value":"---\n\n## 여기서 예측모델이란?\n\n

\n* 특성(종목정보,재무정보,...)을 통해 1년후 수익률을 예측하는 모델\n* 이전 [개미투자공학 2장 확장판](https://observablehq.com/@humble-data-miner/mu-sigma-prediction-model-by-xgboost)에서 간단한 예측모델을 소개\n* 예측모델=(특성으로 표현해본)현실의 축소판\n * 입력(33)\n * 종목정보(2): 거래소(KOSPI/KOSDAQ), 종목그룹\n * 공시정보(2): 보고서순번, 보고후거래일\n * 재무정보(24): (연결C,개별I)×(분기Q,연간Y)×(iPBR,iPLR,iPTR,iPSR,iPER,iPCR)\n * 접두사 i는 inverse를 의미, e.g. PBR=시총/자산, iPBR=자산/시총\n * iPCR=현금흐름/시총, iPLR=부채/시총, iPTR=잉여이익금/시총\n * 가격정보(5): 시가총액, (주간W,월간M)×(최근2기간 가격변동)\n * 출력\n * 1년후 로그수익률(1)\n

\n ${htl.html.fragment`

${svg_model}

`}","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":327,"value":"---\n## 이미 예측모델을 갖고있다면?\n\n

\n\n${md`\n* 활용의 관점\n * 입력으로 원하는 종목의 특성값을 넣고 출력으로 수익률을 예측\n* 분석(해석)의 관점\n * 복잡한 모델에 대한 연역적 이해는 거의 불가능\n * Shapley value(특성의 출력기여도): 적용가능한 귀납적 분석방법\n * 관여하는 입력값들의 조합을 바꿔가면서 출력값의 변화를 본다면?\n * 하나의 데이터포인트에 대해서 각 특성의 출력기여도(변화정도)를 측정\n * 오른쪽 예시를 참고, 최종출력값=(특성별 출력기여도의 누적합)\n * 전체 데이터에 대해 각 특성의 출력기여도 분포와 크기의 평균을 추정\n * 주어진 모델에 대해 SHAP value 측정하는 알고리즘이 존재\n\n### 데이터를 이용해 모델을 역분석시 특성들의 출력영향력 측정가능\n* 영향력이 큰 특성들을 우선순위로 분석\n* 영향력이 미미한 특성들은 모델 입력특성에서 배제하는것도 고려가능\n`}\n

\n\n${htl.html.fragment`

${single_sample_shap_values}

`}","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":93,"value":"$34","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":124,"value":"viewof finseq = Scrubber(series, {\n autoplay: false,\n loop: false,\n delay: anim_dur+1000,\n format: seq=>label_finseq(seq)\n})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":102,"value":"$35","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":807,"value":"$36","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":912,"value":"---\n## 참고자료\n\n* Shapley value 이론 설명: [Wikipedia - Shapley value](https://en.wikipedia.org/wiki/Shapley_value)\n* python SHAP 라이브러리, SHAP value 시각화에 참고: [SHAP(SHapley Additive exPlanations) Documentation](https://shap.readthedocs.io/en/latest/index.html)\n* 트리모델 학습과 SHAP value 계산에 사용한 라이브러리: [XGBoost Documentation](https://xgboost.readthedocs.io/en/stable/)\n* 트리모델 학습과정의 소개, 특히 단일모델편을 참고: [개미투자공학 2장 확장판](https://observablehq.com/@humble-data-miner/mu-sigma-prediction-model-by-xgboost)\n* 로그수익률 참고: [개미투자공학 6장](https://observablehq.com/@humble-data-miner/personal-investment-engineering-ch-6?collection=@humble-data-miner/personal-investment-engineering)","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":942,"value":"---\n## 부연설명 동영상\n\n\n","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":8,"value":"---\n## Appendix","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":75,"value":"raw_data = FileAttachment(\"KRX_TreeSHAP_2018_2022@2.json\").json()","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":78,"value":"series = {\n const beeswarm = beeswarmForce()\n .x(d => axis_x1(d.m[0]))\n .y(axis_y.bandwidth()/2)\n .r(d=>Math.max(1,Math.sqrt(d.c)/16));\n\n const normalized_clip = (v,min,max) => Math.min(1,Math.max(0,(v-min)/(max-min)));\n const series = raw_data.series;\n series.forEach((seq,si)=>{\n seq.stat.forEach((stat,fi)=>{\n stat.clusts.forEach((clust,ci)=>{\n const fext = extent.feats[stat.fi];\n clust.fnorm = normalized_clip(clust.m[1],fext[0],fext[1]);\n });\n stat.swarm = beeswarm(stat.clusts);\n });\n });\n \n return series;\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":83,"value":"extent = raw_data.extent","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":151,"value":"fnamemap = {\n const entries = [\n [\"groupcode\",\"종목그룹\"],\n [\"market\",\"거래소\"],\n [\"totalamount\",\"시가총액\"],\n [\"seq\",\"보고서순번\"],\n [\"offset\",\"보고후거래일\"],\n [\"FX_USDKRW\",\"환율(KRW/USD)\"],\n [\"KRCALLBOKK\",\"콜금리\"],\n [\"KFIA114000\",\"CD금리\"],\n [\"KFIA103009\",\"회사채(3년/AA-)\"],\n [\"KR3YT_RR\",\"한국국채(3년)\"],\n [\"KR10YT_RR\",\"한국국채(10년)\"],\n [\"US3YT_RR\",\"미국국채(3년)\"],\n [\"US10YT_RR\",\"미국국채(10년)\"],\n [\"w0\",\"[직전]주간평균\"],\n [\"w1\",\"[1주전]주간평균\"],\n [\"w2\",\"[2주전]주간평균\"],\n [\"w3\",\"[3주전]주간평균\"],\n [\"m0\",\"[직전]월간평균\"],\n [\"m1\",\"[1주전]월간평균\"],\n [\"m2\",\"[2주전]월간평균\"],\n [\"m3\",\"[3주전]월간평균\"],\n ];\n entries.push(...\n [[\"c\",\"C\"],[\"\",\"I\"]].map(\n cons=>[[\"y\",\"Y\"],[\"q\",\"Q\"]].map(\n period=>[[\"b\",\"자본/시총\"],[\"l\",\"부채/시총\"],[\"t\",\"잉여/시총\"],[\"s\",\"매출/시총\"],[\"e\",\"순익/시총\"],[\"c\",\"현금/시총\"]].map(\n ratio=> [`${cons[0]}p${period[0]}${ratio[0]}r0`,`[${cons[1]},${period[1]}]${ratio[1]}`]\n ))).flat(2));\n return Object.fromEntries(entries);\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":365,"value":"ccodemap = {\n const entries = [\n [\"groupcode\",0],\n [\"market\",0],\n [\"totalamount\",3],\n [\"seq\",1],\n [\"offset\",1],\n [\"w0\",3],\n [\"w1\",3],\n [\"m0\",3],\n [\"m1\",3],\n ];\n entries.push(...\n [[\"c\",\"C\"],[\"\",\"I\"]].map(\n cons=>[[\"y\",\"Y\"],[\"q\",\"Q\"]].map(\n period=>[[\"b\",\"자본/시총\"],[\"l\",\"부채/시총\"],[\"t\",\"잉여/시총\"],[\"s\",\"매출/시총\"],[\"e\",\"순익/시총\"],[\"c\",\"현금/시총\"]].map(\n ratio=> [`${cons[0]}p${period[0]}${ratio[0]}r0`,2]\n ))).flat(2));\n return Object.fromEntries(entries);\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":370,"value":"ccodes = [d3.schemeCategory10[4],d3.schemeCategory10[5],d3.schemeCategory10[9],d3.schemeCategory10[8]]","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":86,"value":"fields = {\n return raw_data.fields.map(f=>[f,fnamemap[f],ccodes[ccodemap[f]]]);\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":89,"value":"groups = raw_data.groups","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":126,"value":"import {Scrubber} from \"@mbostock/scrubber\"","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":189,"value":"import { beeswarmForce } from \"@harrystevens/force-directed-beeswarm\"","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":235,"value":"height = 640","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":240,"value":"max_displayitems = 10","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":223,"value":"axis_x1 = d3.scaleLinear(extent.shap,[margin.left,margin.left+plotdim.width]);","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":238,"value":"axis_y = d3.scaleBand(d3.range(fields.length),[margin.top, margin.top+(height-margin.top-margin.bottom)*fields.length/max_displayitems]).paddingInner(0.2).paddingOuter(0.3);","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":298,"value":"function label_finseq(seq) {\n return `${seq.y}년 ${seq.q}분기 보고서 기간(${seq.d[0]}~${seq.d[1]}) 특성 → 1년후 로그수익률`;\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":229,"value":"plotpad = ({outer:10,inner:80})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":232,"value":"plotdim = ({width:(width-margin.left-margin.right-plotpad.inner)/2, height:height-margin.top-margin.bottom})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":226,"value":"margin = ({left:plotpad.outer,right:plotpad.outer,top:80,bottom:0})","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":131,"value":"anim_dur = 1000","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":145,"value":"{\n chart.update(finseq);\n simple_model_representation.update(finseq);\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":316,"value":"$37","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":319,"value":"svg_model = svg`\n`","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":392,"value":"$38","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":638,"value":"$39","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":10,"value":"html``","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W3a","stdlib":"1","titling":"auto"}} 33:"$32:body"